CN105979589A - Method and system for allocating energy efficient resources of heterogeneous network - Google Patents

Abstract

The invention provides a method and a system for allocating energy efficient resources of a heterogeneous network. The beneficial effect is that each minimum data rate is restricted in the resource optimization allocation process in order to ensure the transmission quality while improving the energy efficiency. In allusion to the above resource optimization allocation problem, the invention provides a cross entropy based RB scheduling algorithm and a KKT condition based power allocation algorithm, and the effectiveness of the algorithms is verified through a simulation experiment.

Description

The Energy Efficient resource allocation methods of heterogeneous network and system

Technical field

The present invention relates to communication technical field, particularly relate to Energy Efficient resource allocation methods and the system of heterogeneous network.

Background technology

The mobile data services amount increased rapidly in global range, result in huge energy expenditure and consequent Environmental pollution.Report points out that the energy expenditure that Information & Communication Technology industry produces already takes up the 5 of whole world total electricity supply %^[1], and the CO2 emission thus caused has occupied the 2% of Global emissions total amount^[2].Therefore, how day by day to increase in support The important problem that energy expenditure, referred to as next generation mobile communication system face is reduced while long mobile data services demand.

Heterogeneous network (Heterogeneous Network, HetNet) is by the 3^rd Generation Partnership Project (3GPP) proposes in Long Term Evolution-Advanced (LTE-A) system.HetNet is at original grand base Stand and be added as needed on the microcell base station of low-power consumption on the basis of eNodeB, fill up macro base station and cover the lower blind spot existed.By Shorten the distance between base station and user in microcell base station, decrease the path loss impact on transmission of wireless signals, because of This can provide the user higher-quality data transport service.Microcell base station does not the most possess the computing function of complexity, and It is affected by the control of macro base station, the most therefore supports the energy expenditure needed for microcell base station runs much smaller compared to macrocell. Article is pointed out in [3] [4], can be obviously improved by the energy efficiency of careful design and deployment HetNet.

Although HetNet in theory has prominent performance advantage, but comprises greatly due to HetNet in a practical situation , therefore there is serious presence of intercell interference, significantly reduce the performance of system in the Microcell of amount dense deployment.Little in order to resist Interval interference, introduces coordinate multipoint (Coordinated Multiple Points, CoMP) transmission technology in LTE-A.? In CoMP transmission, connected base station can be right according to instantaneous channel condition information (Channel State Information, CSI) Send signal and carry out Combined Treatment, to reduce presence of intercell interference, to improve the Signal to Interference plus Noise Ratio (Signal-to-transmitted Interference-plus-Noise Ratio,SINR).Joint transmission (Joint Transmission, JT) CoMP is CoMP One in transmission technology, it allows that the base station that cluster is adjacent is one simultaneously and specifies user to send identical data symbol, one Aspect utilizes space diversity gain to improve the transmission quality of signal, on the other hand the most greatly weakens presence of intercell interference, Therefore JT CoMP transmission can be obviously improved the transmission quality of Cell Edge User.But, for realizing JT CoMP, need between base station By the mutual substantial amounts of data message of back haul link.Therefore, the capacity limit immediate constraint of back haul link passes based on JT CoMP The systematic function of defeated HetNet.

List of references:

[1]S.Tombaz,A.Vastberg,and J.Zander,“Energy-and cost-efficient ultrahigh-capacitywireless access,”IEEE Wireless Communications,vol.18,no.5, pp.18–24,2011.

[2]M..Company,“The impact of ict on global emissions,”on behalf of theGlobaleSustainability Initiative(GeSI),Tech.Rep.,November 2007.

[3]M.Deruyck,W.Joseph,B.Lannoo,D.Colle,and L.Martens, “DesigningEnergy-Efficient Wireless Access Networks:LTE and LTEAdvanced,”IEEE Internet Computing,pp.39–45,2013.

[4]M.Nasimi,F.Hashim,and C.K.Ng,“Characterizing energy efficiencyfor heterogeneous cellular networks,”in 2012IEEE Student Conferenceon Research and Development(SCOReD).IEEE,2012,pp.198–202.

[5]H.-L.M¨a¨att¨anen,K.H¨am¨al¨ainen,J.Ven¨al¨ainen,K.Schober, M.Enescu,and M.Valkama,“System-level performance of lte-advanced with jointtransmission and dynamic point selection schemes,”EURASIP Journalon Advances in Signal Processing,vol.2012,no.1,pp.1–18,2012.

[6]S.Sesia,I.Toufic,and M.Baker,LTE:The UMTS Long Term Evolutionfromtheory to practice.UK WILEY,2009.

[7]W.Yu,T.Kwon,and C.Shin,“Multicell coordination via joint scheduling,beamforming,and power spectrum adaptation,”IEEE Transactionson Wireless Communications,vol.12,no.7,pp.1–14,2013.

[8]Z.Shen,J.G.Andrews,and B.L.Evans,“Adaptive resource allocationin multiuser ofdm systems with proportional rateconstraints,”IEEETransactions on Wireless Communications,vol.4,no.6,pp.2726–2737,2005.

[9]Rubinstein R Y.Optimization of computer simulation models with rare events[J].European Journal of Operational Research,1997,99(1):89-112.

[10]Rubinstein R Y.The cross-entropy method for combinatorial and continuous optimization[J].Methodology and computing in applied probability, 1999,1(2):127-190.

[11] Cover T M, Thomas J A.Elements of Information Theory [M]. Beijing: Tsing-Hua University University press, 2003:12-40,239-262.

[12]D.Palomar and M.Chiang,“A tutorial on decomposition methodsfor network utilitymaximization,”IEEE Journal on Selected Areas inCommunications, vol.24,no.8,pp.1439–1451,August 2006.

[13]J.Salo,G.D.Galdo,J.Salmi,P.Ky¨osti,M.Milojevic,D.Laselva,and C.Schneider,“Matlab implementation ofthe 3gpp spatial channel model,”European Journal ofOperational Research,January 2005.[Online].

Available:http://read.pudn.com/downloads86/sourcecode/app/331591/scm 11-01-2005.pdf

[14]Dinnis A K and Thompson J S.The effects of including wraparound when simulating cellular wireless systems with relaying[C].IEEE 65th Vehicular Technology Conference.Dublin:IEEE,2007:914-918.

Summary of the invention

The invention provides the Energy Efficient resource allocation methods of a kind of heterogeneous network, comprise the steps:

A. receiving step, receives the parameter of input；

B. generating probability step, initializes allocative decision generating probability；

C. generate sample step, generate allocative decision sample according to generating probability；

D. calculation procedure, calculates the target function value of each sample；

E. screen step, sample is screened, and export a scheduling result；

F. judge step, it is judged that whether algorithm restrains, the most so perform to revise step, otherwise according to whole base stations Intermediate object program updates generating probability and then parameter performs to generate sample step；

G. step is revised, according to scheduling result and constraint condition C 4 correction result of whole base stations；

H. scheduling result output step, exports RB scheduling result；

I. resource allocation step, power distribution algorithm based on KKT obtains the power distribution result optimized；

J. resource distribution output step, exports Resource Allocation Formula.

As a further improvement on the present invention, in described generating probability step, TP m is the probability of RB scheduling on RB n Distribution initial value is expressed as

TP m probability distribution q that user selects on RB n is can get according to above formula^m _, ⁿ, and then TP m can be obtained at whole RB On user's select probability distribution

As a further improvement on the present invention, in described screening step, it is known that the capacity of the back haul link of TP m is C_m, During generating sample, if the handling capacity of the TP m of sample generationSampleTo directly be removed；With Sample ground, orderRepresent the thresholding of sample utility function in the t time iteration, the sample of requirement can not be reached for utility function (i.e.) also will be removed；N is generated according to above-mentioned requirements_SAMIndividual effective sample, is designated as

As a further improvement on the present invention, in described screening step, to N_SAMThe utility function of individual effective sample is carried out Descending, it is assumed thatSet a quantile ρ (0≤ρ≤1), for descending The sample of arrangement, intercepts whereinIndividual sample is as significant samples and general using these samples as updating The foundation of rate, and the utility function thresholding generating effective sample will be stepped upEvery time after iteration, this thresholding will be updated to The minima of utility function in significant samples, i.e.

$f_{thres}^{(t+1)}=f_m\left(X_{N_{IM}}^m\right)---\left(14\right).$

As a further improvement on the present invention, resource distribution output step, output Resource Allocation Formula includes calculating and works as The handling capacity in front moment, average fairness coefficient, energy efficiency.

Present invention also offers the Energy Efficient resource allocation system of a kind of heterogeneous network, including:

Receiver module, for receiving the parameter of input；

Generating probability module, is used for initializing allocative decision generating probability；

Generate sample module, for generating allocative decision sample according to generating probability；

Computing module, for calculating the target function value of each sample；

Screening module, for screening sample, and exports a scheduling result；

Whether judge module, restrain for evaluation algorithm, the most so performs correcting module, otherwise according to whole base stations Intermediate object program update generating probability and parameter and then perform to generate sample module；

Correcting module, according to scheduling result and constraint condition C 4 correction result of whole base stations；

Scheduling result output module, is used for exporting RB scheduling result；

Resource distribution module, power distribution algorithm based on KKT obtains the power distribution result optimized；

Resource distribution output module, exports Resource Allocation Formula.

As a further improvement on the present invention, in described generating probability module, TP m is the probability of RB scheduling on RB n Distribution initial value is expressed as

TP m probability distribution q that user selects on RB n is can get according to above formula^m _, ⁿ, and then TP m can be obtained at whole RB On user's select probability distribution

As a further improvement on the present invention, in described screening module, it is known that the capacity of the back haul link of TP m is C_m, During generating sample, if the handling capacity of the TP m of sample generationSampleTo directly be removed；With Sample ground, orderRepresent the thresholding of sample utility function in the t time iteration, the sample of requirement can not be reached for utility function(i.e.) also will be removed；N is generated according to above-mentioned requirements_SAMIndividual effective sample, is designated as

As a further improvement on the present invention, in described screening module, to N_SAMThe utility function of individual effective sample is carried out Descending, it is assumed thatSet a quantile ρ (0≤ρ≤1), for descending The sample of arrangement, intercepts whereinIndividual sample is as significant samples and general using these samples as updating The foundation of rate, and the utility function thresholding generating effective sample will be stepped upEvery time after iteration, this thresholding will be updated to The minima of utility function in significant samples, i.e.

$f_{thres}^{(t+1)}=f_m\left(X_{N_{IM}}^m\right)---\left(14\right).$

As a further improvement on the present invention, in described resource distribution output module, output Resource Allocation Formula wraps Include the handling capacity calculating current time, average fairness coefficient, energy efficiency.

The invention has the beneficial effects as follows: in order to ensure the quality of transmission while improving energy efficiency, the present invention is in money The minimum data rate often transmitted is any limitation as by source optimization assigning process.For above-mentioned resources configuration optimization problem, this Bright propose RB dispatching algorithm based on cross entropy and power distribution algorithm based on KKT condition, and verified by emulation experiment The effectiveness of algorithm.

Accompanying drawing explanation

Fig. 1 is the system model figure of the present invention；

Fig. 2 is the cross-entropy method ultimate principle figure of the present invention；

Fig. 3 is the topological schematic diagram of the emulation of the present invention；

Fig. 4 is the system average throughput spirogram under the different back haul link capacity limit of the present invention；

Fig. 5 is the system average energy efficiency figure under the different back haul link capacity limit of the present invention；

Fig. 6 is the system fairness coefficient figure under the different back haul link capacity limit of the present invention；

Fig. 7 is the system average energy efficiency figure under the different QoS thresholding of the present invention；

Fig. 8 is the system average throughput spirogram under the different QoS thresholding of the present invention；

Fig. 9 be the present invention different QoS thresholding under the fairness coefficient figure of user data rate；

Figure 10 is the method flow diagram of the present invention.

Detailed description of the invention

As shown in Figure 10, the invention discloses the Energy Efficient resource allocation methods of a kind of heterogeneous network, including walking as follows Rapid:

Step S1, receives the parameter of input；

Step S2, initializes allocative decision generating probability；

Step S3, generates allocative decision sample according to generating probability；

Step S4, calculates the target function value of each sample；

Step S5, screens sample, and exports a scheduling result；

Step S6, it is judged that whether algorithm restrains, the most so performs step S7, otherwise according to the middle junction of whole base stations Fruit updates generating probability and then parameter performs step S3；

Step S7, according to scheduling result and constraint condition C 4 correction result of whole base stations；

Step S8, exports RB scheduling result；

Step S9, power distribution algorithm based on KKT obtains the power distribution result optimized；

Step S10, exports Resource Allocation Formula.

In step sl, parameter includes base station number, UE quantity, the accumulation transmission capacity of each UE, RB quantity, constant power Apportioning cost, instantaneous CSI, CoMP set select.

In step slo, output Resource Allocation Formula includes the handling capacity calculating current time, average fairness system Number, energy efficiency.

The present invention, in HetNet downlink transmission system based on DYNAMIC J T CoMP transmission, have studied to improve system energy Amount efficiency is the frequency of target, power joint optimization problem.The resources configuration optimization of network always launched by TP power limited and Back haul link finite capacities etc. retrain.It addition, for the quality ensureing transmission while improving energy efficiency, the present invention is in money The minimum data rate often transmitted is any limitation as by source optimization assigning process.For above-mentioned resources configuration optimization problem, this Bright propose RB dispatching algorithm based on cross entropy and power distribution algorithm based on KKT condition, and verified by emulation experiment The effectiveness of algorithm.

Illustrate for:

1. algorithm produces RB dispatch the generating probability of sample according to global information and the formula (11) of input.

The most each TP generates great amount of samples according to corresponding generating probability, and carries out the screening of sample according to given code. If acquired results meets the condition of convergence, then this result is modified by scheduling result and constraint condition C 4 according to whole TP, I.e. remove the transmission that can not meet qos requirement, the revised RB scheduling result of final output.

If the scheduling result 3. obtained is unsatisfactory for the condition of convergence, then need according to intermediate object program more new samples generating probability and its His relevant parameter.Wherein relevant parameter, need to be updated according to the scheduling result of whole TP.

4. the scheduling result of output is input in power distribution algorithm, it is thus achieved that the power distribution result of optimization

5. the various performance parameters of system is calculated according to final RB scheduling and power distribution result.

The call that the Energy Efficient resource allocation methods of the heterogeneous network of the present invention more prepares is: based on JT CoMP transmission The Energy Efficient resource allocation methods of HetNet.

Consider the downlink transmission system of a HetNet network.Represent for convenience, eNB therein and each Microcell base Stand and be all referred to as TP.Assume that in network, total M TP, each TP are by back haul link and centralized control unit (Control Unit, CU) it is connected.Network topology is as shown in Figure 1.Centralized control unit according to the CSI of user feedback according to give to strategy enter The scheduling of row resource and distribution, then informed result to each TP by back haul link.In order to effectively implement CoMP transmission, Need between TP to keep synchronizing in time, frequency spectrum and phase place.By means of back haul link, such synchronization is held in HetNet very much Easily realize.

Assume that in network, each TP is equipped with N_TIndividual transmitting antenna, and each user is equipped with N_RIndividual reception antenna.According to Definition in LTE standard, the unit of wireless transmission resources is referred to as Resource Block (Resource Block, RB), and each RB comprises 12 continuous print subcarriers, and continue 1 Transmission Time Interval (Transmit Time Interval, TTI) in time.False If the frequency spectrum resource of network can be divided into a N_RBRB, and be used in conjunction with by whole TP.It is considered herein that channel fading is pseudo-steady , i.e. channel is stable within each TTI.It addition, present invention assumes that user and base station all can obtain the wink of zero defect Time CSI.

In lte-a system, TP periodically broadcasts reference signal (Reference Signal, RS), and user is according to RS Receiving intensity the channel condition between TP and user is estimated.Using in the system of CoMP technology, user is according to RS's Intensity according to given one group of TP of policy selection as the participant of CoMP transmission.Such one group of TP is referred to as the CoMP of user Set.Make M_kRepresenting the CoMP set of user k, user k selects M according to following rule_kIn TP:

$\left\{\begin{array}{cc}m\∈M_k& m=thestrongestTP,or\\ & RS\left(thestrongestTP\right)-RS\left(TPm\right)\≤{\mathrm{\Δ}}_{thres}\\ m\∉M_k& otherwise\end{array}\right.---\left(1\right)$

Wherein, RS represents the received signal strength of reference signal RS, Δ_thresIt it is the thresholding of intensity difference in units of dB. Formula (1) represents, M_kIn necessarily comprise the TP corresponding for RS that signal intensity that user k receives is the strongest.It addition, if there is other The intensity of TP, its RS and the difference of maximum intensity be not more than Δ_thresDB, then it is assumed that the channel condition between this TP and user is relatively Good, can be used for carrying out CoMP transmission.Obviously, Δ_thresValue the biggest, M_kThe middle TP that may comprise is the most.Ideally, CoMP TP quantity in set is the most, and the quality of CoMP transmission is the highest.But, TP quantity is the most in a practical situation, for realizing CoMP Control overhead needed for transmission is the biggest.LTE standard thinks rational Δ_thresIt is worth in the periphery of 5～6dB^[5]。

The rule defined according to formula (1), M_kOne or more TP may be comprised.When user k is near some TP, this TP RS at the receiving intensity of user side, may be much larger than the intensity of other RS.In this case, M_kOnly comprise " the strongest TP ", Therefore system can not provide CoMP transmission for user k.Such user is referred to as non-CoMP user in the present invention.It practice, CoMP transmission is the least for the gain that non-CoMP user is brought, and conventional transmission mode can provide enough transmission matter Amount.If on the contrary, user's is located away from TP, even fall within the overlapping region of multiple TP coverage, then user accepts To RS in there is the RS, TP corresponding for these RS of multiple strength similarity and can be selected as the element in CoMP set, for user CoMP transmission is provided.In the present invention, such user is known as CoMP user.

The CoMP set selected based on user, HetNet can provide the user JT CoMP transmission.In JT CoMP transmission, many Individual TP sends identical data symbol to appointment user.Owing between TP, position is enough remote, between the most multiple channels the most solely Vertical.User merges process to multiple signal copies, it is thus achieved that the signal intensity of reinforcement.Meanwhile, JT CoMP transmission eliminates biography Defeated main interference, significantly reduces total interference strength.CoMP set M selected according to formula (1) based on user k_k, adopt On RB n, transmit, with user k when JT CoMP transmission, the data rate obtained to be represented by:

$R_k^n=blog(1+\frac{\underset{m\∈M_k}{\Σ}\left|\right|H_{m,k}^n{w}_m^n|{|}^{2}S}{\underset{m^{\′}\∈M\backslash M_k}{\Σ}\left|\right|H_{m^{\′},k}^n{w}_{m^{\′}}^n|{|}^{2}S+{\mathrm{\σ}}^2})---\left(2\right)$

Wherein,It is N_R×N_TThe channel matrix of dimension, its elementRepresent jth antenna and the user k of TP m I-th antenna between channel coefficients；M represents the set of whole TP composition in network, and M{m} then represents from M removal element { the set after m}；It is N_T× 1 dimensional vector, represents that TP m is to symbolPrecoding, and have Represent The TP m transmitting power to this transmission distribution；Represent white complex gaussian noise at reception antenna to Amount；B represents the total bandwidth of a RB, and in lte-a system, b is generally 180kHz^[6]。

Formula (2) implys that a kind of fixing JT CoMP transmission strategy.The most once user k selectes CoMP set, descending biography Communication system makes M in ensuing TTI_kIn whole TP be that user sends data symbol.But, due to the time variation of channel, Such fixing JT CoMP transmission not always optimum.Present invention employs a kind of dynamic JT CoMP, make system often TTI selects M according to instantaneous CSI_kA subset be user k implement JT CoMP transmission.So on the one hand, can reduce need not The waste wanted, on the other hand M_kIn other TP according to circumstances can send data for other users, thus improve the overall of system Data rate.In order to represent DYNAMIC J T CoMP transmission, the index set of definition scheduling hereWherein It is that user k transmits data, the most then that explanation system distributes the RB n of TP m in ensuing TTI Represent that TPm will not service for user k on RB n.Utilizing scheduling index, user k passes through DYNAMIC J T CoMP transmission on RB n The data rate obtained is:

$R_k^n=blog(1+\frac{\underset{m=1}{\overset{M}{\Σ}}{\mathrm{\β}}_{m,k}^n\left|\right|H_{m,k}^2{w}_m^n|{|}^2{p}_m^n}{\underset{m^{\′}=1}{\overset{M}{\Σ}}(1-{\mathrm{\β}}_{m^{\′},k}^n)\left|\right|H_{m^{\′},k}^n{w}_{m^{\′}}^2|{|}^2{p}_{m^{\′}}^n+{\mathrm{\σ}}^2})---\left(3\right)$

Problem models:

Primary study of the present invention is limited at transfer resource and under the constraint of back haul link finite capacity, to maximize weighting Energy efficiency is the frequency of target, power joint resource optimization problem.For system model previously discussed, here can dose-effect Rate is defined as the ratio of handling capacity and energy expenditure, the throughput of system (bps/ that i.e. every Watt power consumption can obtain Watt).In traditional greedy algorithm, in order to ask most the maximization of object function, system can allocate resources to as much as possible The preferable user of channel condition, the user causing channel condition poor cannot obtain enough services.In order to avoid this extremely Situation, data rate is weighted by the present invention, and weight coefficient user k is at the progressive average obtained to current TTI position According to speed^[7], it is defined as

${\stackrel{\‾}{R}}_k=\mathrm{\α}{\stackrel{\‾}{R}}_k^{before}+(1-\mathrm{\α})R_k---\left(4\right)$

In formula, α (0 < α < 1) is forgetting factor, is used for balancing cumulative mean according to speed and current data rate to resource The impact of distribution；Representing and end to current time, the cumulative mean of user k is according to speed.Introduce the weighting of formula (4) After coefficient, system is more likely the user resource allocation that channel condition is poor when resource is distributed, thus improves in network and use The fairness of user data rate.In order to quantify fairness, fairness coefficient defined in document [8]:

$F={\left(\underset{k=1}{\overset{K}{\&Sigma;}}{R}_k\right)}^2/K\underset{k=1}{\overset{K}{\&Sigma;}}{R}_k^2---\left(5\right)$

In a HetNet network, the performance of system is limited by many-sided condition.First, total transmitting of each TP Power is limited.It addition, connect the back haul link of TP and centralized control unit, its capacity is deposited according to the difference of medium used In different restrictions.Back haul link capacity limit TP can receive and be sent to the total amount of the data of user in every TTI, thus Limit TP total data rate in a TTI.Finally, in order to ensure the Service Quality of user while improving energy efficiency Amount, the present invention is provided with thresholding to the data rate often transmitted, it is ensured that the data rate that obtains of transmission that system is implemented all can be On thresholding.In sum, the frequency towards energy efficiency of present invention research, power joint optimization problem can be modeled as:

$\begin{array}{cc}\underset{{\mathrm{\&beta;}}_{m,k}^n,p_m^n}{\max }& \underset{k=1}{\overset{K}{\&Sigma;}}\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\frac{R_k^n}{{\stackrel{\&OverBar;}{R}}_k}/\underset{m=1}{\overset{M}{\&Sigma;}}\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n{p}_m^n\end{array}$

$\begin{array}{ccc}s.t.& C1:& 0\&le;p_m^n\&le;S,\&ForAll;m,n\end{array}$

$\begin{array}{cc}C2:& {\mathrm{\&beta;}}_{m,k}^n\&Element;\{0,1\},\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n\&le;1,\&ForAll;m,n,k\end{array}$

$\begin{array}{cc}C3:& \underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n{R}_k^n\&le;C_m,\&ForAll;m\end{array}$

$\begin{array}{cc}C4:& R_k^n\&GreaterEqual;R_{thres},\&ForAll;k,n\end{array}$

Wherein, C1 represents the restriction of high emission power, and C2 represents that RB can not be duplicatedly distributed, and C3 represents the flyback line appearance of a street Measure quantitative limitation that every TP is handled up；R_thresFor given data rate thresholding, the Resource Allocation Formula of system should ensure that each biography Defeated data rate is not less than this thresholding, the quality of each transmission during therefore C1 ensure that network.

RB based on cross entropy height algorithm:

The hybrid optimization problem that formula (5) represents contains discrete variable and continuous variable, is that typical NP difficulty is asked Topic, it is impossible to try to achieve optimal solution in polynomial time.In order to hybrid optimization problem be simplified, the present invention is by formula (3-7) It is divided into two subproblems: first, it is assumed that the power averaging of launching of base station distributes to the transmission on each RB, can obtain about RB The subproblem of scheduling.Due to power it is known that optimization problem now becomes and only comprises variableDiscrete/Combinatorial Optimization ask Topic, is prone to solve compared with former problem.Secondly, in known scheduling resultHypothesis under, one can be obtained about merit The continuous optimization problem of rate distribution, can be solved by the analytic method of iteration.

The present invention is first under the hypothesis of constant power distribution, it is proposed that RB dispatching algorithm based on cross-entropy method.

Cross-entropy method brief introduction:

Cross entropy (Cross Entropy, CE) algorithm initially by Rubinstein 1997 propose, for complexity with In machine network, the probability of rare event is estimated^[9].Subsequently, Rubinstein finds simply to repair Cross-Entropy Algorithm Just, just can be used to combinatorial optimization problem is solved^[10]。

Cross entropy is a key concept in present information opinion, namely refers to Kullback-Leibler (K-L) distance Or relative entropy^[11].In random experiment, cross entropy represents " distance " between two probability distribution.Assume f (x) and g (x) Two continuous print probability density functions in the χ of territory, then " distance " between two probability density functions, in other words both Cross entropy is defined as:

$D(f,g)=E_f\ln \frac{f\left(x\right)}{g\left(x\right)}=\&Integral;f\left(x\right)\ln f\left(x\right)dx-\&Integral;f\left(x\right)\ln g\left(x\right)dx---\left(6\right)$

Therefore, cross entropy can be understood as the uncertainty of f (x) during known prior probability g (x).With all " distance " As definition, cross entropy has nonnegativity, i.e. and D (f, g) >=0, and during and if only if f (x)=g (x), equal sign is set up.The opposing party Face, cross entropy does not have a symmetry, i.e. D (f, g) ≠ D (g, f).

The probability density function of random vector x (can also be one-dimensional stochastic variable) can utilize a real parameters to Amount v characterizes, and may comprise real parameters such as average, variance etc. in vector.Assume that S is a real-number function on the χ of territory, to one Individual given real number γ^*, at probability density function f (x；U) under, the value of S is not less than γ^*Probability can be expressed as l=P_u(S(x)≥ γ^*).If the value of probability l is the least, e.g., less than 10^-5, then event { S (x) >=γ can be claimed^*It it is a rare event.Aobvious So, the premise of the probability obtaining rare event is intended to obtain corresponding parameter vector u.Cross-entropy method estimates rare event probability During l, first give the parameter vector w of reference, and according to f (x；W) N number of sample is produced.Then important sampling is utilized Sample is screened by (Important Sampling) strategy, and according to the Sample Refreshment parameter vector w filtered out.If through Dry iteration, w eventually converges to u, thus obtains event { S (x) >=γ^*Probability l.

Cross-entropy method solves combinatorial optimization problem, the appearance of optimization problem optimal solution is considered as rare event and assumes S It is a real-valued utility function on the Χ of territory, optimization problem is usually expected to find on the χ of territory the maximum γ of S^*, Yi Jiling The vector x that S is maximum^*, i.e.

$S\left(x^{*}\right)={\mathrm{\&gamma;}}^{*}=\underset{x\&Element;\mathrm{\&chi;}}{\max }S\left(x\right)---\left(7\right)$

The principle of cross-entropy method is the probability l=P to rare event_u(S(x)≥γ^*) estimate.But it is worth It is noted that γ now^*It is unknown.Consider a certain parameter vector w, be first according to probability distribution f (x；W) produce N number of with Press proof originally, selects several samples that utility function is bigger, according to the Sample Refreshment parameter vector w filtered out.Through some Secondary iteration, the utility function generating sample will converge on γ^*, the random sample simultaneously generated according to parameter vector w will converge on x^*, i.e. the optimal solution of optimization problem.

Cross-entropy method application in combinatorial optimization problem has been widely studied.Typical combination optimization problem, as Change greatly cutting problem, travelling merchants' problem, quadratic assignment problem etc., be all proved to utilize cross-entropy method to obtain optimal solution.

RB dispatching algorithm based on cross entropy:

During known power distribution, RB scheduling problem can be modeled as:

$\begin{array}{ccc}\underset{{\mathrm{\&beta;}}_{m,k}^n}{\max }& & \underset{k=1}{\overset{K}{\&Sigma;}}\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\frac{R_k^n}{{\stackrel{\&OverBar;}{R}}_k}/\underset{m=1}{\overset{M}{\&Sigma;}}\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^{n}S\\ s.t.& C1:& p_m^n\&Element;\{0,S\},\&ForAll;m,n\\ & C2:& {\mathrm{\&beta;}}_{m,k}^n\&Element;\{0,1\},{\&Sigma;}_{k=1}^K{\mathrm{\&beta;}}_{m,k}^n\&le;1,\&ForAll;m,n,k\\ & C3:& \underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n{R}_k^n\&le;C_m,\&ForAll;m\\ & C4:& R_k^n\&GreaterEqual;R_{thres},\&ForAll;k,n\end{array}---\left(8\right)$

Variable in the problems referred to aboveBe a dimension be M × N_RBThe random matrix of × K, whereinIt it is a ratio Special number, can be considered Bernoulli random variable.In each iteration of cross-entropy method, need to produce the sample of q.s.If it is straight It is right to connectSolve, it is clear that amount of calculation is the biggest.

In order to reduce computation complexity, it is considered to generate sample respectively for each TP.A sample on note TP m is X^m =[x^m(1),...,x^m(n),...,x^m(N_RB)], wherein x^mN () represents the user that TP m will service on the n-th RB, i.e.As previously described, user enters the first meeting intensity selection cooperation set M according to reference signal of network_k, thus Centralized control unit is readily available the user of each TP and gathers U_m.Sample X^mIn element x^mN () is then according to given probability distribution From set U_mChoose.Such sample design, can reduce sample space and computation complexity.

The present invention utilizes above-mentioned sample design, it is proposed that a kind of heuristic mutation operations strategy based on cross-entropy method.Should Strategy can be roughly divided into three parts: probability initializes, iterative part and modified result.Wherein iterative part is the core of algorithm Part, includes sample and generates, screens and probability distribution renewal.Iterative process finally makes convergence of probability distribution determine knot in one Really, the optimal solution of scheduling problem it is.

1) probability distribution initializes:

In system, each TPm gathers U according to current channel condition and given strategy from its user_mChoose user suitably RB on be transmitted.When with energy efficiency for optimization aim, exist and make on some RB to improve system energy efficiency It is not transmitted.In order to represent such situation, TP m can be made in set [0, U_mUser is chosen in].When choosing 0, then table Show that TPm does not implement transmission on this RB.

Note U '_m=[0, U_m], on RB n, TP m dispatches user k ∈ U '_mProbability distribution availability vector(1≤i≤|U′_m|) represent.WhereinExpression system select on this RB to Family k=U '_mI () transmits (i.e.) probability, and meetAccording to U '_mDefinition understand U '_m(1) =0, and corresponding probabilityThen represent that TP m does not occur the probability of any transmission on RB n.In initialization procedure, we This probability is set as

$q_1^{m,n}=\mathrm{Pr}\_0---\left(9\right)$

Wherein 0 < Pr_0 < 1 is a given probit.

For U '_mIn nonzero element then according to the estimation of user data rate, probability is composed initial value.Such assignment Convergence rate can be accelerated to a certain extent.Channel situation according to user and the selection of cooperation set, use JT CoMP strategy Time, available according to fixing JTCoMP transmission strategy (i.e. formula (2)), user k data rate on RB n is estimated.

The QoS that every RB is transmitted by constraints C4 in formula (8) is limited, it is impossible to reach thresholding R_thresIt is considered It it is unsuccessful transmission.In order to economize on resources, system will not consider to dispatch on RB nUser k, even k pair The probability answeredIt is 0；And for data rate thresholding R can be reached_thresUser, then estimate according to it Data rate accounts for the proportion of total data rate and carries out probability assignment.Assuming that the data rate of user k' meets qos requirement, it is right The probability answeredIt is defined as:

In above formula, on the right of equal sign, Section 1 illustrates the data rate of user and accounts for the proportion of total data rate, should be noted that Be that total data rate contains only the data rate that disclosure satisfy that QoS；Section 2 on the right of equal sign is then to ensure that probability distribution q^m,nMeet

In sum, TP m probability distribution initial value of RB scheduling on RB n can be expressed as

TP m probability distribution q that user selects on RB n is can get according to above formula^m _, ⁿ, and then TP m can be obtained at whole RB On user's select probability distributionAt the iteration initial stage, algorithm is according to q^mGenerate sample, Further according to the sample situation update probability distribution after screening, until convergence of probability distribution.

2) iterative part:

In the iterative process of algorithm, system is according to given several samples of probability distribution stochastic generation.Stochastic generation Sample do not ensure that and meet constraints and object function requirement, it is therefore desirable to sample is screened.Screening gained Sample is considered as the sample of " good ", and according to the Sample Refreshment probability distribution of " good ", it is possible to make algorithm when next iteration The sample of " more preferably " is obtained with higher probability.By that analogy, final system can obtain and meet constraints and object function Result.

According to known probability distribution for TP m stochastic generation sample, it is designated asIt makes f_mRepresent effect corresponding to TP m With function, according to formula (8) it is known that sampleThe utility function obtained is:

$f_m\left(X_i^m\right)=\underset{n=1,x_i^m\left(n\right)\&NotEqual;0}{\overset{N_{RB}}{\&Sigma;}}\frac{R_{x_i^m\left(n\right)}^n}{{\stackrel{\&OverBar;}{R}}_{x_i^m\left(n\right)}}/\underset{n=1,x_i^m\left(n\right)\&NotEqual;0}{\overset{N_{RB}}{\&Sigma;}}S---\left(12\right)$

In formula,S represents according to sampleThe total power consumption of the TP m obtained.

SampleThe corresponding total throughout on TP m is represented by:

$R_m\left(X_i^m\right)=\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}{R}_{x_i^m\left(n\right)}^n---\left(13\right)$

Qualified sample should meet two conditions: first, and the utility function value of sample is sufficiently high；Second, sample need to be expired Constraints C3 in foot back haul link capacity limit, i.e. formula (8).The sample randomly generated does not ensures that and meets the two Condition, it is therefore desirable to the sample generated is screened.

The capacity of the back haul link of known TP m is C_m, during generating sample, if the TP m's of sample generation gulps down The amount of tellingSampleTo directly be removed.Similarly, orderRepresent sample utility function in the t time iteration Thresholding, the sample of requirement can not be reached for utility function(i.e.) system will not consider yet.System N is generated according to above-mentioned requirements_SAMIndividual effective sample, is designated as

Further, algorithm according to the principle of important sampling at N_SAMIndividual effective sample filters out significant samples.To N_SAM The utility function of individual effective sample carries out descending, without loss of generality, it can be assumed thatIf A fixed quantile ρ (0≤ρ≤1), for the sample of descending, intercepts whereinIndividual sample conduct Significant samples, and using these samples as the foundation of update probability.SymbolRepresent and take the positive number closest to numerical value a.In order to Making sample results closer to optimum object function in iteration, algorithm generates the utility function of effective sample by stepping up every time ThresholdingEvery time after iteration, this thresholding will be updated to the minima of utility function in significant samples, i.e.

$f_{thres}^{(t+1)}=f_m\left(X_{N_{IM}}^m\right)---\left(14\right)$

According to such more new regulation, the utility function of sample will become closer to optimal solution.

It follows that algorithm is distributed according to significant samples update probability so that can be with more preferable probability in next iteration Generate the sample of " good ".SampleMiddle elementProbability distribution q^m _, ⁿ, can be according to N_IMIn each user (includeNo user situation) number of times that occurs is updated, i.e.

Wherein,Represent at N_IMIn individual sample, u occurs in the number of times of n-th of sample.Under In an iteration, algorithm can generate new N according to the probability distribution after updating_SAMIndividual sample.After iteration several times, probability Distribution q^m _, ⁿProgressively restrain.As whole q^m _, ⁿAll convergence with probability 1 is when a certain user, and algorithm i.e. obtains the optimal solution of RB scheduling, And the sample of the determination now obtained (sample generated with probability 1) is the optimal solution of problem.

Algorithm 1 summarizes the process in every single-step iteration.

In algorithm, t represents current iterations,Scheduling result for the TP m of the output of the algorithm when convergence in probability.In each element representation TP m correspondence position RB on scheduling user, represent with scheduling index and be Therefore, byScheduling index set can be obtained

3) modified result:

The result that Cross-Entropy Algorithm described above obtains, strictly meets restrictive condition C1～C3, but it cannot be guaranteed that All transmission all meets given threshold value (i.e. restrictive condition C4).The final mesh of the system resource optimized distribution discussed in the present invention Mark is to improve energy efficiency, for the purpose of the saving energy, the transmission that algorithm will select closedown can not meet qos requirement.

Power distribution algorithm based on KKT condition:

According to the dispatching algorithm based on cross entropy proposed, the solution of the permissible RB obtained schedulingShould Result substitutes into formula (5), i.e. can get an optimization problem about power distribution

$\begin{array}{cccc}\underset{p_m^n}{\max }& & \underset{k=1}{\overset{K}{\&Sigma;}}\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\frac{R_k^n}{{\stackrel{\&OverBar;}{R}}_k}/\underset{m=1}{\overset{M}{\&Sigma;}}\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}{p}_m^n& \\ s.t.& C1:& 0\&le;p_m^n\&le;S& \&ForAll;m,n\\ & C3:& \underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n{R}_k^n\&le;C_m& \&ForAll;m\\ & C4:& R_k^n\&GreaterEqual;R_{thres}& \&ForAll;k,n\end{array}---\left(16\right)$

Object function in formula (16) is a typical non-convex function, and simple method therefore cannot be utilized to obtain entirely Office's optimal solution.The present invention proposes a kind of analytic expression algorithm based on KKT condition, obtains optimization problem with relatively low computation complexity Locally optimal solution.

Setting up formula (16) about the Lagrange's equation of restrictive condition C3 and C4 is:

$\begin{array}{cc}\underset{p_m^n}{\max }& \underset{n=1}{\overset{N_{RB}}{\&Sigma;}}(\underset{k=1}{\overset{K}{\&Sigma;}}\frac{R_k^n}{{\stackrel{\&OverBar;}{R}}_k}/\underset{m=1}{\overset{M}{\&Sigma;}}{p}_m^n)-\underset{m=1}{\overset{M}{\&Sigma;}}{\mathrm{\&lambda;}}_m(\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n{R}_k^n-C_m)+\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&mu;}}_{n,k}(R_k^n-R_{thres})\\ s.t.& 0\&le;p_m^n\&le;S,\&ForAll;m,n\end{array}---\left(17\right)$

Wherein,WithIt it is the Lagrange's multiplier of non-negative.The Lagrange's equation of above formula is solved Imitate the problem set up in solution formula (16).If problem can be separated into the subproblem on each RB, then can reduce each excellent The number of the variable in change problem.Meanwhile, independent subproblem can be with parallel computation, thus time needed for reducing calculating.

OrderUnderstand, in the case of scheduling strategy is fixed, β_mDetermine that.Thus, C_m It is represented by:

$C_m=\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n\frac{C_m}{\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n}=\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n\frac{C_m}{{\mathrm{\&beta;}}_m}---\left(18\right)$

By C in above formula_mSubstitution formula (17), obtain object function

$\begin{array}{cc}\underset{p_m^n}{\max }& \underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\{\underset{k=1}{\overset{K}{\&Sigma;}}\frac{R_k^n}{{\stackrel{\&OverBar;}{R}}_k}/\underset{m=1}{\overset{M}{\&Sigma;}}{p}_m^n-\underset{k=1}{\overset{K}{\&Sigma;}}\underset{m=1}{\overset{M}{\&Sigma;}}{\mathrm{\&lambda;}}_m{\mathrm{\&beta;}}_{m,k}^n(R_k^n-\frac{C_m}{{\mathrm{\&beta;}}_m})+\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&mu;}}_{n,k}(R_k^n-R_{thres})\}\\ s.t.& 0\&le;p_m^n\&le;S,\&ForAll;m,n\end{array}---\left(19\right)$

Because the transmission on each RB is the most uncorrelated, so formula (19) can be resolved into N_RBIndividual independent optimization is asked Topic, and the solution obtained for each subproblem finally forms the optimal solution of formula (19).And independent subproblem can be by also The mode of row carries out calculation process, it is possible to reduce the time run needed for calculating.

Observing formula (19), subproblem independent on the most each RB can be expressed as:

$\begin{array}{cc}\underset{p_m^n}{\max }& \underset{k=1}{\overset{K}{\&Sigma;}}\frac{R_k^n}{{\stackrel{\&OverBar;}{R}}_k}/\underset{m=1}{\overset{M}{\&Sigma;}}{p}_m^n-\underset{k=1}{\overset{K}{\&Sigma;}}\underset{m=1}{\overset{M}{\&Sigma;}}{\mathrm{\&lambda;}}_m{\mathrm{\&beta;}}_{m,k}^n(R_k^n-\frac{C_m}{{\mathrm{\&beta;}}_m})+\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&mu;}}_{n,k}(R_k^n-R_{thres})\\ s.t.& 0\&le;p_m^n\&le;S,\&ForAll;m\end{array}---\left(20\right)$

Make f_nRepresenting the object function in formula (20), it is rightFirst derivative be

$\begin{array}{c}\frac{\&part;f_n}{\&part;p_m^n}=\frac{(\&part;(\underset{k=1}{\overset{K}{\&Sigma;}}{R}_k^n{\sqrt{R}}_k)\&part;p_m^n)p_{tot}^n-\underset{k=1}{\overset{K}{\&Sigma;}}{R}_k^n{\sqrt{R}}_k}{{\left(p_{tot}^n\right)}^2}-\underset{m=1}{\overset{M}{\&Sigma;}}{\mathrm{\&lambda;}}_m{\mathrm{\&beta;}}_{m,k}^n\left(\frac{\&part;\left(\underset{k=1}{\overset{K}{\&Sigma;}}{R}_k^n\right)}{\&part;p_m^n}\right)+\frac{\&part;\left(\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&mu;}}_{n,k}{R}_k^n\right)}{\&part;p_m^n}\\ =(\frac{1}{{\stackrel{\&OverBar;}{R}}_{k^{*}}\&CenterDot;p_{tot}^n}-\underset{m=1}{\overset{M}{\&Sigma;}}{\mathrm{\&lambda;}}_m\&CenterDot;{\mathrm{\&beta;}}_{m,k^{*}}^n+{\mathrm{\&mu;}}_{n,k^{*}})\&CenterDot;\frac{\&part;R_{k^{*}}^n}{\&part;p_m^n}+\underset{k=1;k\&NotEqual;k^{*}}{\overset{K}{\&Sigma;}}(\frac{1}{{\stackrel{\&OverBar;}{R}}_k\&CenterDot;p_{tot}^n}-\underset{m=1}{\overset{M}{\&Sigma;}}{\mathrm{\&lambda;}}_m\&CenterDot;{\mathrm{\&beta;}}_{m,k}^n+{\mathrm{\&mu;}}_{n,k})\&CenterDot;\frac{\&part;R_k^n}{\&part;p_m^n}\\ -\frac{1}{{\left(p_{tot}^n\right)}^2}\&CenterDot;\underset{k=1}{\overset{K}{\&Sigma;}}\frac{R_k^n}{{\stackrel{\&OverBar;}{R}}_k}\end{array}---\left(21\right)$

In formula,Represent and always launch power on every RB,k^*Represent and meetUser；ForWithTo powerFirst derivative, be expressed as follows respectively:

$\frac{\&part;R_{k^{*}}^n}{\&part;p_m^n}=\frac{b\underset{m\&Element;M}{\&Sigma;}{\mathrm{\&beta;}}_{m,k^{*}}^n\left|\right|H_{m,k^{*}}^n{w}_m^n|{|}^2}{\left|\right|H_{m,k^{*}}^n{w}_m^n|{|}^2{p}_m^n+\underset{m^{\&prime;}\&Element;M\backslash \left\{m\right\}}{\&Sigma;}|\left|H_{m^{\&prime;},k^{*}}^n{w}_{m^{\&prime;}}^n\right|{|}^2{p}_{m^{\&prime;}}^n+{\mathrm{\&sigma;}}^2}---\left(22\right)$

With

$\frac{\&part;R_k^n}{\&part;p_m^n}=-\frac{b{\left({\mathrm{\&gamma;}}_k^n\right)}^2}{1+{\mathrm{\&gamma;}}_k^n}\frac{\left|\right|H_{m,k}^n{w}_m^n|{|}^2}{\underset{m^{\&prime;}\&Element;M}{\&Sigma;}{\mathrm{\&beta;}}_{m^{\&prime;},k}^n\left|\right|H_{m^{\&prime;},k}^n{w}_{m^{\&prime;}}^n|{|}^2{p}_{m^{\&prime;}}^n},\&ForAll;k\&NotEqual;k^{*}---\left(23\right)$

In formula,Representing that transmission can obtain Signal to Interference plus Noise Ratio, when using DYNAMIC J T CoMP transmission technology, it is defined as

${\mathrm{\&gamma;}}_k^n=\frac{\underset{m\&Element;M}{\&Sigma;}{\mathrm{\&beta;}}_{m,k}^n\left|\right|H_{m,k}^n{w}_m^n|{|}^2{p}_m^n}{\underset{m^{\&prime;}\&Element;M}{\&Sigma;}(1-{\mathrm{\&beta;}}_{m^{\&prime;},k}^n)\left|\right|H_{m^{\&prime;},k}^n{w}_{m^{\&prime;}}^n|{|}^2{p}_{m^{\&prime;}}^n+{\mathrm{\&sigma;}}^2}---\left(24\right)$

KKT condition according to nonlinear programming problem, to obtain the power distribution of optimum, needs peer-to-peer groupSolve.Obviously, equation set is utilized to obtainAnalytic solutions.Accordingly, it would be desirable to consider iteration It is sought approximate solution by the method for numerical analysis.Assume forFor,It it is known and not phase Close, it is hereby achieved that following approximate solution

${\hat{p}}_m^n=\{\frac{(\frac{1}{{\stackrel{\&OverBar;}{R}}_{k^{*}}\&CenterDot;p_{tot}^n}-\underset{m=1}{\overset{M}{\&Sigma;}}{\mathrm{\&lambda;}}_m\&CenterDot;{\mathrm{\&beta;}}_{m,k^{*}}^n+{\mathrm{\&mu;}}_{n,k^{*}})\&CenterDot;b\underset{m\&Element;M}{\&Sigma;}{\mathrm{\&beta;}}_{m,k^{*}}^n\left|\right|H_{m,k^{*}}^n{w}_m^n|{|}^2}{\frac{1}{{\left(p_{tot}^n\right)}^2}\&CenterDot;\underset{k=1}{\overset{K}{\&Sigma;}}\frac{R_k^n}{{\stackrel{\&OverBar;}{R}}_k}-\underset{k=1;k\&NotEqual;k^{*}}{\overset{K}{\&Sigma;}}(\frac{1}{\stackrel{\&OverBar;}{R_k}\&CenterDot;p_{tot}^n}-\underset{m=1}{\overset{M}{\&Sigma;}}{\mathrm{\&lambda;}}_m\&CenterDot;{\mathrm{\&beta;}}_{m,k}^n+{\mathrm{\&mu;}}_{n,k})\&CenterDot;\frac{\&part;R_k^n}{\&part;p_m^n}}-A_m^n\}\&CenterDot;\frac{1}{\left|\right|H_{m,k^{*}}^n{w}_m^n|{|}^2}---\left(25\right)$

Wherein

$A_m^n=\underset{m^{\&prime;}\&Element;M\backslash \left\{m\right\}}{\&Sigma;}\left|\right|H_{m^{\&prime;},k^{*}}^n{w}_{m^{\&prime;}}^n|{|}^2{p}_{m^{\&prime;}}^n+{\mathrm{\&sigma;}}^2---\left(26\right)$

It should be noted that the power arrived in formula (26)Restrictive condition C1 might not be met, it is possible to existOrSituation.To this end, in order to obtain the distribution of suitable power, it is right to needFurther limited.? EventuallyResult be represented by:

$p_m^n=max\left\{\min \right\{{\hat{p}}_m^n,S\},0\}---\left(28\right)$

For the lagrangian multiplier in formula (26)_mAnd μ_n,kThe subgradient method used in document [12] can be used to enter Row solves.I.e.

${\mathrm{\&lambda;}}_m^{(t+1)}=max\{{\mathrm{\&lambda;}}_m^{\left(t\right)}-{\mathrm{\&nu;}}_{\mathrm{\&lambda;}}^{\left(t\right)}(\underset{n=1}{\overset{N_{RB}}{\&Sigma;}}\underset{k=1}{\overset{K}{\&Sigma;}}{\mathrm{\&beta;}}_{m,k}^n{R}_k^n-C_m),0\}---\left(29\right)$

${\mathrm{\&mu;}}_{n,k}^{(t+1)}=max\{{\mathrm{\&mu;}}_{n,k}^{\left(t\right)}-{\mathrm{\&nu;}}_{\mathrm{\&mu;}}^{\left(t\right)}(R_{thres}-R_k^n),0\}---\left(30\right)$

Wherein,WithRepresent λ respectively_mAnd μ_n,kThe step-length of iteration, t represents iterations.

For each RB, after using power distribution algorithm presented above to carry out the most independent calculating, just can be The power distribution result of system

Emulation experiment and interpretation of result:

In emulation, utilize SCM (Space Channel Model)^[13]The mimo channel of model generation urban area circumstance.In city In district's environment, owing to the barrier such as building, tree is more, it is generally recognized that there is not line-of-sight transmission between TP and user, therefore give birth to When becoming mimo channel, it is contemplated that shadow fading and penetration loss.The major parameter used in emulation is as shown in table 1.In network close Collection deploys 37 hexagonal cell, and the covering radius of every community is 250m.Fig. 3 shows TP and the phase of user in simulated environment To position.Wherein, outermost 18 TP do not produce the transmission of reality.Emulation have employed the mode of community coiling, makes outermost The TP of layer replicates the signal intensity that reality produces the TP of transmission, produces corresponding interference with simulation of real scenes.Community coiling Means often occur in emulating large scale network^[14]。

Table 1 simulation parameter

Fig. 4-6 illustrates the thresholding R as QoS_thresDuring for 540kbps, the different back haul link capacity shadows to systematic function Ring.As it can be seen, when back haul link capacity limit is 50Mbps, the handling capacity of system, energy efficiency and fairness all occur Significant decline.This is because, limited back haul link capacity cannot support enough JT CoMP transmission.Simulation result shows Showing, the JT CoMP transmission implemented in the averagely every TTI of system when the infinite capacity of back haul link is 130 times, and works as flyback line When appearance of a street amount drops to 50Mbps, herein only having 20 times of JT CoMP transmission.It addition, from simulation result it can be seen that work as backhaul When capacity of trunk is 100Mbps, the property indices of system is identical with during unlimited back haul link capacity.Therefore deduce that Conclusion, for the transmission system that the present invention considers, capacity is that the back haul link of 100Mbps can obtain sufficient system Energy.

Fig. 7 shows the energy efficiency of system under different QoS thresholding.It can be seen that QoS thresholding is to system capacity The impact of efficiency is inconspicuous.But, Fig. 8 illustrates that, in the case of QoS thresholding is higher, system is obtained in that higher averagely gulping down The amount of telling.The algorithm proposed due to the present invention closes and can not meet the transmission of QoS thresholding, and when therefore QoS thresholding is higher, system will It is reduced to the service that the bad user of channel condition provides.This mode reduces the energy expenditure of system, it is ensured that can dose-effect Rate, but correspondingly reduce the fairness (as shown in Figure 9) of user data rate.

The invention also discloses the Energy Efficient resource allocation system of a kind of heterogeneous network, including:

Receiver module, for receiving the parameter of input；

Generating probability module, is used for initializing allocative decision generating probability；

Generate sample module, for generating allocative decision sample according to generating probability；

Computing module, for calculating the target function value of each sample；

Screening module, for screening sample, and exports a scheduling result；

Whether judge module, restrain for evaluation algorithm, the most so performs correcting module, otherwise according to whole base stations Intermediate object program update generating probability and parameter and then perform to generate sample module；

Correcting module, according to scheduling result and constraint condition C 4 correction result of whole base stations；

Scheduling result output module, is used for exporting RB scheduling result；

Resource distribution module, power distribution algorithm based on KKT obtains the power distribution result optimized；

Resource distribution output module, exports Resource Allocation Formula.

The present invention, in HetNet downlink transmission system based on JT CoMP transmission, have studied the frequency towards energy efficiency Rate, power joint optimization problem.Optimization problem has considered the spectrum efficiency of system, energy efficiency and user data rate Fairness, have employed the object function maximizing weighted energy efficiency.Meanwhile, consider in resources configuration optimization multiple important Constraint.First, there is the upper limit in total power of launching of base station；Secondly, the present invention considers limited back haul link capacity；Finally, In order to ensure the quality of transmission every time, present invention introduces the constraint to the minimum data rate often transmitted.The present invention is by above-mentioned money Source optimization problem has carried out mathematical modeling, obtains a NP difficulty MIXED INTEGER process (Mixed Integer Programming,MIP).In order to optimization problem, MIP is resolved into two parts by the present invention: one about frequency scheduling Combinatorial problem and one about the continuous optimization problem of power distribution, and propose respectively based on intersection for two parts The dispatching algorithm of entropy and power distribution algorithm based on KKT condition.As shown by the simulation results, the resource using the present invention to propose is divided When joining algorithm, the energy efficiency of HetNet system based on JT CoMP transmission is significantly improved.

Above content is to combine concrete preferred implementation further description made for the present invention, it is impossible to assert Being embodied as of the present invention is confined to these explanations.For general technical staff of the technical field of the invention, On the premise of present inventive concept, it is also possible to make some simple deduction or replace, all should be considered as belonging to the present invention's Protection domain.

Claims (10)

1. the Energy Efficient resource allocation methods of a heterogeneous network, it is characterised in that comprise the steps:

A. receiving step, receives the parameter of input；

B. generating probability step, initializes allocative decision generating probability；

C. generate sample step, generate allocative decision sample according to generating probability；

D. calculation procedure, calculates the target function value of each sample；

E. screen step, sample is screened, and export a scheduling result；

F. judge step, it is judged that whether algorithm restrains, the most so perform to revise step, otherwise according to the centre of whole base stations Result updates generating probability and then parameter performs to generate sample step；

G. step is revised, according to scheduling result and constraint condition C 4 correction result of whole base stations；

H. scheduling result output step, exports RB scheduling result；

I. resource allocation step, power distribution algorithm based on KKT obtains the power distribution result optimized；

J. resource distribution output step, exports Resource Allocation Formula.

Energy Efficient resource allocation methods the most according to claim 1, it is characterised in that in described generating probability step In, TP m probability distribution initial value of RB scheduling on RB n is expressed as

TP m probability distribution q that user selects on RB n is can get according to above formula^m _, ⁿ, and then TP m can be obtained on whole RB User's select probability is distributed

Energy Efficient resource allocation methods the most according to claim 1, it is characterised in that in described screening step, The capacity knowing the back haul link of TP m is C_m, during generating sample, if the handling capacity of the TP m of sample generationSampleTo directly be removed；Similarly, orderRepresent the door of sample utility function in the t time iteration Limit, can not reach the sample of requirement for utility function(i.e.) also will be removed；Raw according to above-mentioned requirements Become N_SAMIndividual effective sample, is designated as

Energy Efficient resource allocation methods the most according to claim 3, it is characterised in that in described screening step, right N_SAMThe utility function of individual effective sample carries out descending, it is assumed thatSet one Individual quantile ρ (0≤ρ≤1), for the sample of descending, intercepts whereinIndividual sample is as important Sample, and using these samples as the foundation of update probability, and the utility function thresholding generating effective sample will be stepped upEvery time after iteration, this thresholding will be updated to the minima of utility function in significant samples, i.e.

$f_{thres}^{(t+1)}=f_m\left(X_{N_{IM}}^m\right)---\left(14\right).$

Energy Efficient resource allocation methods the most according to claim 1, it is characterised in that resource distribution output step, defeated Go out Resource Allocation Formula and include calculating the handling capacity of current time, average fairness coefficient, energy efficiency.

6. the Energy Efficient resource allocation system of a heterogeneous network, it is characterised in that including:

Receiver module, for receiving the parameter of input；

Generating probability module, is used for initializing allocative decision generating probability；

Generate sample module, for generating allocative decision sample according to generating probability；

Computing module, for calculating the target function value of each sample；

Screening module, for screening sample, and exports a scheduling result；

Whether judge module, restrain for evaluation algorithm, the most so performs correcting module, otherwise according in whole base stations Between result update generating probability and parameter and then perform to generate sample module；

Correcting module, according to scheduling result and constraint condition C 4 correction result of whole base stations；

Scheduling result output module, is used for exporting RB scheduling result；

Resource distribution module, power distribution algorithm based on KKT obtains the power distribution result optimized；

Resource distribution output module, exports Resource Allocation Formula.

Energy Efficient resource allocation system the most according to claim 6, it is characterised in that in described generating probability module In, TP m probability distribution initial value of RB scheduling on RB n is expressed as

TP m probability distribution q that user selects on RB n is can get according to above formula^m _, ⁿ, and then TP m can be obtained on whole RB User's select probability is distributed

Energy Efficient resource allocation system the most according to claim 6, it is characterised in that in described screening module, The capacity knowing the back haul link of TP m is C_m, during generating sample, if the handling capacity of the TP m of sample generationSampleTo directly be removed；Similarly, orderRepresent the door of sample utility function in the t time iteration Limit, can not reach the sample of requirement for utility function(i.e.) also will be removed；Raw according to above-mentioned requirements Become N_SAMIndividual effective sample, is designated as

Energy Efficient resource allocation system the most according to claim 8, it is characterised in that in described screening module, right N_SAMThe utility function of individual effective sample carries out descending, it is assumed thatSet one Individual quantile ρ (0≤ρ≤1), for the sample of descending, intercepts whereinIndividual sample is as important Sample, and using these samples as the foundation of update probability, and the utility function thresholding generating effective sample will be stepped upEvery time after iteration, this thresholding will be updated to the minima of utility function in significant samples, i.e.

Energy Efficient resource allocation system the most according to claim 6, it is characterised in that in the distribution output of described resource In module, output Resource Allocation Formula includes the handling capacity calculating current time, average fairness coefficient, energy efficiency.