CN105813208A - Method and system for dynamic resource distribution of multi-user orthogonal frequency division multiplexing system - Google Patents
Method and system for dynamic resource distribution of multi-user orthogonal frequency division multiplexing system Download PDFInfo
- Publication number
- CN105813208A CN105813208A CN201510872169.0A CN201510872169A CN105813208A CN 105813208 A CN105813208 A CN 105813208A CN 201510872169 A CN201510872169 A CN 201510872169A CN 105813208 A CN105813208 A CN 105813208A
- Authority
- CN
- China
- Prior art keywords
- user
- sigma
- subcarrier
- neural network
- neutral net
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W72/00—Local resource management
- H04W72/04—Wireless resource allocation
- H04W72/044—Wireless resource allocation based on the type of the allocated resource
- H04W72/0453—Resources in frequency domain, e.g. a carrier in FDMA
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W72/00—Local resource management
- H04W72/50—Allocation or scheduling criteria for wireless resources
- H04W72/53—Allocation or scheduling criteria for wireless resources based on regulatory allocation policies
Abstract
The invention discloses a neural network based method for dynamic resource distribution of a multi-user orthogonal frequency division multiplexing system. According to the method, a neural network model of a three-dimensional neuron matrix cube is established, wherein the neural network model expresses a mapping relation between one subcarrier and a user and a mapping relation of a bit number modulated by each subcarrier; and according to an energy function of the neural network model, the dynamic resource distribution of the multi-user orthogonal frequency division multiplexing system is carried out, and base station emission total power can be minimized under the condition that the rate demand of each user and an error rate requirement are satisfied. The neural network model of the three-dimensional neuron matrix cube is established, the energy function of the neural network model takes the data rate demand of each user, an error rate and constraint conditions of subcarrier distribution into consideration, under the condition that the rate demand of each user and the error rate requirement are satisfied, the base station emission total power can be minimized, and the requirement for real-time performance of the dynamic resource distribution of the multi-user orthogonal frequency division multiplexing system can be met.
Description
Technical field
The present invention relates to the channel between cellular cell or frequency distribution technique field, particularly to one based on noise chaos
The method of the multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia of neural network algorithm (hereinafter referred to as neutral net) and be
System.
Background technology
In community, the Radio Resource allocative decision of multi-user orthogonal frequency division multiplexing system can be divided into two big classes: fixed resource
Distribution and Dynamic Resource Allocation for Multimedia.
Fixed resource distribution is to distribute fixing subcarrier to each user, and the modulation system of each subcarrier, sends out
Penetrate power the most constant.Its advantage is that algorithm is simple, but the availability of frequency spectrum is the lowest, and network throughput is the biggest.
Dynamic Resource Allocation for Multimedia is according to each subcarrier instantaneous channel conditions information to each user, always by channel condition
Good subcarrier distributes to most suitable user, thus substantially increases the availability of frequency spectrum and network throughput, but dynamically provides
The algorithm of source distribution is complicated, and requirement of real-time is high, and network signaling overhead is big.Need to consider the Real-time Channel decline spy of each subcarrier
Levy, according to bit error rate requirement and the required communication rate of user, to the subcarrier that channel condition is good distribute more bit number,
Or reduce transmitting power;And give the subcarrier of bad channel conditions (deep fade), then distribute less bit number or increase transmitting merit
Rate (does not even distribute bit number, zero launches power)
The optimization aim of Dynamic Resource Allocation for Multimedia involved in the present invention, is the feelings given at each user rate and the bit error rate
Under condition, make cell transmission general power minimum.
For the Dynamic Resource Allocation for Multimedia of multi-user orthogonal frequency division multiplexing system, the subcarrier that each user is distributed, at frequency
Being no longer continuous print on territory, in order to the subcarrier that channel condition is good is distributed to more suitably user, adjacent sub-carrier is likely to
Different users can be distributed to.
In the case of each user rate and the bit error rate are given, in order to make cell transmission general power minimum,
Good optimized algorithm is greedy algorithm, and greedy algorithm first supposes the bit number identical to the distribution of each subcarrier, it is then assumed that give
Each subcarrier increases by one one by one and sends bit, then calculates the required transmitting power increased of each subcarrier, finally selects
Take the subcarrier making transmitting power increment value minimum, this subcarrier increases by one and sends bit.As can be seen here, greedy
Algorithm needs to carry out substantial amounts of sequence and search arithmetic, and computational complexity is the highest, is unfavorable in voice and high-speed data communication
Requirement of real-time;And greedy algorithm may cause certain user to distribute too much or very few subcarrier and bit number, thus shadow
Ring the fairness between user.
But, according to the time-varying characteristics of wireless channel, the coherence time of mobile telecommunication channel is generally several ms, multiplex
The Dynamic Resource Allocation for Multimedia of family ofdm system, it is desirable to will perform once every several ms, domestic and international existing document
Research to multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia problem, efficiency of algorithm is the most not high enough, it is impossible to full well
The requirement of real-time of foot multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia.
Summary of the invention
Efficiency of algorithm for solving in existing multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia is not high enough, it is impossible to
The problem meeting the requirement of real-time of multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia well, it is necessary to one is provided
The method and system of multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net.
A kind of method of multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net, comprising:
Setting up three-dimensional nerve variable matrix cubical neutral net neural network model, neural network model represents a son
Carrier wave and the mapping relations of user, and the mapping relations of bit number that each subcarrier is modulated;
The energy function of neural network model is as follows:
Wherein, N is OFDM OFDM total number of sub-carriers;K is total number of users in community; RkIt is to use
The data rate of family k;BERkBit error rate for user k;hk,nIt it is the subcarrier n fading channel amplitude for user k;
vk,n,bFor the row k in three-dimensional nerve variable matrix cube, the n-th row, the b three-dimensional nerve unit;B represents the b neuron
The numbering of two-dimensional matrix;The bit number distributed on corresponding subcarrier n is C (b), and C (b) is the function of b, when the value of b is
During natural number, then the value of C (b) is set D={0,1 ..., CTIn } one, now corresponding subcarrier modulation modes is 2C (b)The orthogonal amplitude QAM modulation of constellation, CTFor the modulation bit number that subcarrier is maximum allowable.
Energy function according to neural network model carries out multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia so that
In the case of meeting each user rate demand and bit error rate requirement, make Base Transmitter general power minimum.
A kind of system of multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net, comprising:
Establishment of Neural Model unit, is used for setting up three-dimensional nerve variable matrix cubical neutral net neutral net mould
Type, neural network model represents a subcarrier and the mapping relations of user, and the bit number that each subcarrier is modulated
Mapping relations.
The energy function of neural network model is as follows:
Wherein, N is OFDM OFDM total number of sub-carriers;K is total number of users in community;RkIt is user
The data rate of k;BERkBit error rate for user k;hk,nIt it is the subcarrier n fading channel amplitude for user k;vk,n,b
For the row k in three-dimensional nerve variable matrix cube, the n-th row, the b three-dimensional nerve unit;B represents the b neuron two dimension
The numbering of matrix;The bit number distributed on corresponding subcarrier n is C (b), and C (b) is the function of b, when the value of b is nature
During number, then the value of C (b) is set D={0,1 ..., CTIn } one, now corresponding subcarrier modulation modes is 2C(b)Star
The orthogonal amplitude QAM modulation of seat, CTFor the modulation bit number that subcarrier is maximum allowable.
Dynamic Resource Allocation for Multimedia unit, for carrying out multi-user orthogonal frequency division multiplexing according to the energy function of neural network model
System dynamics resource is distributed so that in the case of meeting each user rate demand and bit error rate requirement, make Base Transmitter total
Power is minimum.
The present invention provide multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net method and
System, by setting up three-dimensional nerve variable matrix cubical neutral net neural network model, the energy in this neural network model
Flow function considers the data-rate requirements of each user and the bit error rate and the constraints of subcarrier distribution, every meeting
In the case of individual user rate demand and bit error rate requirement, it is possible to make Base Transmitter general power minimum, can meet many well
The requirement of real-time of user's ofdm system Dynamic Resource Allocation for Multimedia.
Accompanying drawing explanation
Fig. 1 is the multi-user orthogonal frequency division multiplexing system dynamic resource based on neutral net that embodiment of the present invention provides
The method and system structured flowchart of distribution;
Fig. 2 is the wireless channel model figure of single sub-carrier;
Fig. 3 is single Clarke flat fading channel model emulation figure;
Fig. 4 is energy function curve figure;
Fig. 5 is average bit SNR performance chart.
Detailed description of the invention
As it is shown in figure 1, the embodiment of the present invention provides a kind of multi-user orthogonal frequency division multiplexing system based on neutral net to move
The method of state resource distribution, comprising:
S1, setting up three-dimensional nerve variable matrix cubical neutral net neural network model, neural network model represents one
Individual subcarrier and the mapping relations of user, and the mapping relations of bit number that each subcarrier is modulated.
The energy function of neural network model is as follows:
Wherein, N is OFDM OFDM total number of sub-carriers;K is total number of users in community;RkIt is user
The data rate of k;BERkBit error rate for user k;hk,nIt it is the subcarrier n fading channel amplitude for user k;vk,n,b
For the row k in three-dimensional nerve variable matrix cube, the n-th row, the b three-dimensional nerve unit;B represents the b neuron two dimension
The numbering of matrix;The bit number distributed on corresponding subcarrier n is C (b), and C (b) is the function of b, when the value of b is nature
During number, then the value of C (b) is set D={0,1 ..., CTIn } one, now corresponding subcarrier modulation modes is 2C(b)Star
The orthogonal amplitude QAM modulation of seat, CTFor the modulation bit number that subcarrier is maximum allowable.
S2, energy function according to neural network model carry out multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia,
Make in the case of meeting each user rate demand and bit error rate requirement, make Base Transmitter general power minimum.
S3, neural network model is asked three-dimensional nerve unit partial derivative, it is thus achieved that the state equation of neutral net.
S4, by Euler's method, by the state equation of neutral net to time discretization, and add the self feed back chain of decay
Connect power and random noise, obtain neutral net discrete time model.
In the energy function of neural network model, Section 1 represents the total emission power of base station, for each neuron vk,n,b
Likely combine, by the continuous iteration of neutral net, it is thus achieved that minima, the most each user rate and the bit error rate are given
In the case of, make cell transmission general power minimum.
In the energy function of neural network model, Section 2 represents the constraints of the rate requirement of user, the most all distribution
To the bit number modulated on the subcarrier of user k, its summation is just equal to the rate requirement of kth user;When Section 2
When value is equal to zero, then meet the rate requirement of each user.
In the energy function of neural network model, Section 3, Section 4 represent the constraints that the distribution of subcarrier requires.
For the three-dimensional nerve unit v that footnote is nk,n,b, the combination of its all footnote k and b, referred to as k × b plane is neural
Unit;Section 3 in the energy function of neural network model so that footnote is in the two-dimensional matrix of k × b plane neuron of n, right
In the combination of all footnote k and b, make the summation of value of all plane neurons of k × b plane equal to 1.
Section 4 in the energy function of neural network model, makes in k × b plane neuron, most only one of which planes god
It is non-zero through first value, the value all 0 of remaining plane neuron.
The embodiment of the present invention also provides for a kind of multi-user orthogonal frequency division multiplexing system dynamic resource based on neutral net and divides
The system joined, comprising:
Establishment of Neural Model unit, is used for setting up three-dimensional nerve variable matrix cubical neutral net neutral net mould
Type, neural network model represents a subcarrier and the mapping relations of user, and the bit number that each subcarrier is modulated
Mapping relations.
The energy function of neural network model is as follows:
Wherein, N is OFDM OFDM total number of sub-carriers;K is total number of users in community;RkIt is user
The data rate of k;BERkBit error rate for user k;hk,nIt it is the subcarrier n fading channel amplitude for user k;vk,n,b
For the row k in three-dimensional nerve variable matrix cube, the n-th row, the b three-dimensional nerve unit;B represents the b neuron two dimension
The numbering of matrix;The bit number distributed on corresponding subcarrier n is C (b), and C (b) is the function of b, when the value of b is nature
During number, then the value of C (b) is set D={0,1 ..., CTIn } one, now corresponding subcarrier modulation modes is 2C(b)Star
The orthogonal amplitude QAM modulation of seat, CTFor the modulation bit number that subcarrier is maximum allowable.
Dynamic Resource Allocation for Multimedia unit, for carrying out multi-user orthogonal frequency division multiplexing according to the energy function of neural network model
System dynamics resource is distributed so that in the case of meeting each user rate demand and bit error rate requirement, make Base Transmitter total
Power is minimum.
The state equation acquiring unit of neutral net, with the partial derivative that neural network model is asked three-dimensional nerve unit, it is thus achieved that
The state equation of neutral net.
Neutral net discrete time model unit, for by Euler's method, by the state equation of neutral net to the time
Discretization, and add self feed back link power and the random noise of decay, obtain neutral net discrete time model.
In the energy function of neural network model, Section 1 represents the total emission power of base station, for each neuron vk,n,b
Likely combine, by the continuous iteration of neutral net, it is thus achieved that minima, the most each user rate and the bit error rate are given
In the case of, make cell transmission general power minimum.
In the energy function of neural network model, Section 2 represents the constraints of the rate requirement of user, the most all distribution
To the bit number modulated on the subcarrier of user k, its summation is just equal to the rate requirement of kth user;When Section 2
When value is equal to zero, then meet the rate requirement of each user.
In the energy function of neural network model, Section 3, Section 4 represent the constraints that the distribution of subcarrier requires.
For the three-dimensional nerve unit v that footnote is nk,n,b, the combination of its all footnote k and b, referred to as k × b plane is neural
Unit;Section 3 in the energy function of neural network model so that footnote is in the two-dimensional matrix of k × b plane neuron of n, right
In the combination of all footnote k and b, make the summation of value of all plane neurons of k × b plane equal to 1.
Section 4 in the energy function of neural network model, makes in k × b plane neuron, most only one of which planes god
It is non-zero through first value, the value all 0 of remaining plane neuron.
Hereinafter the principle of the embodiment of the present invention is described further:
In order to make the total emission power of base station in multi-user orthogonal frequency division multiplexing system community minimum, formulae express can be used
As follows:
To formula (1), there are two constraintss, as follows:
Constraints 1:
Constraints 2:ck,n≠ 0, (ck′,n=0, k ≠ k ', n=1,2,3 ... N) (3)
In the most various, PTIt it is the total emission power of base station;N is OFDM total number of sub-carriers;K is total in community
Number of users;RkIt it is the data rate (unit: bits/ symbol) of user k;BERkBit error rate for user k;ck,nFor user k
The bit number of distribution on subcarrier n, it is determined by modulation system, if the data of subcarrier c to be sent bit, then corresponding
Modulation system is 2cThe QAM modulation of constellation, ck,nFor gathering D={0,1 ..., the element in C};hk,nIt is that channel fading causes
Amplitude gain, i.e. subcarrier n are for the fading channel amplitude of user k.
Constraints 1 represents the rate requirement of user k, and it is distributed equal on the subcarrier of all user of distributing to k
The summation of bit number;Constraints 2 represents that the distribution requirement of subcarrier, a subcarrier can only distribute to a user.
Function f (g) in formula (1) represents when channel gain is 1, under the conditions of certain error rate constraint, receives c bit number
According to required transmitting power, it with the relational expression of c and the bit error rate is:
In formula (4), N0For noise power spectral density, Q-1G () is the inverse function of Q (g), and have:
Whether one channel (or frequency) is distributed to a user, only need to represent with 0 and 1, such as neuron vxi=1
It is (or unallocated: v that (or 0) represents that channel i is allocated to cell xxi=0).
But for the subcarrier of multi-user orthogonal frequency division multiplexing system, bit, power joint assignment problem, not only need to examine
Consider whether a subcarrier distributes to a user, in addition it is also necessary to according to real-time channel condition information, distribute to each subcarrier
Corresponding bit number.So the information representated by the neuron of a neutral net, do not require nothing more than and can represent a subcarrier
With the mapping relations (two-dimensional matrix) of user, also requirement can map out the bit number that each subcarrier is modulated, and this just needs
Three-dimensional matrice to be used represents.
Therefore, neural network model can not use a two-dimentional neuron matrix again, and should use the nerve of three-dimensional
Variable matrix, each neuron of this three-dimensional matrice is expressed as v by the embodiment of the present inventionk,n,b, footnote k represents Customs Assigned Number, n table
Show that subcarrier number, b represent the bit number that subcarrier is distributed.Analyze based on above, the neutral net of optimized algorithm of the present invention
Energy function is as follows:
In formula (6), vk,n,bFor the row k in three-dimensional nerve variable matrix cube, the n-th row, the b neuron.The present invention
Simply represent the numbering of the b neuron two-dimensional matrix with b, but the bit number distributed on the subcarrier n of correspondence is C
B (), C (b) is the function of b, and its functional value is required to determine by concrete subcarrier-modulated.It is so natural number when the value of b
Time (TLAB specifies the row, column of three-dimensional matrice, the numbering of layer must be natural number), then the value of C (b) for set D={0,
1,…,CTIn } one, now corresponding subcarrier modulation modes is 2C(b)The QAM modulation of constellation, CTPermit for subcarrier maximum
The modulation bit number permitted.
In formula (6), Section 1 is that the energy function of formula (1) realizes, and is the total emission power of base station, as neuron vk,n,b
=1, represent that the n-th subcarrier is allocated to kth user, and the bit number that this subcarrier is distributed is C (b), f (C (b),
BERk) definition identical with formula (4), be only merely independent variable and become C (b) by c, the bit error rate becomes BERk;hk,nDefinition
Identical with formula (1).The meaning of Section 1 is, for each neuron vk,n,bLikely combine, by neutral net not
Disconnected iteration is run, and is eventually found a minima, and namely minimum Base Transmitter general power, to realize the target optimized.
Section 2 is that the energy function of formula (2) realizes, and the namely realization of constraints 1 the most all distributes to user k's
The bit number modulated on subcarrier, its summation is just equal to the rate requirement of kth user.When the value of Section 2 is equal to zero
Time, then meet the rate requirement of each user.
Section 3 and Section 4 are that the energy function of formula (3) realizes, and namely the energy function of constraints 2 realizes, its
Purpose is to allow each subcarrier at most be assigned to only a user, or a user does not distributes, for idle condition.For ease of rear
Continued is stated, for the neuron v that footnote is nk,n,b, the combination of its all footnote k and b, the present invention referred to as " (k × b) plane god
Through unit ", these neurons can be considered as in the case of footnote n is some fixed value, the two-dimentional neuron of k row b row
Matrix.In (k × b) plane neuron, can only have the value of a neuron equal to 1, remaining neuron is all necessarily equal to 0
(i.e. one subcarrier is assigned to only a user), only in this way, just can ensure that neutral net found is an efficient solution,
It is unlikely to be absorbed in a trivial solution.
In order to reach this purpose, formula (6) is divided into two parts and realizes by energy function, i.e. Section 3 and Section 4.The
In three two-dimensional matrixs being to ensure that (k × b) plane neuron that footnote is n, for the combination of all footnote k and b, make (k ×
B) summation of the value of all neurons of plane is equal to 1.But Section 3 also can only ensure the value of (k × b) plane neuron
Summation equal to 1, and it cannot be guaranteed that the value of only one of which neuron is equal to 1 in (k × b) plane neuron, remaining is equal to 0
(i.e. one subcarrier is assigned to only a user), so only only having Section 3, its constraint dynamics is inadequate.Because at (k
× b) in plane neuron, it is possible to there is the value of multiple neuron less than 1, but itself and the value situation equal to 1, such as certain
The value of two neurons is respectively equal to 0.5, and value is 0.5+0.5=1.So, guarantee (k × b) plane neuron in Section 3
And equal on the basis of 1, then by Section 4, make in (k × b) plane neuron, the value of most only one of which neurons is
Non-zero, remaining is entirely 0.So in Section 4, allowing (k × b) plane neuron be multiplied two-by-two, the most each neuron, all
It is multiplied with other neurons in addition to self, if the value of Section 4 is equal to zero, the most only two kinds situations: is that (k × b) puts down
The value of all neurons in face is equal to 0;Two is that only one of which neuron is equal to non-zero, and remaining is entirely 0.
As a result, the Section 3 of formula (6) guarantee (k × b) plane neuron and equal to 1, then guaranteed by the Section 4 of formula (6)
In (k × b) plane neuron, only one of which is non-zero, and the common effect of two ensures that in (k × b) plane neuron, only
Having one is 1 (non-zero), and remaining is all 0.
First subitem of the Section 4 of formula (6) is that (k × b) plane neuron is multiplied (except self) two-by-two at b row;
Second subitem is that (k × b) plane neuron is multiplied (except self) two-by-two in row k;3rd subitem is (k × b) plane
Neuron is multiplied (except self) at different row two-by-two with different row.As a result, these three subitem, just make (k × b) plane god
It is multiplied with other neurons in addition to self through first each.If so the value of Section 4 is equal to zero, then inevitable
Being that the value of each neuron is equal to 0, or only one of which neuron is equal to non-zero, remaining is entirely 0.
Although Section 3 and Section 4 ensure that each subcarrier is assigned to only a user, but also require that each subcarrier
At least must distribute to a user, it is impossible to do not distributed by the free time.So, if the rate requirement of user is little, communication load is relatively
Time low, unnecessary subcarrier the most just can be made not distribute, remain idle condition, always launch the mesh of power with minimizing
??
In order to solve this problem, the present invention, in the Section 1 and Section 2 of energy function formula (6), is not to use god
Through unit vk,n,bFootnote b (natural number) directly represent the bit number that subcarrier is distributed, but have employed the function C of footnote b
B () represents the bit number that subcarrier is distributed.The most not only ensure that the seriality (natural number) of footnote b, but also ensure
The noncontinuity of function C (b) of footnote b.Because the excursion of footnote b, inside matrix, it is desirable to be from 1 company started
Continuous natural number, and the bit number that each subcarrier is distributed, then be C (b) bit.If C (b)=0, then it is assumed that this subcarrier
Unallocated bit number, the most idle unallocated.Programming for convenience, the present invention C (1) defined in energy function formula (6)=
0, i.e. with neuron vk,n,1=1 (footnote b=1), represents the unallocated bit number of subcarrier, is i.e. not yet assigned to any user, for
Idle.
For formula (6), the definition of optimal solution is: energy function second and third, four be all zero, even if Section 1 is neural
Network reruns down, and its value declines the most again, then it is assumed that neutral net have found an optimal solution.
The neuron of formula (6) is not one-dimensional, but have employed a three-dimensional matrice V (k row n row b layer) and represent,
vk,n,bFor matrix V row k, n-th row, an element of b layer, then neutral net operates in the shape that (m × n × b) ties up
State space, the vector of the m × n that all neurons in matrix V are constituted × b dimension, then represent in this state space
One point.
Energy function formula (6) is sought vk,n,bPartial derivative, the state equation just obtained:
In formula (7), uk,n,bFor neuron vk,n,bInternal state, neuron vk,n,bOutput vk,n,b(t) and internal state
uk,n,bT the relational expression of (), i.e. neuron activation functions be:
Using Euler's method, by formula (7) to time discretization, and the self feed back link adding decay is weighed and random noise,
Obtain can software emulation realize discrete time model:
In formula (9), α is the zooming parameter of neuron input, simulated annealing function zk,n,b(t), the random noise of decay
nk,n,bT () is defined as follows:
zk,n,b(t)=(1-β1)zk,n,b(t-1) (10)
Am[nk,n,b(t+1)]=(1-β2)Am[nk,n,b(t)] (11)
In above two formulas, β1And β2The self feed back decay factor being respectively and the decay factor of noise amplitude.
It is below application example and proof of algorithm
In order to the optimized algorithm of multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia in community is emulated, need
To each mobile subscriber in each subcarrier of OFDM and community, simulate a real mobile communication ring
Border, this is accomplished by the channel gain simulating each subcarrier to each mobile subscriber, i.e. carries out Channel Modeling.On this basis,
Further according to rate requirement and the bit error rate requirement of each mobile subscriber, optimized algorithm is emulated, and carry out Performance Evaluation.
Multidiameter fading channel models
In general, the path that wireless signal arrives receiver from transmitter has a plurality of, referred to as multipath signal.And it is every
The signal of paths arrives after time of receiver has and first have, and there is relative time delay, if these relative time delays are much smaller than one
The time of OFDM symbol, then it is believed that these multipath signals arrive receiver the most simultaneously.This situation
Lower multipath transmisstion does not results in the interference between OFDM symbol, but signal amplitude fluctuating in time can be caused to become
Changing, this change is referred to as decline, and owing to this decline makes the frequency response of channel be smooth in frequency range used, institute
With referred to as flat fading because its change with frequency it doesn't matter, so declining also known as non-frequency-selective.
If on the contrary, can not ignore compared with the time of a symbol relative time delay of multipath signal, then work as multichannel
During signal superposition, the symbol of different time will overlap together, causes intersymbol interference.This decline is referred to as frequency and selects
Property decline because now the frequency response of channel is not smooth in frequency range used.
Emulate each OFDM subcarrier to each mobile subscriber's channel gain, first have to analyze single sub-load
The ripple channel model to certain user.Fig. 2 gives a wireless channel model being applicable to single sub-carrier.
In Fig. 2, transmitting signal is S0T (), reception signal is R (t), and whole channel model is made up of, often n bar propagation path
The propagation delay of paths is τn, channel gain is an, every paths is all modeled as a Clarke flat fading channel model.
It should be noted that τ in Fig. 2nValue more than an OFDM symbol time (non-flat forms decline), and every
In individual Clarke channel model, having assumed that M bar multipath signal, the relative time delay of these multipath signals is less than an orthogonal frequency
The multiplexed symbols time (flat fading).
The transmitting signal assuming base station is:
In formula (12), ω0For carrier frequency, φ0For first phase, S0T the signal amplitude of () is(volt), power is (watt),
It is assumed here that signal amplitude is(volt), is to make S0T the power of () is (watt).
As a example by nth bar path, the propagation delay in this path is τn, path loss is an, it is assumed that the signal warp on this path
Arriving receiver after crossing secondary reflection or scattering, i.e. have paths to arrive receiver, accepting signal is signal sn(t);And assume to be somebody's turn to do
On path, the amplitude of this reflection (or scattering) signal is(volt), power is(watt), and assume that this reflects
The relative delay of (or scattering) signal is the least, is not enough to bring intersymbol interference, thus it is believed that paths signal is one
Individual flat fading signal, here it is famous flat fading channel model.
The amplitude assuming above-mentioned signal and the azimuth arriving reception antenna are random and meet statistical iteration, then arrive
The signal reaching receiver can be expressed as:
In formula (13), v is the translational speed of user, and unit is m/s (meter per second);θiIt it is the i reflection (or scattering) signal
Angle of incidence, i.e. the initial phase of i-th reflection (or scattering) signal, obeys being uniformly distributed between π;λ is the ripple of carrier wave
Long.The Doppler frequency shift of then i-th reflection (or scattering) signal is:
For the sake of easy analysis, use A herenT () represents because of Doppler frequency shift, and the twinkling signal amplitude caused becomes
Change:
Then formula (13) can be simplified shown as:
According in Fig. 2, be added by n bar multipath signal, then obtaining receiving signal R (t) can be expressed as:
Because of the embodiment of the present invention it has assumed that launch signal S0T the power of () is equal to 1 (watt), thus receive the power of signal i.e.
For channel gain, it may be assumed that
From formula (18), channel gain can change by t over time, and this change is caused by Doppler frequency shift, belongs to
In time selective fading;It addition, channel gain also can be along with the frequencies omega of carrier wave0Delay time T with channelnChange, this
Change is caused by multipath transmisstion, belongs to frequency selective fading.
It is pointed out that for easy analysis, the embodiment of the present invention is it has assumed that the channel propagation impairments of the 0th paths is
Zero, i.e. channel gain is a0=e0=1, other paths (1 to 5) loss both greater than zero, so the statistical average of whole channel gain
Value, also will approximate more than 1And the assembly average of the time dispersive channel gain in Practical Project
It is only possible to less than 1, not more than 1 (instantaneous channel gain is likely larger than 1).
In order to verify above-mentioned phantom, the present invention frequencies omega to a carrier frequency0=900MHz, translational speed are v=
50km/h, the single flat fading channel model that number is M=200 of reflection (or scattering) signal is emulated, and emulation is held
The continuous time is 1 second, and time sampling interval is 1 millisecond, then obtain channel gain such as Fig. 3 institute of single flat fading channel model
Show.
It can be seen from figure 3 that when channel conditions is good, the channel gain of flat fading signal can reach more than 5dB, deeply
During decline, channel gain is less than-25dB, and the excursion of channel gain reaches more than 30dB, in this result and Practical Project
Channel gain change substantially the most consistent.
On the basis of Fig. 3, according still further to shown in Fig. 2, n bar Clarke flat fading channel is overlapped, then obtain as
The channel gain of the single OFDM subcarrier shown in formula (18), i.e. frequency selective fading channels gain is (non-flat
Smooth).
On the basis of the single sub-carrier channel model shown in Fig. 2, the present invention is for 32 subcarrier, 8 users
Multi-user orthogonal frequency division multiplexing system carries out channel simulator, and the channel gain just obtained is as shown in table 1:
The OFDM sub-carrier channels gain of table 1 32 subcarrier 8 user
The duration that data in table 1 being for emulation, certain time sampling point is randomly drawed and obtain, change over time
Changing, the data in table 1 also can change, but for the channel gain of a special time sampled point, just represents current multi-user
The each subcarrier of the ofdm system instantaneous channel gain to each user.
In table 1, line number represents 32 subcarriers, and column number represents 8 users, by observing every a line and every string
Data can be shown that following features:
1) data of every a line do not have dependency, this is because the position of each mobile subscriber is random, because of
And the channel gain of any one OFDM subcarrier is for each user, also it is random, so each subcarrier
The channel gain of each user the most do not had dependency.
2) data of every string have dependency, and all OFDM subcarriers are for same user, adjacent
The frequency interval of subcarrier is relatively near, the impact of multipath fading, causes frequency selective fading, but the channel between adjacent sub-carrier
Gain, numerically will not produce big saltus step;
3) all row and the numerical value of all row, its meansigma methods be slightly larger than 1 (1 should be less than in Practical Project), this be because of
For in fig. 2, present invention assumes that the channel propagation impairments of the 0th paths is zero, make the meansigma methods of whole channel gain to be more than
1;
4) any one subcarrier, while the least to certain subscriber channel gain (being in deep fade), also tends to it
The channel gain that he certain user comes is compared with big (channel conditions is good);Any one user at certain subcarrier to its channel gain
The while of the least, also tend to also other subcarriers bigger to its channel channel gain.
Energy function is verified
For the channel gain data in table 1, when resetting b=1,2,3,4 in formula (6), the modulation ratio of each subcarrier
Special number is respectively C (b)=0,2,4,6 four kinds of values, four kinds of modulation systems of i.e. corresponding subcarrier: do not modulate (0 bit), 4-
QAM (2 bit), 1QAM (4 bit), 64-QAM (6 bit) modulate.Then each user is distributed the data of 16bit, uses god
The energy function formula (6) proposed the present invention through network is verified, just obtains the multi-user orthogonal frequency of 32 subcarriers, 8 users
Division multiplexing system subcarrier, bit, power co-allocation result as shown in table 2, the nonzero digit (2,4,6) in table 2 represents
The bit number modulated on the subcarrier distributed, numeral 0 represents does not distributes subcarrier.
Parameter for neutral net is arranged, by each ginseng in formula (8), formula (9), formula (10), formula (11) in this example
Number setting is respectively as follows: Ae=0.02, Be=6, Ce=45, De=4;Other parameters are set to: k=0.95, α=0.005, β1=β2
=0.06, μ0=1, zi(0)=0.8, I0=0.5, Am=0.5;Maximum iteration time is set to 1000.
Table 2 is visible, every string nonzero digit and be equal to 16, the most each user is assigned with the data of 16 bits, full
The foot Section 2 of energy function formula (6);All subcarriers are at most assigned to only a user or the free time does not distributes (such as No. 16
Subcarrier is idle), meet the constraints 2 of formula (3), i.e. meet third and fourth constraint bar of energy function formula (6)
Part.
The OFDM Dynamic Resource Allocation for Multimedia table of table 2 32 subcarrier 8 user
In table 2, the 16th sub carriers is not yet assigned to any user, it appears that seem the Section 3 of the formula that violates (6), its
Real, neuron value now is vk,16,1=1 (n=16, b=1,1≤k≤8), as it was previously stated, in this simulation example,
In setting formula (6) when b=1,2,3,4, the modulation bit number of each subcarrier is respectively C (b)=0,2,4,6 four kinds of values, i.e.
Have been defined for C (1)=0, i.e. with neuron vk,n,1=1 represents the unallocated bit number of subcarrier, so not violating formula
(6) Section 3.
For the Section 1 of energy function formula (6), can verify in terms of two.
One is visible by the numerical value in synopsis 2 and table 1, that every numerical value in Table 1 is bigger grid, same at table 2
Position (identical row and column), then be typically all assigned with the bit number of non-zero;And those subcarriers that numerical value is less in Table 1
Channel gain, the most in table 2, does not the most distribute subcarrier (bit number is 0).16th sub carriers why can the free time not
Distribution be this subcarrier for all users, channel gain is not all very big, although the channel gain of No. 3 users is reached by it
0.9737, but this channel yield value is for No. 3 user, does not preponderate, say, that No. 3 user is come
Saying, sub-carrier channels gain also having more than 0.9737 is a lot, as a result the 16th sub carriers be also the failure to distribute to No. 3 with
Family.Certainly, in Practical Project, all users are declined the biggest situation by a subcarrier, or the most rare.
The method of the Section 1 of another checking energy function formula (6) is to observe energy function curve, and Fig. 4 gives nerve
The energy function curve figure of network.
In Fig. 4, more than energy function to 300 iteration step lengths, the most no longer have dropped, but the most do not reach zero, because of
Will not be zero for total emission power.Compared with the energy function curve of foregoing character of the present invention, the energy function curve of Fig. 4 rises and falls
More violent, this reason is neuron vk,n,bThe weights more than 1 are had been assigned in energy function formula (6).In formula (6)
In Section 1, neuron vk,n,bIt is multiplied with channel gain function f (g), in Section 2, neuron vk,n,bAgain with the letter of footnote b
Number C (b) is multiplied.So, although some neuron vk,n,bValue be consecutive variations between 0 to 1, but carry the most therefrom
The excursion of the energy function come has exceeded well between 0 to 1, thus when result in neutral net operation, the change of energy function
Change scope occurs in that the saltus step that comparison is violent.The Average Iteration step-length of this example neutral net is about 300.
The parameter of neutral net is arranged
For the setting of neural network parameter, puppet adjacent with optimal solution in neutral net state space to be suppressed
State, pseudo-state refers to trivial solution.
First assume that neutral net has searched out an optimal solution, then changes the output of some neuron, is allowed to from 1
To 0.As a result, last value of energy function formula (6) will not change, and its value remains 0.This is because formula (6) is last
One is to be multiplied two-by-two by (k × b) plane neuron (except each neuron self), then by all of product phase
Add.So the essence of last item constraint, it is to require in (k × b) plane neuron, at most can only have the value of a neuron
For non-zero, remaining neuron is all necessarily equal to 0 (i.e. one subcarrier is assigned to only a user), if so at (k × b)
In plane neuron, even if the value of all neurons all becomes zero, but last value remains 0.
Except last value is constant, remaining every value of formula (6) all can become big or diminish, and is analyzed as follows.
The value of formula (6) Section 1 can reduce, and when the output of some neuron of change, is allowed to from 1 to 0, is equivalent to few point
Joined a subcarrier give certain user, result general power just have dropped, and the value of decline is expressed as follows:
Assume neuron vk,n,bOutput change to 0 from 1, then be equivalent to be assigned with less a sub-carrier wave n to user, and should
On subcarrier, the bit number of modulation is C (b), owing to the transmitting power needed for this subcarrier-modulated is f (C (b), BERk), then examine
Consider punishment parameter A of energy function formula (6) Section 1e, then the value that Section 1 reduces is exactly the value of formula (19).
The value of formula (6) Section 2 the most likely keeps constant, it is also possible to increase.If neuron vk,n,bFootnote b=
1, then bit number C (the b)=C (1)=0 distributed on subcarrier n, i.e. subcarrier n the most do not distribute bit number, in other words
Subcarrier n is idle, is not assigned to any user.Now, as neuron vk,n,bOutput change to 0 from 1, each user
The bit number distributed does not has this change in fact, thus the value of Section 2 is also maintained for constant.But, if neuron vk,n,b's
Footnote b is not equal to 1, then bit number C (b) distributed on subcarrier n is the most just no longer equal to 0, if now allowing neuron vk,n,b
Output change to 0 from 1, then the bit number that user k is distributed just decreases C (b) bit, thus the value of formula (6) Section 2 will
Increase:
The value of formula (6) Section 3 can increase, due to as neuron vk,n,bOutput change to 0 from 1, cause (k × b) put down
In nervus facialis unit, the output valve of neuron decreases one, so the value of formula (6) Section 3 will increase:
It is pointed out that the change of above-mentioned neuron, only need to consider that neuron exports from changing to, and without consider from
The change of 0 to 1 because from 0 change to 1 time, the every value of energy function all can increase, and only the output of neuron is from 1 change
During to 0, the energy function existing increase of every just meeting, it is reduced again.
In order to suppress pseudo-state, improve the convergence efficiency of neutral net, be necessary for guaranteeing the energy function value of pseudo-state, begin
Be greater than eventually the energy function value of optimal solution, i.e. formula (6) second and third increase value, itself and be greater than Section 1 reduce value,
It is expressed as follows:
It should be noted that during due to b=1, bit number C (b) that subcarrier n is distributed=C (1)=0, so god
Through unit vk,n,bOutput from 1 change to 0 time, it should only considering the situation that b is not equal to 1, energy function Section 2 can increase, this is just
The possible value of requirement formula (22) left side minimum is more than the possible value of the right maximum, when being now considered as b=2, on subcarrier n
The situation of bit number C (b) distributed=C (2)=2;The possible value that the right is maximum, mainly considers sub-carrier channels gain
Excursion, as it can be seen from table 1 sub-carrier channels gain hk,nValue, in the case of 99%, be all greater than 0.1, because of
This is in formula (22), and the embodiment of the present invention chooses hk,nThe situation that minima is 0.1, BERkIt is set to 10 for unified-4, formula (4)
Noise power spectral density be set to N0=1, then, formula (22) can be reduced to:
Formula (23) is exactly that the present invention is to three punishment parameters Ae、Be、CeInstallation warrants, it is also possible to be rewritten into 4Be+Ce>
240Ae.For last punishment parameter D in formula (6)eSetting, the embodiment of the present invention takes first to arrange Ae=0, then set
Put 4Be=Ce=De=1 (4Be+Ce>240Ae), neutral net of reruning, and observe the every of energy function, if it find that energy
The value of a certain item of function, the value more every than other is much bigger, and the iteration step length of neutral net increases, and it is invalid to be even absorbed in
Solve (step-length exceedes maximum set value 1000), the most correspondingly increase the punishment parameter that it is corresponding, to reduce this value, accelerate
The convergence of neutral net;Otherwise then keep constant, or reduce its value.
In punishment parameter Be、Ce、DeAfter all setting, set A further according to formula (23)eValue.
At the end of every single-step iteration, from (k × b) plane neuron, choose the nerve that wherein output valve is maximum
Unit, is set to 1 by the output of this neuron, and the output of remaining neuron is both configured to 0, is then updated to energy function formula
(6) in, if now second and third, the value of four, be equal to 0, even and if neutral net iteration continues again, the value of Section 1
Reduce the most again, then neuron vk,n,bWith current centrifugal pump (0 or 1) output, neutral net terminates iteration;Otherwise recover god
Through unit vk,n,bSuccessive value, neutral net proceeds next step iteration.
System emulation and performance evaluation
System emulation is based on the single OFDM subcarrier rayleigh fading channel (Rayleigh shown in Fig. 2
Fading Channels) model, embodiment of the present invention hypothesis system bandwidth is 5MHz, and the sub-carrier number of OFDM is
128, the bit error rate requirement of each user is unified is assumed to 10-4, the bit number that each subcarrier can distribute respectively gathers D=
{ element in 0,2,4,6}, i.e. modulation system are respectively, and do not modulate (subcarrier is idle), 4-QAM, 1QAM, 64-QAM.Each
Total bit number of OFDM symbol is 512 bits, and this 512 bit is randomly assigned to each user, and number of users sets respectively
Be set to 2,4,8,16,32 5 kind of situation.
Performance evaluation
In order to assess the performance of the optimized algorithm that the present invention proposes, and conveniently comparing with achievement in advance, the present invention adopts
Assess, if the transmitting of each OFDM symbol is total with average bit SNR (average bit SNR, abSNR)
Power is PT, the data of each OFDM symbol are 512 bits, and then average bit SNR is defined as:
In formula (24), N0For the noise power spectral density in the range of simulation bandwidth, in this example, in order to simplify calculating, and
Convenient and existing document result compares, and noise power spectral density is set to N0=1.Be 2 the most respectively to number of users,
4, five kinds of situations of 8,16,32, total bit number is 512bit, emulates, and obtains the curve of abSNR as shown in Figure 5.
In Fig. 5, Wong algorithm, Kim algorithm and Shin algorithm are that the comparison that three kinds of foreign scholars propose is classical
AbSNR performance curve, for comparing the performance of the inventive method and system, approach is the abSNR performance that the present invention proposes
Curve.
Neutral net approach curve in Fig. 5 is when number of users increases successively from 2,4,8,16,32, and abSNR is gradually
Decline, this is because when number of users is less, each OFDM subcarrier is in deep fade to all users
Probability bigger, such as only 2 users time, 2 users are deep by any one OFDM subcarrier
The probability of decline, significantly larger than number of users is probability when 4,8,16,32.When certain subcarrier to all users all in deeply declining
When falling, its result necessarily causes more powerful (the meeting under same bit distribution condition) distributed on this subcarrier, thus
Cause abSNR bigger, so when only 2 users, the performance advantage of neural network algorithm does not highlight.But at number of users
When increasing to 4,8,16, the probability that each OFDM subcarrier is in deep fade to all users is gradually lowered, this
Time neural network algorithm advantage start to embody.
When number of users increases to 32 from 16, owing to neutral net neuron number purpose increases, the energy letter of neutral net
Number space becomes more complicated, causes the performance of neutral net to be declined slightly, thus causes abSNR the most slightly to increase.
The efficiency of neutral net
In order to assess the Searching efficiency of neutral net, the present invention is further to five kinds of feelings that number of users is 2,4,8,16,32
Under condition, when subcarrier is 128, each user all distributes the data of 16bit, neutral net in the case of every kind is run 100 times,
As shown in table 3 to the Searching efficiency of neutral net.
The ratio of table 3 neutral net optimal solution
From table 3 it can be seen that the Searching efficiency of neutral net is gradually reduced along with the increase of number of users, this is because neural
The neuron number of network increases, and energy function space is more complicated, and neutral net is absorbed in the probability of local minimum and adds.
The embodiment of the present invention, to the Dynamic Resource Allocation for Multimedia problem of multi-user orthogonal frequency division multiplexing system in community, has carried out wide
General in-depth study.The optimized algorithm that multi-user orthogonal frequency division multiplexing system subcarrier, bit, power joint are distributed, from
From the point of view of the document published both at home and abroad at present, all there is algorithm the most complicated, the problem that amount of calculation is the biggest, and algorithm is real-time
Property is not high enough, the most all can only obtain suboptimal solution.
The embodiment of the present invention uses neutral net, it is proposed that a kind of new multi-user orthogonal frequency division multiplexing system subcarrier,
The optimized algorithm of bit, power joint distribution, this algorithm have devised a kind of new energy function, and this energy function considers often
The constraints that the data-rate requirements of individual user and the bit error rate and subcarrier are distributed, is meeting each user rate demand
And in the case of bit error rate requirement, make Base Transmitter general power minimum.
Then by the multi-user orthogonal frequency division multiplexing system example of 8 users of one 32 subcarriers, this chapter is carried
The new energy function gone out is verified, result indicates the effective of the new energy function that the embodiment of the present invention proposed
Property.
Finally by the multi-user orthogonal frequency division multiplexing system example of 128 subcarriers, number of users be 2,4,8,16,
In the case of five kinds of 32, altogether distribute 512bit data to whole system, carry out system emulation.Result verification this chapter is proposed
The effectiveness of optimized algorithm, relative to domestic and international on going result, the algorithm that the embodiment of the present invention is proposed has meter faster
Calculating speed, and the optimizing ability of neutral net is higher, system obtains lower average every bit signal to noise ratio (abSNR).
The method described in conjunction with the embodiment that disclosed in this invention or the step of algorithm can directly use hardware, processor
The software module performed, or the combination of the two implements.Software module can be placed in random access memory, internal memory, read-only storage
Institute in device, electrically programmable ROM, electrically erasable ROM, depositor, hard disk, moveable magnetic disc, CD-ROM or technical field
In the storage medium of known arbitrarily other forms.
It is understood that for the person of ordinary skill of the art, can conceive according to the technology of the present invention and do
Go out other various corresponding changes and deformation, and all these change all should belong to the protection model of the claims in the present invention with deformation
Enclose.
Claims (10)
1. the method for a multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net, it is characterised in that
Comprising:
Setting up three-dimensional nerve variable matrix cubical neutral net neural network model, neural network model represents a subcarrier
With the mapping relations of user, and mapping relations of bit number that each subcarrier is modulated;
The energy function of neural network model is as follows:
Wherein, N is OFDM OFDM total number of sub-carriers;K is total number of users in community;RkIt it is the number of user k
According to speed;BERkBit error rate for user k;hk,nIt it is the subcarrier n fading channel amplitude for user k;vk,n,bFor three-dimensional
Row k in neuron matrix cube, the n-th row, the b three-dimensional nerve unit;B represents the b neuron two-dimensional matrix
Numbering;The bit number distributed on corresponding subcarrier n is C (b), and C (b) is the function of b, when the value of b is natural number, then
The value of C (b) is set D={0,1 ..., CTIn } one, now corresponding subcarrier modulation modes is 2C(b)Constellation is just
Hand over amplitude QAM modulation, CTFor the modulation bit number that subcarrier is maximum allowable;
Energy function according to neural network model carries out multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia so that full
In the case of each user rate demand of foot and bit error rate requirement, make Base Transmitter general power minimum.
2. the side of multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net as claimed in claim 1
Method, it is characterised in that
In the energy function of neural network model, Section 1 represents the total emission power of base station, for each neuron vk,n,bInstitute
Likely combine, by the continuous iteration of neutral net, it is thus achieved that minima, the situation that the most each user rate and the bit error rate give
Under, make cell transmission general power minimum.
3. the side of multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net as claimed in claim 1
Method, it is characterised in that
In the energy function of neural network model, Section 2 represents that the constraints of the rate requirement of user, the most all distributing to are used
The bit number modulated on the subcarrier of family k, its summation is just equal to the rate requirement of kth user;Value etc. when Section 2
In zero time, then meet the rate requirement of each user.
4. the side of multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net as claimed in claim 1
Method, it is characterised in that
In the energy function of neural network model, Section 3, Section 4 represent the constraints that the distribution of subcarrier requires;
For the three-dimensional nerve unit v that footnote is nk,n,b, the combination of its all footnote k and b, referred to as k × b plane neuron;God
Section 3 in the energy function of network model so that footnote is in the two-dimensional matrix of k × b plane neuron of n, for all
The combination of footnote k and b, makes the summation of value of all plane neurons of k × b plane equal to 1;
Section 4 in the energy function of neural network model, makes in k × b plane neuron, most only one of which plane neurons
Value be non-zero, the value all 0 of remaining plane neuron.
5. the multi-user orthogonal frequency division multiplexing system dynamic resource based on neutral net as described in any one of Claims 1-4
The method of distribution, it is characterised in that
Also include: neural network model is sought the partial derivative of three-dimensional nerve unit, it is thus achieved that the state equation of neutral net;
By Euler's method, by the state equation of neutral net to time discretization, and add decay self feed back link power and
Random noise, obtains neutral net discrete time model.
6. the system of a multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net, it is characterised in that
Comprising:
Establishment of Neural Model unit, is used for setting up three-dimensional nerve variable matrix cubical neutral net neural network model,
Neural network model represents a subcarrier and the mapping relations of user, and the reflecting of the bit number modulated of each subcarrier
Penetrate relation;
The energy function of neural network model is as follows:
Wherein, N is OFDM OFDM total number of sub-carriers;K is total number of users in community;RkIt it is the number of user k
According to speed;BERkBit error rate for user k;hk,nIt it is the subcarrier n fading channel amplitude for user k;vk,n,bFor three-dimensional
Row k in neuron matrix cube, the n-th row, the b three-dimensional nerve unit;B represents the b neuron two-dimensional matrix
Numbering;The bit number distributed on corresponding subcarrier n is C (b), and C (b) is the function of b, when the value of b is natural number, then
The value of C (b) is set D={0,1 ..., CTIn } one, now corresponding subcarrier modulation modes is 2C(b)Constellation is just
Hand over amplitude QAM modulation, CTFor the modulation bit number that subcarrier is maximum allowable;
Dynamic Resource Allocation for Multimedia unit, for carrying out multi-user orthogonal frequency division multiplexing system according to the energy function of neural network model
Dynamic Resource Allocation for Multimedia so that in the case of meeting each user rate demand and bit error rate requirement, make Base Transmitter general power
Minimum.
7. multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net as claimed in claim 6 is
System, it is characterised in that
In the energy function of neural network model, Section 1 represents the total emission power of base station, for each neuron vk,n,bInstitute
Likely combine, by the continuous iteration of neutral net, it is thus achieved that minima, the situation that the most each user rate and the bit error rate give
Under, make cell transmission general power minimum.
8. multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net as claimed in claim 6 is
System, it is characterised in that
In the energy function of neural network model, Section 2 represents that the constraints of the rate requirement of user, the most all distributing to are used
The bit number modulated on the subcarrier of family k, its summation is just equal to the rate requirement of kth user;Value etc. when Section 2
In zero time, then meet the rate requirement of each user.
9. multi-user orthogonal frequency division multiplexing system Dynamic Resource Allocation for Multimedia based on neutral net as claimed in claim 6 is
System, it is characterised in that
In the energy function of neural network model, Section 3, Section 4 represent the constraints that the distribution of subcarrier requires;
For the three-dimensional nerve unit v that footnote is nk,n,b, the combination of its all footnote k and b, referred to as k × b plane neuron;God
Section 3 in the energy function of network model so that footnote is in the two-dimensional matrix of k × b plane neuron of n, for all
The combination of footnote k and b, makes the summation of value of all plane neurons of k × b plane equal to 1;
Section 4 in the energy function of neural network model, makes in k × b plane neuron, most only one of which plane neurons
Value be non-zero, the value all 0 of remaining plane neuron.
10. the multi-user orthogonal frequency division multiplexing system dynamic resource based on neutral net as described in any one of claim 6 to 9
The system of distribution, it is characterised in that
Also include: the state equation acquiring unit of neutral net, with neural network model is asked three-dimensional nerve unit partial derivative, obtain
Obtain the state equation of neutral net;
Neutral net discrete time model unit, for by Euler's method, by the state equation of neutral net to time discrete
Change, and add self feed back link power and the random noise of decay, obtain neutral net discrete time model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510872169.0A CN105813208A (en) | 2015-12-01 | 2015-12-01 | Method and system for dynamic resource distribution of multi-user orthogonal frequency division multiplexing system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510872169.0A CN105813208A (en) | 2015-12-01 | 2015-12-01 | Method and system for dynamic resource distribution of multi-user orthogonal frequency division multiplexing system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105813208A true CN105813208A (en) | 2016-07-27 |
Family
ID=56465664
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510872169.0A Pending CN105813208A (en) | 2015-12-01 | 2015-12-01 | Method and system for dynamic resource distribution of multi-user orthogonal frequency division multiplexing system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105813208A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019144701A1 (en) * | 2018-01-25 | 2019-08-01 | 腾讯科技(深圳)有限公司 | Neural network computing method and apparatus, and related device |
CN111328133A (en) * | 2018-12-17 | 2020-06-23 | 上海大学 | V2X resource allocation method based on deep neural network |
CN113411106A (en) * | 2021-05-31 | 2021-09-17 | 海南大学 | Power distribution method based on deep learning in safe space modulation system |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101854725A (en) * | 2010-06-30 | 2010-10-06 | 西安交通大学 | Resource distribution and power distribution method for cellular multi-cell orthogonal frequency division multiple access (OFDMA) system |
CN102891822A (en) * | 2012-09-21 | 2013-01-23 | 北京邮电大学 | Resource distribution method of multi-user OFDM (Orthogonal Frequency Division Multiplexing) system |
-
2015
- 2015-12-01 CN CN201510872169.0A patent/CN105813208A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101854725A (en) * | 2010-06-30 | 2010-10-06 | 西安交通大学 | Resource distribution and power distribution method for cellular multi-cell orthogonal frequency division multiple access (OFDMA) system |
CN102891822A (en) * | 2012-09-21 | 2013-01-23 | 北京邮电大学 | Resource distribution method of multi-user OFDM (Orthogonal Frequency Division Multiplexing) system |
Non-Patent Citations (1)
Title |
---|
赵承志: "未来移动通信动态资源分配算法的研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019144701A1 (en) * | 2018-01-25 | 2019-08-01 | 腾讯科技(深圳)有限公司 | Neural network computing method and apparatus, and related device |
CN110083448A (en) * | 2018-01-25 | 2019-08-02 | 腾讯科技(深圳)有限公司 | A kind of computing resource method of adjustment, device and relevant device |
US11507812B2 (en) | 2018-01-25 | 2022-11-22 | Tencent Technology (Shenzhen) Company Limited | Neural network operational method and apparatus, and related device |
CN110083448B (en) * | 2018-01-25 | 2023-08-18 | 腾讯科技(深圳)有限公司 | Computing resource adjusting method and device and related equipment |
CN111328133A (en) * | 2018-12-17 | 2020-06-23 | 上海大学 | V2X resource allocation method based on deep neural network |
CN113411106A (en) * | 2021-05-31 | 2021-09-17 | 海南大学 | Power distribution method based on deep learning in safe space modulation system |
CN113411106B (en) * | 2021-05-31 | 2023-03-24 | 海南大学 | Power distribution method based on deep learning in safe space modulation system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Su et al. | Obtaining full-diversity space-frequency codes from space-time codes via mapping | |
CN102868459B (en) | For mobile terminal emulator and the emulation mode of radio telecommunication network | |
CN106162659A (en) | A kind of LEO system DCS signal reconfiguring method of energy efficiency priority time delay tolerance | |
CN102164108A (en) | Multi-carrier communication with time division multiplexing and carrier-selective loading | |
CN104243370B (en) | A kind of time-domain channel estimating method applied in multiaerial system | |
CN103051583B (en) | A kind of OFDMA resource allocation methods based on rate adaptation | |
CN102413487B (en) | Performance evaluation system aiming at Relay technology in LTE-A (Long Term Evolution-Advanced) system | |
CN105940652A (en) | Method and apparatus for estimating communication channel in mobile communication system | |
CN101557643B (en) | Method and system of radio resource allocation | |
CN109861731A (en) | A kind of mixing precoder and its design method | |
CN105813208A (en) | Method and system for dynamic resource distribution of multi-user orthogonal frequency division multiplexing system | |
CN102781101B (en) | Resource distribution method in orthogonal frequency division multiple access (OFDMA) relay system | |
CN107734697B (en) | Low-complexity multi-user grouping method based on NOMA system | |
CN104702326A (en) | MSE-based (mean square error-based) virtual MIMO (multiple input multiple output) user pairing and resource allocating method | |
CN106452534A (en) | Pilot optimization method for large-scale MIMO channel estimation based on structural compressed sensing | |
CN102891822B (en) | Resource distribution method of multi-user OFDM (Orthogonal Frequency Division Multiplexing) system | |
CN101511125A (en) | Discrete velocity layer-striding power distribution method suitable for distributed antenna system | |
CN110191476A (en) | A kind of non-orthogonal multiple cut-in method based on reconfigurable antenna array | |
CN104135769A (en) | Method of OFDMA (Orthogonal Frequency Division Multiple Access) ergodic capacity maximized resource allocation under incomplete channel state information | |
Carvajal et al. | Performance analysis of MC-CDMA systems in Rayleigh fading channel with inter-cell and co-cell interference | |
Xuehua et al. | Study and implementation of MIMO-OFDM system based on Matlab | |
Kim et al. | Selection of CDMA and OFDM using DNN in underwater acoustic channels | |
HashibSiddique et al. | Performance study of IEEE 802.16 d under Stanford University Interim (SUI) channel | |
CN103354658B (en) | Multiple cell multi-input multi-output system considers the optimized for energy efficiency power distribution algorithm that co-channel interference suppresses | |
CN103796294B (en) | Power distribution method of DAS |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160727 |