CN112929977A - Deep learning amplification forwarding cooperative network energy efficiency resource allocation method - Google Patents

Deep learning amplification forwarding cooperative network energy efficiency resource allocation method Download PDF

Info

Publication number
CN112929977A
CN112929977A CN202110184433.7A CN202110184433A CN112929977A CN 112929977 A CN112929977 A CN 112929977A CN 202110184433 A CN202110184433 A CN 202110184433A CN 112929977 A CN112929977 A CN 112929977A
Authority
CN
China
Prior art keywords
node
power
relay
source node
vector
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.)
Granted
Application number
CN202110184433.7A
Other languages
Chinese (zh)
Other versions
CN112929977B (en
Inventor
郭艳艳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanxi University
Original Assignee
Shanxi University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanxi University filed Critical Shanxi University
Priority to CN202110184433.7A priority Critical patent/CN112929977B/en
Publication of CN112929977A publication Critical patent/CN112929977A/en
Application granted granted Critical
Publication of CN112929977B publication Critical patent/CN112929977B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • H04W72/0473—Wireless resource allocation based on the type of the allocated resource the resource being transmission power
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W40/00—Communication routing or communication path finding
    • H04W40/02—Communication route or path selection, e.g. power-based or shortest path routing
    • H04W40/22—Communication route or path selection, e.g. power-based or shortest path routing using selective relaying for reaching a BTS [Base Transceiver Station] or an access point
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention discloses an energy efficiency resource allocation method for an amplification forwarding cooperative network for deep learning. First, a Combined Deep Neural Network (CDNN) model is constructed consisting of two independent fully-connected (FC) subnetworks, one of which is used to determine the power allocation of the source node and the other is used to determine the power allocation of the relay node. In addition, an activation function based on an arc tangent function is designed for a CDNN model output layer to approximate the distribution of power factors, and a rectification linear unit (ReLU) function is adopted to further filter the output negative relay node power weight, so that the selection of the relay nodes and the power distribution of the relay nodes are realized at the same time. The designed activation function proves to be capable of allocating a more appropriate power value to the wireless node according to the channel gain than the Sigmoid-based activation function. Simulation results also prove that compared with the traditional method of fixing the source node power, the CDNN model provided by the invention has improved performance in two aspects of energy efficiency and interruption probability.

Description

Deep learning amplification forwarding cooperative network energy efficiency resource allocation method
Technical Field
The invention belongs to the technical field of wireless cooperative communication, and particularly relates to a deep learning energy efficiency resource allocation method for an amplification forwarding cooperative network.
Background
With the rapid development of the 5G communication technology, the high energy consumption brought by the complex network structure becomes a road barricade for 5G scale commercial use, and how to improve the technology and greatly reduce the high energy consumption of the 5G network is not slow. In the cooperative communication technology, under the multi-user environment, each single-antenna user shares antennas with each other, so that a virtual multi-antenna system is formed, transmission diversity is realized, and the transmission performance of the system is improved. The cooperative communication technology has very wide application prospect in the field of 5G communication. Wherein the amplify-and-forward (AF) collaboration is cheaper to deploy than decode-and-forward (DF) collaboration network and is very attractive for network operators. In the research of the AF cooperation technology, relay selection and joint power optimization have very important significance for improving network energy efficiency. On one hand, however, joint power optimization of a source node and a plurality of relay nodes in an AF cooperative network includes continuous variables and combined variables, which is a non-convex problem, and the current research method can be solved only on the premise of knowing the power of the source node; on the other hand, when a group of relay nodes with an unfixed number is selected from available relay nodes for cooperation and attempts are made to optimize beamforming vectors under respective power constraints, only an Exhaustive Search (ES) method can be used for the solution, which results in extremely high complexity.
Deep learning, which approximates complex problems with sufficient neurons and hidden layers, is an ideal choice for performing a large number of non-convex and non-linear problems. Therefore, the deep learning is applied to the multi-relay AF cooperation system, and the method has important significance for optimizing the resource allocation problem.
Disclosure of Invention
Aiming at the problems, the invention provides a deep learning energy efficiency resource allocation method for an amplification forwarding cooperative network.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides a Combined Deep Neural Network (CDNN): the relay node power distribution system comprises two independent Fully Connected (FC) sub-networks, namely SF1 and SF2, wherein the two sub-networks are formed by connecting limited sub-modules in series, the SF1 sub-network is used for determining the power distribution of relay nodes, other sub-modules of the sub-network except the last sub-module are formed by a fully connected layer followed by a hyperbolic tangent function, the last sub-module of the sub-network is formed by a fully connected layer followed by a self-defined activation function, the number of hidden nodes of the last sub-module is the same as the number of potential relay nodes in an amplifying and forwarding cooperative network, and each output corresponds to the power distribution of one relay node; the SF2 sub-network is used for determining the power distribution of the source node, the sub-network except the last sub-module consists of a full connection layer followed by a rectification linear unit (ReLU), the last sub-module consists of a full connection layer followed by a self-defined activation function, and the output is the power of the source node.
The invention also provides a deep learning method for allocating the energy efficiency resources of the amplify-and-forward cooperative network, which comprises the following steps:
step 1, establishing a source node and relay node power distribution joint optimization model for amplifying and forwarding cooperative network minimum energy consumption;
step 2, constructing a combined deep neural network model;
step 3, training the constructed combined deep neural network model;
and 4, testing by using the trained and learned combined deep neural network model.
Further, the establishing of the source node and relay node power distribution joint optimization model for amplifying and forwarding the minimum energy consumption of the cooperative network in the step 1 includes the following steps:
(1) calculating the total rate of the amplification forwarding cooperative network system;
(2) and establishing a power distribution joint optimization model of the source node and the relay node for minimizing energy consumption of the amplify-and-forward cooperative network.
Further, the amplify-and-forward cooperative network system in step (1) includes a source node and a destination node, N potential relay nodes; the calculation of the total rate of the amplifying and forwarding cooperative network system comprises the following processes:
assuming that the channel states from a source node to a destination node, from the source node to all potential relay nodes and from all potential relay nodes to the destination node are known;
further, assuming that a transmission time slot is divided into two parts, in a first part of time period, a source node broadcasts information to a destination node and all potential relay nodes, which is called a broadcasting stage; in the second part of time period, the relay node forwards the information received from the source node, which is called a cooperation stage;
in the broadcast phase, a source node transmits a signal x, and signals received by a destination node and any relay node are respectively:
Figure BDA0002942441090000031
Figure BDA0002942441090000032
in the formula, psFor the transmission power of the source node, assume that the amplitude of the signal x transmitted by the source node is 1, hs,dAnd hs,iChannel gains, w, from source node to destination node, and from source node to relay node i, respectivelys,dAnd ws,iIn the broadcasting stage, the receiving noise of a target node and a relay node i; assuming that the noise received by all the relay nodes and the destination node obeys complex Gaussian distribution, and the mean value is zero;
in the cooperation stage, the receiving, by the destination node, of the signal amplified and forwarded by the relay node is:
Figure BDA0002942441090000041
in the formula (I), the compound is shown in the specification,
Figure BDA0002942441090000042
the relay node receives the variance of the noise; h isi,dIs the channel gain from the relay node i to the destination node; w is ar,dIs the reception noise of the destination node in the cooperation phase,
Figure BDA0002942441090000043
is the power normalization factor of the relay node i; vector quantity
Figure BDA0002942441090000044
Is the forward power vector of all potential relay nodes [. C]TRepresenting transpose of matrix or vector, alphaiRepresenting the forward power, vector, of the relay node i
Figure BDA0002942441090000045
(Vector)
Figure BDA0002942441090000046
Combining formulae (1) and (3) can result in:
Figure BDA0002942441090000047
according to equation (4), the signal-to-noise ratio of the destination node is obtained:
Figure BDA0002942441090000048
in the formula (I), the compound is shown in the specification,
Figure BDA0002942441090000049
variance, H, representing the received noise of the destination nodeHRepresenting the vector H by transposition, fHTaking a transposition of the expression vector f; wHThe expression vector W is transposed;
the total rate obtained by the destination node of the amplifying and forwarding cooperative network system is as follows:
Rd＝Blog2(1+γd) (6)
where B is the bandwidth of the channel.
The source node and relay node power distribution joint optimization model and constraint conditions for amplifying and forwarding the minimum energy consumption of the cooperative network in the step (2) are as follows:
Figure BDA0002942441090000051
satisfies the following conditions: rd≥Rth (7b)
0＜ps＜Ps max (7c)
Figure BDA0002942441090000052
Figure BDA0002942441090000053
In the formula (f)Hf represents the sum of the powers of all relay nodes, RthIndicating the system rate requirement, Ps maxIs the power threshold of the source node,
Figure BDA0002942441090000054
a threshold value representing the forwarding power of each relay node,
Figure BDA0002942441090000055
is the threshold value of the sum of the forwarding power of all the relay nodes; (7b) representing a rate constraint required by the amplifying and forwarding cooperative network system; (7c) representing a power constraint of the source node; (7d) represents a power limit for each relay node; (7e) representing a total power constraint for all relay nodes.
Further, the specific steps of the combined deep neural network model constructed in step 2 of the deep learning method for allocating the energy efficiency resources of the amplify-and-forward cooperative network for deciding the power allocation of the source node and the relay node are as follows:
(1) one-dimensional vector of normalized channel gain
Figure BDA00029424410900000511
Channel gains from a source node to a destination node, from the source node to all potential relay nodes and from all potential relay nodes to the destination node are respectively input into SF1 sub-networks and SF2 sub-networks;
Figure BDA00029424410900000512
element (1) of
Figure BDA00029424410900000513
Is derived from the channel gain | hi,j|2Converted to dB and then unified into a vector with mean value of zero and variance of 1
Figure BDA0002942441090000058
The calculation formula is as follows:
Figure BDA0002942441090000059
(2) input to SF1 one-dimensional vector
Figure BDA00029424410900000510
The vectors output by the full-connection layer of the last submodule of the SF1 are fed to a self-defined activation function through the processing of the submodules connected in series in the SF1, the transmission power of each relay node is output, the activation function sets the negative output power value to be 0, and when the output power value is 0, the corresponding potential relay node does not participate in cooperative transmission, so that the relay node selection and the power distribution are completed simultaneously;
(3) input to SF2 one-dimensional vector
Figure BDA0002942441090000061
The output of the full connection layer of the last submodule of the SF2 is fed to a self-defined activation function through the processing of the submodules connected in series in the SF2, the number of the output is 1, and the power distribution of the source node is completed.
Further, the specific steps of relay node selection and power allocation in step (2) are as follows:
vector x of outputs of full-link layer of last submodule of SF1SF1Feeds into an activation function
Figure BDA0002942441090000062
In the formula, alphaiSetting the output negative value as 0 by equation (9) for the power allocated to the relay node i, and ensuring that the power of the relay node i satisfies the requirement
Figure BDA0002942441090000063
The restriction condition (7d) is satisfied; [ x ] ofSF1]iRepresenting a vector xSF1If α isi0, indicates that the relay node i does not participate in cooperative transmission.
The specific steps of the power distribution of the source node in the step (3) are as follows:
output x of full link layer in last submodule of SF2SF2Is fed to a custom activation function, the output is represented as
Figure BDA0002942441090000064
In the formula (I), the compound is shown in the specification,
Figure BDA0002942441090000065
the number of hidden nodes for the last submodule of SF 2; p is a radical ofs∈(0,Ps max) Is the transmission power of the source node, and the limitation condition (7c) is satisfied; [ x ] ofSF2]iRepresenting a vector xSF2The ith value of (a).
Further, in the method for allocating energy efficiency resources of an amplification forwarding cooperative network for deep learning, the process of training the constructed combined deep neural network model in step 3 is as follows: the multiple samples are input into the combined deep neural network model as a set of training data, then training is performed by using the adaptive moment estimation, and when the combined deep neural network model cannot continuously exceed the set baseline, the training algorithm stops. In order to achieve the goal of minimizing the system energy consumption and simultaneously meet the system rate requirement and the total power constraint of the relay nodes, the network parameters of the two sub-networks need to be synchronously updated by the following loss functions in the training process of the combined deep neural network model:
Figure BDA0002942441090000071
wherein E {. denotes averaging a set of training data; lambda is more than or equal to 01≤1，0≤λ 21 or less and 0 or less3Control parameters are less than or equal to 1; [ x ] of]+＝max(x,0)；[Rth-Rd]+Corresponding to the constraint (7b), if satisfied, the value is 0;
Figure BDA0002942441090000072
corresponding to the constraint (7e), if satisfied, the value is 0; (f)Hf+ps) Corresponding to equation (7 a); theta1Set of all weights and offsets for SF1, θ2Set of all weights and offsets for SF 2;
in the process of minimizing the loss function, the direction is continuously opposite to theta1And theta2The update is performed as follows:
Figure BDA0002942441090000073
Figure BDA0002942441090000074
wherein, 0 < beta 11 and 0 beta 21 respectively represents the corresponding learning rate; t represents the number of updates;
Figure BDA0002942441090000075
representing the loss function L vs. theta1Taking a partial derivative;
Figure BDA0002942441090000076
representing the loss function L vs. theta2Taking a partial derivative; and the number of the first and second electrodes,
Figure BDA0002942441090000077
and
Figure BDA0002942441090000078
initial weights and offsets for SF1 and SF2, respectively.
Compared with the prior art, the invention has the following advantages:
1. the invention constructs a Combined Deep Neural Network (CDNN) which comprises two independent fully-connected (FC) sub-networks which are respectively used for determining the power distribution of a source node and a single relay node. Then, network parameters of the two sub-networks are synchronously updated by using a loss function, and the design of the loss function considers the maximum forwarding power and the constraint so as to minimize the overall energy consumption of the system without violating the rate requirement of the system, thereby jointly learning the power distribution strategies of the source node and the relay node.
2. The invention uses an improved arctangent function as an activation function of the output layer of the CDNN model to approximate the distribution of the power factor. In addition, the negative output power value of the power distribution sub-network for determining the relay nodes is further removed by the improved rectifying and linear unit (ReLU), so that the selection of the relay nodes and the power distribution thereof can be completed simultaneously, and the number of the relay nodes to be selected does not need to be fixed.
3. The simulation result proves that the energy efficiency performance of the CDNN model provided by the invention is improved compared with that of the traditional method.
Drawings
FIG. 1 is an AF collaboration system model;
FIG. 2 is a diagram of a combined deep neural network model according to the present invention;
FIG. 3 is a system average energy efficiency graph under different numbers of potential relay nodes;
fig. 4 is a system outage probability map for different numbers of potential relay nodes;
fig. 5 is a graph of the average energy efficiency of the system under various system rate requirements for a potential number of relays of 6.
Detailed Description
The technical solution in the embodiments of the present invention will be specifically and specifically described below with reference to the embodiments of the present invention and the accompanying drawings. It should be noted that variations and modifications can be made by those skilled in the art without departing from the principle of the present invention, and these should also be construed as falling within the scope of the present invention.
Example 1
Establishing a source node and relay node power distribution joint optimization model for amplifying and forwarding cooperative network minimum energy consumption
As shown in fig. 1, the AF system includes a Source node (Source) and a Destination node (Destination), and N potential Relay nodes (Relay) participate in cooperation.
1. And (3) calculating the total rate of the amplification forwarding cooperative network system:
the channel state from the source node to the destination node, from the source node to all potential relay nodes and from all potential relay nodes to the destination node of the AF system is assumed to be known.
Further, assuming that a transmission time slot is divided into two parts, in a first part of time period, a source node broadcasts information to a destination node and all potential relay nodes, which is called a broadcasting stage; in the second part of time period, the relay node forwards the information received from the source node, which is called a cooperation stage;
in the broadcasting stage, a source node transmits a signal x, and a destination node and any relay node receive signals respectively
Figure BDA0002942441090000091
Figure BDA0002942441090000092
In the formula, psFor the transmission power of the source node, assume that the amplitude of the signal x transmitted by the source node is 1, hs,dAnd hs,iChannel gains, w, from source node to destination node, and from source node to relay node i, respectivelys,dAnd ws,iIn the broadcasting stage, the receiving noise of a target node and a relay node i; assuming that the noise received by all the relay nodes and the destination node obeys complex Gaussian distribution, and the mean value is zero;
in the cooperation stage, the receiving, by the destination node, of the signal amplified and forwarded by the relay node is:
Figure BDA0002942441090000093
in the formula (I), the compound is shown in the specification,
Figure BDA0002942441090000094
the relay node receives the variance of the noise; h isi,dIs the channel gain from the relay node i to the destination node; w is ar,dIs the reception noise of the destination node in the cooperation phase,
Figure BDA0002942441090000095
is the power normalization factor of the relay node i; vector quantity
Figure BDA0002942441090000096
Is the forward power vector of all relay nodes [ ·]TRepresenting transpose of matrix or vector, alphaiForward power, vector of relay node i
Figure BDA0002942441090000097
(Vector)
Figure BDA0002942441090000098
By combining the formulae (1) and (3), a compound having a structure represented by the formula
Figure BDA0002942441090000101
According to the formula (4), the signal-to-noise ratio of the destination node is obtained
Figure BDA0002942441090000102
In the formula (I), the compound is shown in the specification,
Figure BDA0002942441090000103
variance of received noise, H, at destination nodeHRepresenting vector H by taking turnsF is put onHTaking a transposition of the expression vector f; wHThe representative vector W takes the transpose.
The total rate obtained by the destination node of the amplifying and forwarding cooperative network system is as follows:
Rd＝Blog2(1+γd) (6)
where B is the bandwidth of the channel.
2. Establishment of source node and relay node power distribution joint optimization model for amplifying and forwarding cooperative network minimum energy consumption
The power distribution joint optimization model and the constraint conditions of the source node and the relay node for the amplification forwarding cooperative network to minimize energy consumption are as follows:
Figure BDA0002942441090000104
satisfies the following conditions: rd≥Rth (7b)
0＜ps＜Ps max (7c)
Figure BDA0002942441090000105
Figure BDA0002942441090000106
In the formula (f)Hf represents the sum of the powers of all relay nodes, RthIndicating the system rate requirement, Ps maxIs the power threshold of the source node,
Figure BDA0002942441090000107
a threshold value representing the forwarding power of each relay node,
Figure BDA0002942441090000108
is the threshold value of the sum of the forwarding power of all the relay nodes; (7b) representing a rate constraint required by the amplifying and forwarding cooperative network system; (7c) representing a power constraint of the source node; (7d)represents a power limit for each relay node; (7e) representing a total power constraint for all relay nodes.
Example 2
Constructing a combined deep neural network model
As shown in fig. 2, the constructed CDNN model includes two independent fully-connected sub-networks, denoted SF1 and SF2, each consisting of M1And M2The number of hidden nodes of the ith sub-block in SF1 and SF2 is represented as and. The SF1 sub-network is used for determining the power distribution of the relay node, except the last sub-module, the other sub-modules of the sub-network are all composed of a full connection layer followed by a hyperbolic tangent function, and the hyperbolic tangent function is expressed as
Figure BDA0002942441090000111
The last sub-module of the sub-network consists of a full connection layer followed by a self-defined activation function, the number of hidden nodes of the last sub-module is the same as the number of potential relay nodes in the amplification forwarding cooperative network, and each output determines the power distribution of one relay node; the SF2 sub-network is used for determining the power distribution of the source node, the sub-network except the last sub-module consists of a full connection layer followed by a rectification linear unit (ReLU), the last sub-module consists of a full connection layer followed by a self-defined activation function, and the output is the power of the source node.
Example 3
The combined deep neural network model decides the power distribution of the source node and the relay node, and comprises the following steps:
step 1, normalizing the channel gain one-dimensional vector
Figure BDA0002942441090000112
The length is 2N +1, and the channel gains from the source node to the destination node, from the source node to all potential relay nodes and from all potential relay nodes to the destination node are respectively input into SF1 and SF2 sub-networks;
Figure BDA0002942441090000119
element (1) of
Figure BDA0002942441090000118
Is derived from the channel gain | hi,j|2Converted to dB and then unified into a vector with mean value of zero and variance of 1
Figure BDA0002942441090000115
The calculation formula is as follows:
Figure BDA0002942441090000116
step 2, inputting the vector into SF1 one-dimensional vector
Figure BDA0002942441090000117
The vectors output by the full-connection layer of the last submodule of the SF1 are fed to a self-defined activation function through the processing of the submodules connected in series in the SF1, the transmission power of all the relay nodes is output, the activation function sets the negative output power value to be 0, and when the output power value is 0, the corresponding relay node does not participate in cooperative transmission, so that the selection and the power distribution of the relay nodes are completed simultaneously;
the method specifically comprises the following steps: vector x of outputs of full-link layer of last submodule of SF1SF1Feeds into an activation function
Figure BDA0002942441090000121
In the formula, alphaiSetting the output negative value as 0 by equation (9) for the power allocated to the relay node i, and ensuring that the power of the relay node i satisfies the requirement
Figure BDA0002942441090000122
The restriction condition (7d) is satisfied; [ x ] ofSF1]iRepresenting a vector xSF1If α isi0, indicates that the relay node i does not participate in cooperationAnd (5) transmitting.
And 3, feeding the output of the full connection layer of the last submodule of the SF2 to a self-defined activation function, wherein the number of the output is 1, and completing the power distribution of the source node.
The method specifically comprises the following steps: output x of full link layer in last submodule of SF2SF2Is fed to a custom activation function, the output is represented as
Figure BDA0002942441090000123
In the formula (I), the compound is shown in the specification,
Figure BDA0002942441090000124
the number of hidden nodes for the last submodule of SF 2; p is a radical ofs∈(0,Ps max) Is the transmission power of the source node, and the limitation condition (7c) is satisfied; [ x ] ofSF2]iRepresenting a vector xSF2The ith value of (a).
Example 4
Training of combined deep neural network models
Inputting K samples as a group of training data into a combined deep neural network model, then training by using adaptive moment estimation, and synchronously updating network parameters of two sub-networks by using the following loss functions in the training process of the combined deep neural network model in order to achieve the aim of minimizing the energy consumption of a system and simultaneously meet the requirement of the system rate and the total power constraint of a relay node:
Figure BDA0002942441090000131
wherein E {. denotes averaging K samples; lambda is more than or equal to 01≤1，0≤λ 21 or less and 0 or less3Control parameters are less than or equal to 1; [ x ] of]+＝max(x,0)；[Rth-Rd]+Corresponding to the formula (7b), when the constraint condition (7b) is satisfied, the loss caused by this portion becomes zero;
Figure BDA0002942441090000132
corresponding to equation (7e) when the constraint condition (7e) is satisfied, the loss caused by this portion becomes zero; (f)Hf+ps) Corresponding to the formula (7a), the smaller the loss caused by the part is, the better the loss is, on the premise that (7b) (7e) is satisfied; theta1Set of all weights and offsets for SF1, θ2Set of all weights and offsets for SF 2; it should be noted that λ is too large2Or λ3The system performance of the CDNN is degraded because the optimization will focus on satisfying the constraint terms rather than optimizing the feature function, and λ2Or λ3Too small will not satisfy the constraints in the optimization.
In the extreme process of minimizing the loss function, the direction of the extreme value is reversed to theta1And theta2The update is performed as follows:
Figure BDA0002942441090000133
Figure BDA0002942441090000134
wherein, 0 < beta 11 and 0 beta 21 respectively represents the corresponding learning rate; t represents the number of updates;
Figure BDA0002942441090000135
representing the loss function L vs. theta1Taking a partial derivative;
Figure BDA0002942441090000136
representing the loss function L vs. theta2Taking a partial derivative; and the number of the first and second electrodes,
Figure BDA0002942441090000137
and
Figure BDA0002942441090000138
initial weights and offsets for SF1 and SF2, respectively.
During the training processWhen the CDNN model can not continuously exceed the set baseline (the interruption probability threshold P)out) The training algorithm will stop. And then, storing the learned CDNN model for testing.
Example 5
Resource allocation for energy efficiency performance of simulation test
Let Ps max＝20dBm,
Figure BDA0002942441090000139
B＝4Hz,and
Figure BDA00029424410900001310
Figure BDA00029424410900001311
The coordinates of the source node are set to be (x is 0, y is 0), the coordinates of the target node are (x is 20m, y is 20m), and then the N potential relays are randomly distributed in a square area with the coordinates x being 0-20 m and y being 0-20 m. Small scale fading uses independent and uniformly distributed complex gaussian random variables with a mean of zero and a variance of 1. The path loss exponent is set to k-3. In addition, 10000 samples were generated for training and testing.
The parameter settings of the CDNN model are shown in table 1.
TABLE 1 CDNN model parameter configuration
Parameter(s) Value of
K 100
P out 5％
M1,M 2 3
The last submodule of the SF1 is N 256
β1，β2 10-4
λ1，λ2，λ 3 1
Comparative example
In order to evaluate the performance superiority of the energy efficiency resource allocation method of the deep learning amplification forwarding cooperative network, the method is compared with the existing 'fixed source node power' scheme, wherein the 'fixed source node power' scheme is that the source node power is set as a fixed value, and only the power of the relay node is optimized. Meanwhile, in order to evaluate the performance of an arctan function-based active layer (referred to as "arctan CDNN"), it was also compared with a Sigmoid function-based active layer referred to as "Sigmoid CDNN", which are respectively defined as follows:
Figure BDA0002942441090000141
Figure BDA0002942441090000142
first, the system rate requirement is set to R th2 bit/s. Fig. 3 and 4 are performance graphs of average energy efficiency and outage probability of the system at different numbers of relay nodes. As can be seen from the figure, the average energy efficiency of the system is also improved as the number of potential relays increases. Also, in the case of interruption probabilityIn the case of deterioration, the proposed CDNN model may achieve higher energy efficiency than the "fixed source node power" approach. For a "fixed source node power" scheme, when the source node sets a low transmit power (p)s0.05mW), the performance of the outage probability becomes poor and the average energy efficiency of the system becomes very low.
Fig. 5 is a diagram of the average energy efficiency of the system under different system rate requirements with the number N of potential relays being 6. As can be seen from the figure, the proposed arctan function based active layer has a higher system average energy efficiency than the Sigmoid function based active layer. When the system speed requirement is increased, the distance between the two curves is gradually reduced.
TABLE 2Rth2bit/s and RthAn example of power allocation at 10 bit/s.
Table 2 example of power allocation (N ═ 6, | h)s,d|2＝4.0*10-5)
Figure BDA0002942441090000151
From fig. 5 and table 2, the following conclusions can be drawn: it is proposed that an arctan function based active layer may achieve a better power allocation factor than a Sigmoid function based active layer, especially when the wireless nodes communicate at low transmit power.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims (10)

1. A combined deep neural network model, comprising: the system comprises two independent fully-connected sub-networks, namely SF1 and SF2, wherein the two sub-networks are formed by connecting limited sub-modules in series, the SF1 sub-network is used for determining the power distribution of the relay nodes, the sub-networks except the last sub-module are formed by a fully-connected layer followed by a hyperbolic tangent function, the last sub-module of the sub-network is formed by a fully-connected layer followed by a self-defined activation function, the number of hidden nodes of the last sub-module is the same as the number of potential relay nodes in the amplification forwarding cooperation network, and each output determines the power distribution of one relay node; the SF2 sub-network is used for determining the power distribution of the source node, the sub-network except the last sub-module consists of a full connection layer followed by a rectification linear unit, the last sub-module consists of a full connection layer followed by a self-defined activation function, and the output is the power of the source node.
2. A deep learning energy efficiency resource allocation method for an amplification forwarding cooperative network comprises the following steps: the method is characterized by comprising the following steps:
step 1, establishing a source node and relay node power distribution joint optimization model for amplifying and forwarding cooperative network minimum energy consumption;
step 2, constructing a combined deep neural network model;
step 3, training the constructed combined deep neural network model;
and 4, testing by using the trained and learned combined deep neural network model.
3. The method for allocating the energy efficiency resources of the deep learning amplify-and-forward cooperative network according to claim 2, wherein: the establishment of the source node and relay node power distribution joint optimization model for amplifying and forwarding the minimum energy consumption of the cooperative network in the step 1 comprises the following steps:
step 1, calculating the total rate of an amplification forwarding cooperative network system;
and 2, establishing a power distribution joint optimization model of the source node and the relay node for amplifying and forwarding the minimum energy consumption of the cooperative network.
4. The method for allocating the energy efficiency resources of the deep learning amplify-and-forward cooperative network according to claim 3, wherein: the amplifying and forwarding cooperative network system in the step 1 comprises a source node, a destination node and N potential relay nodes; the calculation of the total rate of the amplifying and forwarding cooperative network system comprises the following processes:
assuming that the channel states from a source node to a destination node, from the source node to all potential relay nodes and from all potential relay nodes to the destination node are known;
further, assuming that a transmission time slot is divided into two parts, in a first part of time period, a source node broadcasts information to a destination node and all potential relay nodes, which is called a broadcasting stage; in the second part of time period, the relay node forwards the information received from the source node, which is called a cooperation stage;
in the broadcast phase, a source node transmits a signal x, and signals received by a destination node and any relay node are respectively:
Figure FDA0002942441080000021
Figure FDA0002942441080000022
in the formula, psFor the transmission power of the source node, assume that the amplitude of the signal x transmitted by the source node is 1, hs,dAnd hs,iChannel gains, w, from source node to destination node, and from source node to relay node i, respectivelys,dAnd ws,iIn the broadcasting stage, the receiving noise of a target node and a relay node i; assuming that the noise received by all the relay nodes and the destination node obeys complex Gaussian distribution, and the mean value is zero;
in the cooperation stage, the receiving, by the destination node, of the signal amplified and forwarded by the relay node is:
Figure FDA0002942441080000023
in the formula (I), the compound is shown in the specification,
Figure FDA0002942441080000024
the relay node receives the variance of the noise; h isi,dIs the channel gain from the relay node i to the destination node; w is ar,dIs the reception noise of the destination node in the cooperation phase,
Figure FDA0002942441080000031
is the power normalization factor of the relay node i; vector quantity
Figure FDA0002942441080000032
Is the forward power vector of all potential relay nodes [. C]TRepresenting transpose of matrix or vector, alphaiRepresenting the forward power, vector, of the relay node i
Figure FDA0002942441080000033
(Vector)
Figure FDA0002942441080000034
Combining formulae (1) and (3) can result in:
Figure FDA0002942441080000035
according to equation (4), the signal-to-noise ratio of the destination node is obtained:
Figure FDA0002942441080000036
in the formula (I), the compound is shown in the specification,
Figure FDA0002942441080000037
variance, H, representing the received noise of the destination nodeHRepresenting the vector H by transposition, fHTaking a transposition of the expression vector f; wHThe expression vector W is transposed;
the total rate obtained by the destination node of the amplifying and forwarding cooperative network system is as follows:
Rd＝Blog2(1+γd) (6)
where B is the bandwidth of the channel.
5. The method for allocating the energy efficiency resources of the deep learning amplify-and-forward cooperative network according to claim 3, wherein: the power distribution joint optimization model and the constraint conditions of the source node and the relay node for amplifying the minimum energy consumption of the forwarding cooperative network in the step 2 are as follows:
Figure FDA0002942441080000038
satisfies the following conditions: rd≥Rth (7b)
0＜ps＜Ps max (7c)
Figure FDA0002942441080000041
Figure FDA0002942441080000042
In the formula (f)Hf represents the sum of the powers of all relay nodes, RthIndicating the system rate requirement, Ps maxIs the power threshold of the source node,
Figure FDA0002942441080000043
a threshold value representing the forwarding power of each relay node,
Figure FDA0002942441080000044
is the threshold value of the sum of the forwarding power of all the relay nodes; (7b) representing a rate constraint required by the amplifying and forwarding cooperative network system; (7c) representing a sourcePower constraints of the nodes; (7d) represents a power limit for each relay node; (7e) representing a total power constraint for all relay nodes.
6. The method for allocating the energy efficiency resources of the deep learning amplify-and-forward cooperative network according to claim 2, wherein: the specific steps of the combined deep neural network model constructed in the step 2 for deciding the power distribution of the source node and the relay node are as follows:
step 1, normalizing the channel gain one-dimensional vector
Figure FDA0002942441080000046
Channel gains from a source node to a destination node, from the source node to all potential relay nodes and from all potential relay nodes to the destination node are respectively input into SF1 sub-networks and SF2 sub-networks;
step 2, inputting the vector into SF1 one-dimensional vector
Figure FDA0002942441080000047
The vector output by the full connection layer in the last submodule of the SF1 is fed to a self-defined activation function through the processing of the submodules connected in series in the SF1, the transmission power corresponding to each potential relay node is output, the activation function sets the negative output power value to be 0, and when the output power value is 0, the corresponding potential relay node does not participate in cooperative transmission, so that the relay node selection and the power distribution are completed simultaneously;
step 3, inputting the vector into SF2 one-dimensional vector
Figure FDA0002942441080000045
And through the processing of the submodules connected in series in the SF2, the output of the full-connection layer in the last submodule of the SF2 is fed to a self-defined activation function, the number of the output is 1, and the power distribution of the source node is completed.
7. The method for allocating the energy efficiency resources of the deep learning amplify-and-forward cooperative network according to claim 6, wherein: the specific steps of relay node selection and power allocation in step 2 are as follows:
vector x of full connected layer output in last submodule of SF1SF1Feeds into an activation function
Figure FDA0002942441080000051
In the formula, alphaiSetting the output negative value as 0 by equation (8) for the power allocated to the relay node i, and ensuring that the power of the relay node i satisfies the requirement
Figure FDA0002942441080000052
The restriction condition (7d) is satisfied; [ x ] ofSF1]iRepresenting a vector xSF1If α isi0, indicates that the relay node i does not participate in cooperative transmission.
8. The method for allocating the energy efficiency resources of the deep learning amplify-and-forward cooperative network according to claim 6, wherein: the specific steps of the power distribution of the source node in the step 3 are as follows:
output x of full link layer in last submodule of SF2SF2Is fed to a custom activation function, the output is represented as
Figure FDA0002942441080000053
In the formula (I), the compound is shown in the specification,
Figure FDA0002942441080000054
the number of hidden nodes for the last submodule of SF 2; p is a radical ofs∈(0,Ps max) Is the transmission power of the source node, and the limitation condition (7c) is satisfied; [ x ] ofSF2]iRepresenting a vector xSF2The ith value of (a).
9. The method for allocating energy efficiency resources of the deep learning amplify-and-forward cooperative network according to claim 2, wherein the process of training the built combined deep neural network model in the step 3 is as follows: the multiple samples are input into the built combined deep neural network model as a set of training data, then training is carried out by using the adaptive moment estimation, and when the combined deep neural network model cannot continuously exceed the set baseline, the training algorithm stops.
10. The method for allocating the energy efficiency resources of the deep learning amplify-and-forward cooperative network according to claim 9, wherein: in order to achieve the goal of minimizing the system energy consumption and simultaneously meet the system rate requirement and the total power constraint of the relay nodes, the network parameters of the two sub-networks need to be synchronously updated by the following loss functions in the training process of the combined deep neural network model:
Figure FDA0002942441080000061
wherein E {. denotes averaging a set of training data; lambda is more than or equal to 01≤1，0≤λ21 or less and 0 or less3Control parameters are less than or equal to 1; [ x ] of]+＝max(x,0)；[Rth-Rd]+Corresponding to the constraint (7b), if satisfied, the value is 0;
Figure FDA0002942441080000062
corresponding to the constraint (7e), if satisfied, the value is 0; (f)Hf+ps) Corresponding to equation (7 a); theta1Set of all weights and offsets for SF1, θ2Set of all weights and offsets for SF 2;
in the process of minimizing the loss function, the direction is continuously opposite to theta1And theta2The update is performed as follows:
Figure FDA0002942441080000063
Figure FDA0002942441080000064
wherein, 0 < beta11 and 0 beta21 respectively represents the corresponding learning rate; t represents the number of updates;
Figure FDA0002942441080000065
representing the loss function L vs. theta1Taking a partial derivative;
Figure FDA0002942441080000066
representing the loss function L vs. theta2And (4) taking partial derivatives.
CN202110184433.7A 2021-02-10 2021-02-10 Deep learning amplification forwarding cooperative network energy efficiency resource allocation method Active CN112929977B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110184433.7A CN112929977B (en) 2021-02-10 2021-02-10 Deep learning amplification forwarding cooperative network energy efficiency resource allocation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110184433.7A CN112929977B (en) 2021-02-10 2021-02-10 Deep learning amplification forwarding cooperative network energy efficiency resource allocation method

Publications (2)

Publication Number Publication Date
CN112929977A true CN112929977A (en) 2021-06-08
CN112929977B CN112929977B (en) 2022-05-31

Family

ID=76169742

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110184433.7A Active CN112929977B (en) 2021-02-10 2021-02-10 Deep learning amplification forwarding cooperative network energy efficiency resource allocation method

Country Status (1)

Country Link
CN (1) CN112929977B (en)

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140269989A1 (en) * 2013-03-14 2014-09-18 Arizona Board of Regents, a body corporate of the state of Arizona, acting for and on behalf of Energy efficient distributed estimation using nonlinear amplifiers
CN108156034A (en) * 2017-12-22 2018-06-12 武汉噢易云计算股份有限公司 A kind of message forwarding method and message forwarding system based on deep neural network auxiliary
CN109726200A (en) * 2018-12-06 2019-05-07 国网甘肃省电力公司信息通信公司 Grid information system fault location system and method based on two-way deep neural network
CN109936399A (en) * 2019-03-07 2019-06-25 西北工业大学 A kind of insincere junction network antenna selecting method based on deep neural network
WO2019154992A1 (en) * 2018-02-08 2019-08-15 Sony Corporation Position estimation device and communication device
US20190258919A1 (en) * 2017-04-17 2019-08-22 Cerebras Systems Inc. Wavelet representation for accelerated deep learning
CN111191769A (en) * 2019-12-25 2020-05-22 中国科学院苏州纳米技术与纳米仿生研究所 Self-adaptive neural network training and reasoning device
CN111741483A (en) * 2019-10-29 2020-10-02 青岛科技大学 Interrupt probability performance prediction method for mobile communication system
CN111865844A (en) * 2020-05-29 2020-10-30 北京百卓网络技术有限公司 Channel estimation method and device for large-scale MIMO full-duplex relay system
CN111885671A (en) * 2020-07-17 2020-11-03 燕山大学 Underwater joint relay selection and power distribution method based on deep reinforcement learning
CN111970718A (en) * 2020-07-22 2020-11-20 西北工业大学 Deep learning-based power distribution method in energy collection untrusted relay network
CN112243252A (en) * 2020-09-08 2021-01-19 西北工业大学 Safety transmission enhancement method for relay network of unmanned aerial vehicle

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140269989A1 (en) * 2013-03-14 2014-09-18 Arizona Board of Regents, a body corporate of the state of Arizona, acting for and on behalf of Energy efficient distributed estimation using nonlinear amplifiers
US20190258919A1 (en) * 2017-04-17 2019-08-22 Cerebras Systems Inc. Wavelet representation for accelerated deep learning
CN108156034A (en) * 2017-12-22 2018-06-12 武汉噢易云计算股份有限公司 A kind of message forwarding method and message forwarding system based on deep neural network auxiliary
WO2019154992A1 (en) * 2018-02-08 2019-08-15 Sony Corporation Position estimation device and communication device
CN109726200A (en) * 2018-12-06 2019-05-07 国网甘肃省电力公司信息通信公司 Grid information system fault location system and method based on two-way deep neural network
CN109936399A (en) * 2019-03-07 2019-06-25 西北工业大学 A kind of insincere junction network antenna selecting method based on deep neural network
CN111741483A (en) * 2019-10-29 2020-10-02 青岛科技大学 Interrupt probability performance prediction method for mobile communication system
CN111191769A (en) * 2019-12-25 2020-05-22 中国科学院苏州纳米技术与纳米仿生研究所 Self-adaptive neural network training and reasoning device
CN111865844A (en) * 2020-05-29 2020-10-30 北京百卓网络技术有限公司 Channel estimation method and device for large-scale MIMO full-duplex relay system
CN111885671A (en) * 2020-07-17 2020-11-03 燕山大学 Underwater joint relay selection and power distribution method based on deep reinforcement learning
CN111970718A (en) * 2020-07-22 2020-11-20 西北工业大学 Deep learning-based power distribution method in energy collection untrusted relay network
CN112243252A (en) * 2020-09-08 2021-01-19 西北工业大学 Safety transmission enhancement method for relay network of unmanned aerial vehicle

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
YULIA GAVRILOVA: "A Guid to Deep Learning and Neural Networks", 《HTTPS://SEROKELL.IO/BLOG/DEEP-LEARNING-AND-NEURAL-NETWORK-GUIDE》 *
刘琨: "基于能量效率的无线网络协作通信技术研究", 《信息科技辑》 *
胡文杰: "基于深度强化学习的协作通信中继选择", 《电讯技术》 *
贾鹤萍: "认知网络能效性协作功率分配算法", 《测试技术学报》 *
郭艳艳: "协作中继高效性传输技术研究", 《信息科技辑》 *
陈峰燕: "认知无线电网络中协作中继技术的研究", 《信息科技辑》 *

Also Published As

Publication number Publication date
CN112929977B (en) 2022-05-31

Similar Documents

Publication Publication Date Title
CN109302250B (en) Method for performing relay selection and transmitting power distribution by energy transmission full duplex relay
CN108924935A (en) A kind of power distribution method in NOMA based on nitrification enhancement power domain
CN111385011B (en) Wireless power supply relay network system based on full duplex and optimization method
Cao et al. Deep reinforcement learning for channel and power allocation in UAV-enabled IoT systems
CN113411105B (en) AP selection method of non-cell large-scale antenna system
CN111431646B (en) Dynamic resource allocation method in millimeter wave system
CN104767556A (en) Beamforming coordination method and device based on partial interference alignment
CN106231665B (en) Resource allocation methods based on the switching of RRH dynamic mode in number energy integrated network
CN115278707A (en) NOMA terahertz network energy efficiency optimization method based on assistance of intelligent reflecting surface
CN106304111A (en) Isomery cellular network power allocation method based on energy acquisition relay station
Liu et al. Distributed relay selection for heterogeneous UAV communication networks using a many-to-many matching game without substitutability
CN115175220A (en) Communication resource allocation method and device based on unmanned aerial vehicle ad hoc network
CN106788934B (en) The pilot distribution method of multi-plot joint in a kind of extensive mimo system
CN108880633B (en) Joint design optimization method for beamforming antenna selection grouping algorithm
CN107087279B (en) Base station activation and beam forming method based on stable access relation
CN111741520B (en) Cognitive underwater acoustic communication system power distribution method based on particle swarm
Song et al. Maximizing packets collection in wireless powered IoT networks with charge-or-data time slots
CN112929977B (en) Deep learning amplification forwarding cooperative network energy efficiency resource allocation method
CN114745032B (en) Honeycomb-free large-scale MIMO intelligent distributed beam selection method
CN107182116A (en) Interference control method based on power distribution in Full-duplex cellular GSM
Jothi et al. Nelder mead-based spider monkey optimization for optimal power and channel allocation in MANET
CN116321236A (en) RIS-assisted safe honeycomb-free large-scale MIMO system energy efficiency optimization method
CN103873126A (en) Power optimization method based on genetic algorithm in multi-hop collaborative network
Li et al. Distributed RIS-enhanced cell-free NOMA networks
CN113595609A (en) Cellular mobile communication system cooperative signal sending method based on reinforcement learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant