CN111162888B - Distributed antenna system, remote access unit, power distribution method, and medium - Google Patents
Distributed antenna system, remote access unit, power distribution method, and medium Download PDFInfo
- Publication number
- CN111162888B CN111162888B CN201911323599.1A CN201911323599A CN111162888B CN 111162888 B CN111162888 B CN 111162888B CN 201911323599 A CN201911323599 A CN 201911323599A CN 111162888 B CN111162888 B CN 111162888B
- Authority
- CN
- China
- Prior art keywords
- power
- state information
- channel state
- model
- trained
- 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.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L5/00—Arrangements affording multiple use of the transmission path
- H04L5/003—Arrangements for allocating sub-channels of the transmission path
- H04L5/0053—Allocation of signaling, i.e. of overhead other than pilot signals
- H04L5/0057—Physical resource allocation for CQI
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W72/00—Local resource management
- H04W72/50—Allocation or scheduling criteria for wireless resources
- H04W72/54—Allocation or scheduling criteria for wireless resources based on quality criteria
- H04W72/542—Allocation or scheduling criteria for wireless resources based on quality criteria using measured or perceived quality
-
- 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
Abstract
The invention discloses a power distribution method, which comprises the following steps: acquiring channel state information between a user terminal and a remote access unit; acquiring target power corresponding to the channel state information according to a preset mapping relation between the channel state information and the power; and changing the transmitting power of the antenna corresponding to the user terminal into the target power. The invention also provides a remote access unit, a distributed antenna system and a medium. The invention has reasonable power distribution of the remote access unit to the user terminal.
Description
Technical Field
The present invention relates to the field of communications technologies, and in particular, to a distributed antenna system, a remote access unit, a power distribution method, and a medium.
Background
With the rapid development of communications, the rapid increase in communication rate and energy consumption has brought a great challenge to modern wireless communication networks. In recent years, researchers have continuously sought new solutions to reform the conventional Co-located Antenna Systems (CAS) to meet the rapidly increasing user demand. In addition, with the rapid development of communication systems and intelligent terminals, the number of intelligent terminals is increasing explosively, so that the energy consumption of future communication networks is immeasurable. Researchers are constantly looking for new schemes to improve Spectral Efficiency (SE) and Energy Efficiency (EE). Therefore, how to improve The spectrum efficiency and reduce The energy consumption becomes an important point of The next generation communication network (5G).
Power allocation for remote access units in CAS is also important in order to provide high rate data transmission. In an exemplary technology, an antenna system adopts an average distribution scheme for power distribution of a user terminal, such a power distribution mode cannot maximize spectrum efficiency or energy consumption efficiency of a remote access unit, and power distribution of the remote access unit to the user terminal is not reasonable.
Disclosure of Invention
The invention mainly aims to provide a distributed antenna system, a remote access unit, a power distribution method and a medium, and aims to solve the problem that the power distribution of the remote access unit to a user terminal is unreasonable.
In order to achieve the above object, the present invention provides a power allocation method, including the steps of:
acquiring channel state information between a user terminal and a remote access unit;
acquiring target power corresponding to the channel state information according to a preset mapping relation between the channel state information and the power; and
and changing the transmitting power of the antenna corresponding to the user terminal into the target power.
In an embodiment, the mapping relationship is realized by a power distribution model, the power distribution model is obtained by training a plurality of pieces of channel state information and tag powers corresponding to the channel state information, and the tag powers are determined according to the channel state information.
In an embodiment, before the step of obtaining the channel state information between the user terminal and the remote access unit, the method further includes:
sequentially inputting a plurality of pieces of channel state information in a data set and label power corresponding to each piece of channel state information into a model to be trained so as to train the model to be trained;
stopping training the model when the training parameters of the trained model meet the preset conditions, wherein after the trained model outputs the distributed power of the channel state information, judging whether the training parameters of the trained model meet the preset conditions or not;
and saving the model which stops training as a power distribution model.
In an embodiment, the step of sequentially inputting the plurality of pieces of channel state information in the data set and the tag power corresponding to each piece of channel state information into a model to be trained to train the model to be trained includes:
inputting the current channel state information into the model to be trained for training, and acquiring the distribution power output by the trained model;
acquiring tag power corresponding to current channel state information, and determining the distribution power and the mean square error between the tag powers;
and judging whether the mean square error is less than or equal to a preset mean square error or not, wherein the training parameters of the trained model comprise the mean square error, and when the mean square error is less than or equal to the preset mean square error, judging that the trained model meets a preset condition.
In an embodiment, after the step of determining whether the mean square error is smaller than a preset mean square error, the method further includes:
when the mean square error is larger than a preset mean square error, adjusting parameters of a network in the trained model;
and taking the next channel state information in the data set as the current channel state information, and returning to the step of acquiring the distributed power output by the trained model.
In an embodiment, the training parameter includes a convergence value of a neural network, a mean square error between a distributed power output by a trained model and a tag power of channel state information corresponding to the distributed power, or a training number, and the preset condition includes:
the convergence value of the neural network is smaller than a preset convergence value;
the mean square error is less than a preset mean square error;
or the training times reach preset times.
In an embodiment, the mapping relationship is obtained by learning, by a neural network, a plurality of pieces of channel state information and tag powers corresponding to the channel state information, where the tag powers are determined according to the channel state information.
To achieve the above object, the present invention further provides a remote access unit, which includes a processor, a memory and a power allocation program stored in the memory and operable on the processor, the remote access unit is provided with a mapping relationship between channel state information and power, and the power allocation program, when executed by the processor, implements the steps of the power allocation method as described above.
In order to achieve the above object, the present invention further provides a distributed antenna system, where the distributed antenna system includes multiple remote access units, each remote access unit is provided with a mapping relationship between channel state information and power, and the remote access unit is configured to implement the above steps of the power allocation method.
To achieve the above object, the present invention also provides a medium storing a power allocation program, which when executed by a processor, implements the steps of the power allocation method as described above.
In the distributed antenna system, the remote access unit, the power distribution method and the medium provided by the embodiment of the invention, the remote access unit acquires channel state information between the user terminal and the remote access unit, determines the target power of the user terminal according to the mapping relation between the channel state information and the power, and then changes the transmitting power of the antenna corresponding to the user terminal into the target power. The mapping relation between the channel state information and the power is stored in the remote access unit, so that the remote access unit can reasonably determine the target power of the user terminal according to the mapping relation and the channel state information, and the power of the user terminal is reasonably distributed by the remote access unit.
Drawings
Fig. 1 is a schematic diagram of a hardware architecture of a remote access unit according to an embodiment of the present invention;
fig. 2 is a schematic system architecture diagram of a distributed antenna system according to an embodiment of the present invention;
FIG. 3 is a flowchart illustrating a power allocation method according to a first embodiment of the present invention;
FIG. 4 is a flowchart illustrating a power allocation method according to a second embodiment of the present invention;
FIG. 5 is a detailed flowchart of step S30 in FIG. 4;
fig. 6 is a flowchart illustrating a power allocation method according to a third embodiment of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not to be construed as limiting the invention.
The main solution of the embodiment of the invention is as follows: acquiring channel state information between a user terminal and a remote access unit; acquiring target power corresponding to the channel state information according to a preset mapping relation between the channel state information and the power; and changing the transmitting power of the antenna corresponding to the user terminal into the target power.
The mapping relation between the channel state information and the power is stored in the remote access unit, so that the remote access unit can reasonably determine the target power of the user terminal according to the mapping relation and the channel state information, and the remote access unit can reasonably distribute the power of the user terminal.
As shown in fig. 1, fig. 1 is a schematic diagram of a hardware structure of a remote access unit according to an embodiment of the present invention.
As shown in fig. 1, the remote access unit may include: a processor 1001, such as a CPU, a communication bus 1002, and a memory 1003. Wherein a communication bus 1002 is provided to enable connective communication between these components. The memory 1003 may be a high-speed RAM memory or a non-volatile memory (e.g., a disk memory). The memory 1003 may alternatively be a storage device separate from the processor 1001.
Those skilled in the art will appreciate that the configuration shown in fig. 1 does not constitute a limitation of the remote access unit and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.
As shown in fig. 1, the memory 1003, which is a kind of computer storage medium, may include therein an operating system and a power allocation program.
In the apparatus shown in fig. 1, the processor 1001 may be arranged to call a power allocation program stored in the memory 1003 and perform the following operations:
acquiring channel state information between a user terminal and a remote access unit;
acquiring target power corresponding to the channel state information according to a preset mapping relation between the channel state information and the power; and
and changing the transmitting power of the antenna corresponding to the user terminal into the target power.
In one embodiment, processor 1001 may invoke a power allocation program stored in memory 1003 and further perform the following operations:
the mapping relation is realized through a power distribution model, the power distribution model is obtained through training of a plurality of pieces of channel state information and label power corresponding to each piece of channel state information, and the label power is determined according to the channel state information.
In one embodiment, processor 1001 may invoke a power allocation program stored in memory 1003 and further perform the following operations:
sequentially inputting a plurality of pieces of channel state information in a data set and label power corresponding to each piece of channel state information into a model to be trained so as to train the model to be trained;
stopping training the model when the training parameters of the trained model meet the preset conditions, wherein after the trained model outputs the distributed power of the channel state information, judging whether the training parameters of the trained model meet the preset conditions or not;
and saving the model which stops training as a power distribution model.
In one embodiment, processor 1001 may invoke a power allocation program stored in memory 1003 and further perform the following operations:
inputting the current channel state information into the model to be trained for training, and acquiring the distribution power output by the trained model;
acquiring tag power corresponding to current channel state information, and determining the distribution power and the mean square error between the tag powers;
and judging whether the mean square error is less than or equal to a preset mean square error or not, wherein the training parameters of the trained model comprise the mean square error, and when the mean square error is less than or equal to the preset mean square error, judging that the trained model meets a preset condition.
In one embodiment, processor 1001 may invoke a power allocation program stored in memory 1003 and further perform the following operations:
when the mean square error is larger than a preset mean square error, adjusting parameters of a network in the trained model;
and taking the next channel state information in the data set as the current channel state information, and returning to the step of acquiring the distributed power output by the trained model.
In one embodiment, processor 1001 may invoke a power allocation program stored in memory 1003 and further perform the following operations:
the convergence value of the neural network is smaller than a preset convergence value;
the mean square error is less than a preset mean square error;
or the training times reach preset times.
In one embodiment, processor 1001 may invoke a power allocation program stored in memory 1003 and further perform the following operations:
the mapping relation is obtained by learning a plurality of channel state information and label power corresponding to each channel state information through a neural network, and the label power is determined according to the channel state information.
As one implementation, a distributed antenna system may be as shown in fig. 2.
Referring to fig. 2, the distributed antenna system includes a plurality of Remote Access Units (RAUs), specifically RAUs 1, RAUs 2, RAUs 3, RAUs 4, and RAUs 5, the RAU1 located in the middle can be regarded as a special Central Unit (CU), and other RAUs are connected to the RAU1 through optical fibers. All RAUs are low power single antenna Base Stations (BS). In addition, there are K cellular Users (UE) randomly distributed in the area where the distributed antenna system is located, specifically, UE1, and UE3. Each remote access unit may implement power allocation to the user terminal.
Based on the hardware construction, various embodiments of the power distribution method of the invention are provided.
Referring to fig. 3, a first embodiment of the present invention provides a power allocation method, including the steps of:
step S10, acquiring channel state information between a user terminal and a remote access unit;
in this embodiment, the execution subject is a remote access unit. The remote access unit may be provided with a channel model, and the remote access unit obtains channel state information between the user terminal and the remote access unit through the channel model.
In particular, use is made of n,k To represent channel state information between the nth RAU and the kth UE, the channel state information including small-scale fading and large-scale fading. The information about the channel state may be expressed as: h is a total of n,k ＝g n,k *w n,k Wherein g is n,k Representing small-scale fading w between the nth RAU to the kth UE n,k Representing large scale fading, w n,k Independent of small-scale fading, it can be expressed as:
where c is the average path gain at a reference distance of 1 km. d is a radical of n,k Indicating the distance between the nth RAU to the kth UE. Alpha is a path fading coefficient, and usually takes a value in the range of [3,5 ]]。s n,k Is a lognormal distributed fading variable, i.e. 10log 10 s n,k Has a mean value of 0 and a standard deviation of σ n,k . The remote access unit can acquire large-scale fading, small-scale fading and distance between the remote access unit and the user terminal, so that the channel state information between the user terminal and the remote access unit can be determined according to the acquired parameters and formulas.
Step S20, acquiring target power corresponding to the channel state information according to a preset mapping relation between the channel state information and the power;
the remote access unit is provided with a mapping relation between the channel state information and the power, and the mapping relation is determined according to a power distribution algorithm. The power distribution algorithm comprises a sub-step algorithm, a water filling algorithm or a concave-convex process algorithm. The relation between the channel state information and the power can be learned through a neural network, meanwhile, the channel state information is calculated through a power distribution algorithm to obtain the label power, the relation between the channel state information and the power is corrected through the label power, and finally the mapping relation capable of outputting the target power of the user terminal more accurately is obtained, and the mapping relation can be stored in a remote access unit.
Further, the user may set the problem to be solved by the power allocation of the remote access unit. The problem is divided into maximizing spectral efficiency and maximizing energy efficiency. Therefore, the power labels adopted for different problems enable the neural network to carry out the relation between the channel state information and the power. For example, when the problem to be solved by the remote access unit is to maximize energy efficiency, the power corresponding to the channel state information under the condition of maximizing the energy efficiency is firstly performed through a power classification algorithm, the power is used as tag power, then the tag power is adopted to compare the power output after the neural network learns the relationship between the channel state information and the power, and the parameter of the neural network is adjusted according to the comparison result, so that the power output by the neural network is closer and closer to the power corresponding to the channel state information under the condition of maximizing the energy efficiency, which is calculated by a power distribution algorithm, thereby the neural network learns the final mapping relationship between the channel state information and the power, and the error between the power determined under the mapping relationship and the tag power is within an acceptable range. After the remote access unit acquires the channel state information between the user terminal and the remote access unit, the target power of the user terminal can be determined according to the stored mapping relation and the channel state information.
And step S30, changing the transmitting power of the antenna corresponding to the user terminal into the target power.
After the target power is determined, the remote access unit can determine a target antenna corresponding to the user terminal, so that the transmitting power of the target antenna is changed into the target power, the target power of the user terminal is distributed, the problem of maximizing energy efficiency or the problem of maximizing spectrum efficiency, which is set in the remote access unit, is solved, and the power configuration is reasonably performed on the user terminal.
In the technical solution provided in this embodiment, the remote access unit obtains channel state information between the user terminal and the remote access unit, determines a target power of the user terminal according to a mapping relationship between the channel state information and the power, and then changes a transmission power of a corresponding antenna of the user terminal to the target power. The mapping relation between the channel state information and the power is stored in the remote access unit, so that the remote access unit can reasonably determine the target power of the user terminal according to the mapping relation and the channel state information, and the power of the user terminal is reasonably distributed by the remote access unit.
Referring to fig. 4, fig. 4 is a second embodiment of the power allocation method of the present invention, and based on the first embodiment, before the step S10, the method further includes:
step S40, sequentially inputting a plurality of pieces of channel state information in a data set and label power corresponding to each piece of channel state information into a model to be trained so as to train the model to be trained;
the power allocation algorithm can be regarded as user channel state information h n,k Corresponding power allocation p n,k A non-linear mapping relationship between the two, and the neural network can solve the problem of the non-linear mapping. Thus, the user channel state information h can be used n,k Power allocation p as input to neural networks and based on conventional sub-gradient algorithms n,k Training as a label for the neural network, thereby enabling the neural network to learn the user channel state information h n,k With corresponding power allocation p n,k Fitting the traditional times by the relationship of nonlinear mapping between the twoThe gradient power distribution algorithm is adopted, so that the trained neural network model has the performance of power distribution similar to that of the traditional sub-gradient algorithm.
According to the above principle, the mapping relationship between the channel state information and the power may be set in a power distribution model of the remote access unit, that is, the mapping relationship is obtained by training the model, and the trained model is the power distribution model. The power distribution model is a neural network model with a power distribution function.
The power allocation algorithm can be regarded as user channel state information h n,k Corresponding power allocation p n,k A non-linear mapping relationship between the two, and the neural network can solve the problem of the non-linear mapping. Thus, user channel state information h can be used n,k Power allocation p as input to neural networks and based on conventional sub-gradient algorithms n,k Training as labels of the neural network, thereby enabling the neural network to learn the user channel state information h n,k Corresponding power allocation p n,k Fitting a traditional sub-gradient power distribution algorithm by the nonlinear mapping relation between the two sub-gradient power distribution algorithms, so that the trained neural network model has the performance of power distribution similar to that of the traditional sub-gradient algorithm.
Specifically, a plurality of pieces of channel state information are obtained first, and a power distribution algorithm is adopted to calculate each piece of channel state information, so that the tag power corresponding to each piece of channel state information is obtained. And taking the various channel state information and the power label corresponding to the channel state information as a data set.
And inputting the state information of each channel in the data set and the label power into the model to be trained so as to train the model to be trained. The model to be trained is provided with a neural network, and the neural network comprises an input layer, three hidden layers and an output layer. The number of neurons in the feature layer of the neural network is determined according to a training scene, wherein the training scene is the number of remote access units in the distributed antenna system and the number of user terminals in the area where the distributed antenna system is located. For example, in the training scenario where the number of remote access units in the distributed antenna system is 5, and the number of user terminals is 3, the numbers of neurons in one input layer, three hidden layers, and one output layer of the neural network are 15, 32, 64, 128, and 15, respectively. And the training scenario is that the number of remote access units in the distributed antenna system is 10, and the number of user terminals is 10, the number of neurons in an input layer, three hidden layers and an output layer of the neural network is 50, 64, 128, 256 and 50 respectively.
The model to be trained is divided into a model corresponding to the maximized energy efficiency and a model corresponding to the maximized spectral efficiency. If the remote access unit needs to meet the maximum energy efficiency, taking a model corresponding to the maximum energy efficiency as a model to be trained; and if the remote access unit needs to meet the maximum spectrum efficiency, taking the model corresponding to the maximum frequency efficiency as the model to be trained. The model to be trained is provided with a prototype relationship between the channel state information and the power, the power is obtained by inputting the channel state information into the prototype relationship, and the output power is compared with the label power to adjust the parameters of the neural network. Specifically, referring to fig. 5, that is, step S40 includes:
step S41, inputting the current channel state information into the model to be trained for training, and acquiring the distribution power output by the trained model;
step S42, obtaining the label power corresponding to the current channel state information, and determining the mean square error between the distribution power and the label power;
and S43, judging whether the mean square error is less than or equal to a preset mean square error or not, wherein the training parameters of the trained model comprise the mean square error, and judging that the trained model meets a preset condition when the mean square error is less than or equal to the preset mean square error.
And taking the channel state information in the data set as the current channel state information, and inputting the current channel state information into the model to be trained for training so as to obtain the distributed power output by the model to be trained. And calculating the mean square error between the label power corresponding to the current channel state information and the distributed power. In particular, the neural network comprises an inputA fully connected neural network structure of layers, three hidden layers and one output layer. Because any iterative algorithm corresponding to the conventional power allocation algorithm can be regarded as the channel state information h n,k Corresponding power allocation p n,k A non-linear mapping relationship therebetween. For the neural network, the relationship of the nonlinear mapping is learned and the traditional algorithm is fitted. Therefore, in order to increase the nonlinearity of the neural network, the ReLU function is added as an activation function of the hidden layer of the neural network. For the output layer, under the training scenario, p of the output of the neural network n,k Should be between 0 and the maximum transmit power of the remote access unit, is unlikely to be less than 0, is unlikely to be greater than the maximum transmit power of the remote access unit, and outputs p n,k Should be a continuous value between 0 and maximum transmit power and not be a discrete value. Therefore, the neural network does a nonlinear regression task under the model to be trained. The activation function of the output layer chooses to use the ReLU function. Thus, the allocated power of the output of the neural network can be expressed as:
after the model to be trained outputs the distributed power of the channel state information, calculating the mean square error between the distributed power and the tag power, and if the mean square error is larger, indicating that the difference between the distributed power and the tag power is larger, namely the distributed power is more inaccurate. And adjusting parameters of the neural network when the mean square error is judged to be larger than the preset mean square error, taking the next channel state information as the current channel state information, returning to execute the steps of inputting the current channel state information into the model to be trained for training and acquiring the distributed power output by the model to be trained. When the mean square error is less than or equal to the preset mean square error, the error between the distributed power and the target power is smaller, that is, the distributed power output by the model to be trained is within the acceptable range, and at this time, the model to be trained meets the preset condition, that is, the training of the model to be trained can be stopped.
Further, the RMSprop algorithm can be used as an optimization algorithm of a neural network, which is an efficient way of gradient descent in small batches, and uses 0.9 as an attenuation rate. In order to improve the performance of the neural network, the Xavier method can be selected to initialize the weights. The parameters that have a large influence on the training effect of the neural network include learning rate and batch size. Through experimentation, the loss value of the first 500 trainings of the neural network determines the appropriate learning rate and the appropriate batch size, for example, the batch size is 512, and the initial learning rate is selected to be 0.001. The batch size is the training times of the model to be trained.
In order to enable the trained neural network model to perform power distribution under different maximum transmission powers of the RAUs, a plurality of neural networks with the same structure can be trained respectively to be responsible for power distribution work of different maximum transmission powers of the RAUs respectively. In this embodiment, DNN1, DNN2, \8230andDNNn are used for these neural networks having the same structure. In addition, for convenience of representation, the maximum transmission power of different RAUs is represented as:
then, the same channel state information H is used to perform calculation through a conventional sub-gradient algorithm, and different optimal power allocation schemes may be obtained under different maximum transmit powers, which may be respectively expressed as: { p (Pmax1) ,p (Pmax2) ,...,p (Pmaxn) }. The optimal power distribution schemes can be used as label power when different neural networks are trained respectively.
And S50, stopping training the model when the training parameters of the trained model meet preset conditions, and storing the model which stops training as a power distribution model, wherein after the trained model outputs the distribution power of the channel state information, judging whether the training parameters of the trained model meet the preset conditions or not.
The training parameters of the trained model meet the preset conditions and can be characterized in that the model training is finished, namely the power output by the model is closer to the label power, and the training parameters comprise the convergence value of the neural network, the mean square error between the distribution power output by the trained model and the label power of the channel state information corresponding to the distribution power or the training times. The preset conditions include that the convergence value of the network of the trained model is smaller than a preset convergence value; the mean square error between the distributed power output by the trained model and the label power of the channel state information corresponding to the distributed power is smaller than a preset mean square error; or the training times of the trained model reach preset times. In the training scenario with 5 remote access units and 3 users, the preset number of times may be 512, i.e. there are 512 channel state information in the data set.
After the training of the trained model is stopped, the model of which the training is stopped can be saved as a power distribution model. The power distribution model and the power distribution algorithm are adopted to carry out tests under the same scene and channel state information, and the following test results are obtained, and are specifically shown in table-1, table-2, table-3 and table-4:
TABLE-1
TABLE-2
TABLE-3
TABLE-4
The method comprises the steps of obtaining channel state information, DNN, sub-gradient, maxSE and maxEE, wherein PP is channel state information, sub-gradient is a neural network in a power distribution model, sub-gradient is a Sub-step algorithm in a power distribution algorithm, maxSE is maximum spectrum efficiency, and maxEE is maximum energy efficiency.
The scenarios of Table-1 and Table-3 are the same, and the scenarios of Table-2 and Table-4 are the same. As can be seen from tables-1 and-2, the target power of the ue output by the power allocation model is very close to the power obtained by the sub-step algorithm. As can be seen from the data in Table-1 and Table-2, the performance of the power distribution model under maxSE and maxEE can reach more than 91% of the performance of the conventional sub-gradient algorithm.
Tables-3 and-4 are experimental data of the time when the power distribution model outputs power and the time when the power is obtained by the sub-step algorithm. Based on the data, the calculation time of the distribution model output power is saved by at least 470 times compared with the calculation time of the sub-gradient algorithm, and can be saved by more than 200 times and ten thousand times at most. In order to better understand the difference between the distribution model algorithm and the sub-step algorithm, the time complexity of the two algorithms is analyzed, the time complexity of a trained fully-connected neural network is O (n), and when the objective function is the maximization system SE, the time complexity of the sub-gradient power distribution algorithm is O (n) 3 ) (ii) a And under the condition that the goal is to maximize the EE of the system, the fractional programming is firstly used for converting the nonlinear non-convex optimization problem into the convex function optimization problem, then the sub-gradient algorithm is used for obtaining the optimal solution of the system, and the total process time complexity of the sub-gradient algorithm is O (n) 4 )。
In conclusion, the power distribution model based on deep learning improves the online operation speed of the system at least in three levels and greatly reduces the time complexity of the algorithm under the condition of ensuring that the performance reaches more than 91% of the performance of the traditional sub-gradient algorithm.
In the technical solution provided in this embodiment, the plurality of pieces of channel state information in the data set and the label power corresponding to each piece of channel state information are sequentially output to the model to be trained to train the model to be trained, and when the training parameters of the trained model satisfy the preset conditions, the training of the trained model is stopped, so that the model from which the training is stopped is stored as the power distribution model. The label power is obtained by a power distribution algorithm, so that a power distribution model for outputting accurate power can be obtained by training, and the remote access unit can reasonably distribute power for the user terminal.
Referring to fig. 6, fig. 6 is a third embodiment of the power allocation method of the present invention, and based on the second embodiment, before the step S40, the method further includes:
and S60, setting loss functions of the network in the preset model to obtain the model to be trained, wherein the loss functions comprise a first part of loss functions and a second part of loss functions.
In this embodiment, the training of the model can be divided into two parts, one part is a feedforward algorithm, and the other part is a back propagation operation. The feed-forward operation is to calculate a loss value from the input of the neural network to the output of the neural network and then calculate a loss value according to a loss function. The back propagation operation is to continuously adjust the parameters of the neural network according to the loss value, so that the loss value of the neural network is smaller and smaller. The entire loss function contains two parts: the mean square error loss between the output of the neural network and the tag and the loss of the constraints. Therefore, it is necessary to set a loss function of the neural network, where the loss function includes a first partial loss function and a second partial loss function, that is, the first partial loss function and the second partial loss function are set for the loss function of the network in the model to obtain the model to be trained.
The loss function plays an important role in the training of the neural network, and the design of the loss function can be expressed as
Where M represents the batch size at the time of training, p n,k Representing the optimal power allocation calculated by the sub-gradient algorithm,
representing the output of the neural network, loss mse Is a neural network output
And tag p n,k Mean square error between, loss limit Is to ensure that the neural network can strictly adhere to the maximum SE problemThe inner constraints and the constraints within the maximize EE problem,
maximum transmit power of the remote access unit.
Scale factor lambda 1 And λ 2 Are respectively loss mse (first partial loss function) and loss limit (second partial loss function) loss fading coefficient for maintaining loss mse And loss limit Balancing to ensure that the effect of neural network training is good enough. That is, when setting the first partial loss function and the second partial loss function, it is necessary to set the loss fading coefficients corresponding to the first partial loss function and the second partial loss function. As can be seen from the above design formula of the loss function, the maximum transmit power of the remote access unit needs to be set for the model. It can be understood that, after the first partial loss function and the second partial loss function of the network in the preset model are set, the loss fading coefficients of the first partial loss function and the second partial loss function are set, and the maximum transmission power of the remote access unit is input into the model to obtain the model to be trained.
In addition, the second fractional loss function is determined based on a constraint comprising a constraint corresponding to a maximum spectral efficiency of the remote access unit or a constraint corresponding to a maximum energy efficiency of the remote access unit.
The limiting conditions for maximizing the spectral efficiency are as follows:
the constraint conditions for maximizing energy efficiency are as follows:
wherein P is d Representing constant circuit power consumption, P, per RAU c Representing a constant base power consumption, P o Representing the power consumed by the fiber transmission connecting the RAUs. τ denotes the power amplifier efficiency.
If it is not
Then the constraint is not satisfied then
Because the loss function of the neural network is gradually reduced in the training process if loss limit Greater than 0, then the training forces the neural network parameters to update so that loss limit Becoming smaller and smaller until less than 0. Conversely, if
Then
Therefore loss limit The training of the neural network is not affected. loss mse Is to ensure that the performance of the neural network is good enough, loss limit Are constraints of the system model that ensure that the output of the neural network is satisfied.
In the technical scheme provided by this embodiment, a model to be trained is obtained by setting a first partial loss function and a second partial loss function for the model, and thus the model to be trained is trained through feedforward operation and back propagation operation.
The present invention also provides a remote access unit, which includes a processor, a memory and a power allocation program stored in the memory and operable on the processor, wherein the remote access unit is provided with a mapping relationship between channel state information and power, and the power allocation program, when executed by the processor, implements the steps of the power allocation method according to the above embodiments.
The present invention further provides a distributed antenna system, where the distributed antenna system includes a plurality of remote access units, each remote access unit is provided with a mapping relationship between channel state information and power, and the remote access unit is configured to implement each step of the power allocation method according to the above embodiment.
The present invention also provides a medium storing a power allocation program that, when executed by a processor, implements the steps of the power allocation method according to the above embodiment.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of other like elements in a process, method, article, or system comprising the element.
The above-mentioned serial numbers of the embodiments of the present invention are only for description, and do not represent the advantages and disadvantages of the embodiments.
Through the description of the foregoing embodiments, it is clear to those skilled in the art that the method of the foregoing embodiments may be implemented by software plus a necessary general hardware platform, and certainly may also be implemented by hardware, but in many cases, the former is a better implementation. Based on such understanding, the technical solution of the present invention or portions thereof contributing to the exemplary technology may be embodied in the form of a software product, which is stored in a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) as described above and includes several instructions for enabling a terminal device (e.g., a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only an alternative embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (9)
1. A power allocation method, characterized in that the power allocation method comprises the steps of:
acquiring channel state information between a user terminal and a remote access unit;
acquiring target power corresponding to the channel state information according to a preset mapping relation between the channel state information and the power; and
changing the transmitting power of an antenna corresponding to the user terminal into the target power;
the mapping relation is achieved through a power distribution model, the power distribution model is obtained through training of a plurality of pieces of channel state information and label power corresponding to the channel state information, the label power is determined through power classification algorithm calculation according to power corresponding to the channel state information when energy efficiency is maximized, and when output power obtained through inputting the channel state information to the power distribution model is obtained, parameters of the power distribution model are adjusted according to a comparison result of the output power and the label power, so that an error of the output power and the power corresponding to the channel state information when the energy efficiency is maximized is within a preset range.
2. The power allocation method of claim 1, wherein said step of obtaining channel state information between the user terminal and the remote access unit is preceded by the step of:
sequentially inputting a plurality of pieces of channel state information in a data set and label power corresponding to each piece of channel state information into a model to be trained so as to train the model to be trained;
stopping training the model when the training parameters of the trained model meet the preset conditions, wherein after the trained model outputs the distributed power of the channel state information, judging whether the training parameters of the trained model meet the preset conditions or not;
and saving the model which stops training as a power distribution model.
3. The power allocation method according to claim 2, wherein the step of sequentially inputting the plurality of pieces of channel state information in the data set and the tag power corresponding to each piece of channel state information into the model to be trained so as to train the model to be trained comprises:
inputting the current channel state information into the model to be trained for training, and acquiring the distribution power output by the trained model;
acquiring tag power corresponding to current channel state information, and determining the mean square error between the distributed power and the tag power;
and judging whether the mean square error is less than or equal to a preset mean square error or not, wherein the training parameters of the trained model comprise the mean square error, and when the mean square error is less than or equal to the preset mean square error, judging that the trained model meets a preset condition.
4. The power allocation method according to claim 3, wherein said step of determining whether the mean square error is smaller than a predetermined mean square error further comprises:
when the mean square error is larger than a preset mean square error, adjusting parameters of a network in the trained model;
and taking the next channel state information in the data set as the current channel state information, and returning to the step of acquiring the distributed power output by the trained model.
5. The power distribution method according to any one of claims 2 to 4, wherein the training parameters include a convergence value of a neural network, a mean square error between a distribution power of a trained model output and a label power of channel state information corresponding to the distribution power, or a training number, and the preset condition includes:
the convergence value of the neural network is smaller than a preset convergence value;
the mean square error is less than a preset mean square error;
or the training times reach preset times.
6. The power allocation method according to claim 1, wherein the mapping relationship is obtained by learning, by a neural network, a plurality of pieces of channel state information and a tag power corresponding to each piece of channel state information, and the tag power is determined according to the channel state information.
7. A remote access unit comprising a processor, a memory, and a power allocation program stored in the memory and executable on the processor, the remote access unit being provided with a mapping between channel state information and power, the power allocation program when executed by the processor implementing the steps of the power allocation method according to any one of claims 1-6.
8. A distributed antenna system comprising a plurality of remote access units, each remote access unit having a mapping relationship between channel state information and power, the remote access units comprising a processor, a memory, and a power allocation program stored in the memory and operable on the processor, the power allocation program when executed by the processor implementing the steps of the power allocation method as claimed in any one of claims 1 to 6.
9. A computer-readable storage medium, storing a power distribution program which, when executed by a processor, performs the steps of the power distribution method of any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911323599.1A CN111162888B (en) | 2019-12-20 | 2019-12-20 | Distributed antenna system, remote access unit, power distribution method, and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911323599.1A CN111162888B (en) | 2019-12-20 | 2019-12-20 | Distributed antenna system, remote access unit, power distribution method, and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111162888A CN111162888A (en) | 2020-05-15 |
| CN111162888B true CN111162888B (en) | 2023-01-31 |
Family
ID=70557520
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911323599.1A Active CN111162888B (en) | 2019-12-20 | 2019-12-20 | Distributed antenna system, remote access unit, power distribution method, and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111162888B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP4162617A4 (en) * | 2020-06-08 | 2024-03-20 | Nokia Technologies Oy | Applying uplink channel information to determine data processing model deployed for downlink use |
| CN111970718B (en) * | 2020-07-22 | 2022-03-11 | 西北工业大学 | Deep learning-based power distribution method in energy collection untrusted relay network |
| CN114079939A (en) * | 2020-08-14 | 2022-02-22 | 中国移动通信有限公司研究院 | Method, device, equipment and storage medium for predicting base station transmitting power |
| CN114071516B (en) * | 2021-11-12 | 2024-02-06 | 三维通信股份有限公司 | Carrier configuration method, system, processing equipment and chip of distributed antenna system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109495139A (en) * | 2018-10-11 | 2019-03-19 | 深圳大学 | A kind of distributing antenna system and power distribution method |
| CN110401964B (en) * | 2019-08-06 | 2021-08-03 | 北京邮电大学 | Power control method based on deep learning for user-oriented center network |
-
2019
- 2019-12-20 CN CN201911323599.1A patent/CN111162888B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111162888A (en) | 2020-05-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111162888B (en) | Distributed antenna system, remote access unit, power distribution method, and medium | |
| Xu et al. | A deep reinforcement learning based framework for power-efficient resource allocation in cloud RANs | |
| Yoshida et al. | MAB-based client selection for federated learning with uncertain resources in mobile networks | |
| Dai et al. | Energy‐efficient resource allocation for device‐to‐device communication with WPT | |
| CN113365312B (en) | Mobile load balancing method combining reinforcement learning and supervised learning | |
| Shen et al. | Transfer learning for mixed-integer resource allocation problems in wireless networks | |
| Zhang et al. | A dynamic power allocation scheme in power-domain NOMA using actor-critic reinforcement learning | |
| CN114641076A (en) | Edge computing unloading method based on dynamic user satisfaction in ultra-dense network | |
| Shodamola et al. | A machine learning based framework for KPI maximization in emerging networks using mobility parameters | |
| Dong et al. | Dynamic cross-layer beamforming in hybrid powered communication systems with harvest-use-trade strategy | |
| US11570063B2 (en) | Quality of experience optimization system and method | |
| TWI756438B (en) | Electronic apparatus and method for wireless communications | |
| CN106921998A (en) | A kind of energy mix is powered the access control system and method for mobile communications network | |
| Alizadeh et al. | Power control with qos guarantees: A differentiable projection-based unsupervised learning framework | |
| Zhang et al. | Bayesian and multi-armed contextual meta-optimization for efficient wireless radio resource management | |
| WO2021120199A1 (en) | Distributed antenna system, remote access unit, power distribution method, and medium | |
| CN112752327A (en) | Power regulation method and access network equipment | |
| US20230345451A1 (en) | Multi-user scheduling method and system based on reinforcement learning for 5g iot system | |
| Zhi‐jin et al. | Spectrum handoff based on adaptive weights adjustment | |
| CN113660674B (en) | Method for intelligently realizing network construction and network optimization | |
| CN113395757A (en) | Deep reinforcement learning cognitive network power control method based on improved return function | |
| CN115665767A (en) | Wireless network operation situation regulation and control method, device, equipment and storage medium | |
| Hou et al. | Multicell power control under QoS requirements with CNet | |
| CN108901074A (en) | A kind of mobile subscriber's frequency spectrum distributing method based on cuckoo searching algorithm | |
| Iqbal et al. | Convolutional neural network-based deep Q-network (CNN-DQN) resource management in cloud radio access network |
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 |