CN113919483A - Method and system for constructing and positioning radio map in wireless communication network - Google Patents
Method and system for constructing and positioning radio map in wireless communication network Download PDFInfo
- Publication number
- CN113919483A CN113919483A CN202111114699.0A CN202111114699A CN113919483A CN 113919483 A CN113919483 A CN 113919483A CN 202111114699 A CN202111114699 A CN 202111114699A CN 113919483 A CN113919483 A CN 113919483A
- Authority
- CN
- China
- Prior art keywords
- training
- user
- edge
- neural network
- participating
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The application discloses a method and a system for constructing and positioning a radio map in a wireless communication network, wherein the method for constructing and positioning the radio map in the wireless communication network specifically comprises the following steps: performing collection of radio spectrum data; in response to completing the collection of radio-spectrum data, selecting an edge user to participate in training; responding to the edge user selected to participate in the training, and performing federal learning training to obtain a trained neural network model; outputting and saving the neural network model; and performing radio map positioning according to the stored neural network model. According to the method, the privacy of the user is protected, the high-precision radio map capable of meeting the requirements is accurately generalized by using the neural network, and the method is more suitable for intelligent scenes.
Description
Technical Field
The present application relates to the field of mobile communication networks, and in particular, to a method and system for constructing and positioning a radio map in a wireless communication network.
Background
With the rapid development of global economy, modern mobile communication devices are increasingly characterized by being ultra-dense and heterogeneous. The wireless communication network environment is also more and more complex, and effective representation of the wireless communication network environment is difficult to perform in a real physical environment, so that the coverage condition of the wireless network and the channel quality information of the wireless network in a real scene are difficult to obtain, which brings great challenges to effective management of the wireless communication network in future communication. Therefore, efficient and accurate construction of a Radio Map (Radio Map) of a known area has great significance for the development of future wireless communication networks.
But since constructing an accurate radio map requires a large amount of radio spectrum data and the method of acquisition is very limited. The existing method for solving data is to process the radio spectrum data of multiple data owners in a centralized way, and finally generalize a global radio map. However, once the data is shared, the data has a risk of leakage, which brings a great challenge to construct an accurate radio map under the premise of protecting the data security. In this context, Federal Learning (FL) helps to accurately train a global network model while protecting user data privacy. Federal learning ensures that data cannot go out of the local, training participants only exchange model parameters for local network training, and meanwhile, multiple data owning parties can be supported to participate in training, and the speed of model training is accelerated. Once the radio map is accurately constructed, the radio map can be directly applied to a real scene, wherein an important application is that a user carries out accurate positioning according to a known radio map and searches for a path with a good channel environment to complete a task.
Although federal learning provides an effective solution for building a global radiomap network model that protects data privacy and is nearly lossless. However, there are still many problems to be solved, mainly including:
1) in the process of federally learning and training a radio map model, model parameters need to be exchanged for many times, which occupies a large amount of communication resources. Due to the limited communication resource blocks, a reasonable allocation of communication resources is required to optimize the whole training process. 2) The local users have limited airborne energy, and when the radio map model is too long in training time and high in training precision requirement, part of the airborne energy of the local users cannot support the whole training process, so that when the radio map model is learned and trained by the federation, the airborne energy limitation of the users is an important factor influencing the training quality. Therefore, how to select the high-quality users to participate in the federal learning training reduces the overall training time, thereby saving the local energy of the users. 3) In the process of training a radio map model by federal learning, model transmission and local training in uplink and downlink communication processes bring communication delay problems, so that the time of the whole training process is increased, and model parameters are not updated timely.
Therefore, how to implement reasonable resource allocation, low energy consumption, high precision, low delay radio maps through federal learning is a problem which needs to be solved urgently in the field.
Disclosure of Invention
The invention aims to provide a mobile base station calculation and cache combined resource allocation method combining internet of things slicing service, which can balance the occupation of other physical resources on the premise that cache resources occupy the main position, can realize better performance, can perform balanced and flexible allocation on various physical resources on the basis, and has promotion significance for better realization of various strict communication requirements of mobile internet of things and better communication utility for users under the service-oriented characteristic.
In order to achieve the above object, the present invention provides a method for constructing and positioning a radio map in a wireless communication network, which specifically comprises the following steps: performing collection of radio spectrum data; in response to completing the collection of radio-spectrum data, selecting an edge user to participate in training; responding to the edge user selected to participate in the training, and performing federal learning training to obtain a trained neural network model; outputting and saving the neural network model; and performing radio map positioning according to the stored neural network model.
The method comprises the steps that the edge users participating in the overall training are locally subjected to radio spectrum information collection through system equipment, wherein the radio spectrum information collection mainly comprises position information and channel related information, and the signal-to-interference-and-noise ratio and large-scale channel gain information of corresponding geographic position information are mainly collected.
As above, wherein, the selecting the edge users participating in the training specifically includes the following sub-steps: initializing parameters; determining the size of the selected probability of all edge users participating in the training of the current round and all edge users not participating in the training of the current round in the next round according to the initialized parameters; and determining the edge users participating in the next round of training according to the selection probability of all the edge users participating in the current round of training and all the edge users not participating in the current round of training in the next round of training.
As above, wherein initializing parameters includes initializing a maximum number of user selections N, determining users/who are participating in training all the way through*And randomly selecting N-1 edge users from {1,2, L, L } users to participate in the training round.
As above, wherein, according to the initialized parameters, the determining the magnitude of the selected probability of each edge user specifically comprises the following sub-steps: training batch data according to the initialization parameters to obtain gradient transformation information before and after the edge user training participating in the training of the current round; the edge users participating in the current round of training send the obtained gradient change information before and after the edge users participating in the current round of training to the unmanned aerial vehicle aggregation terminal through the upper data transmission; the unmanned aerial vehicle aggregation end uses a neural network to predict the gradient change information of the user not participating in the current round of training by using the received gradient change information before and after training; and obtaining the magnitude of the selected probability of the next round of training of each edge user according to the gradient change information of the edge users not participating in the current round of training.
As above, the gradient change information before and after training includes gradient change information | | | e generated when the edge user l performs the μ th round of traininglμ| | and user l who participates in training in the whole course*Gradient change information of
As above, wherein each edge user next round of training is selected with the magnitude of the probability, namely:
wherein l*For the users who always participate in the training process, l' is the user who does not participate in the training of the current round, N is the maximum number of users allowed to participate in the training of the system, | | el'μ| l represents gradient change information of the edge user not participating in the training of the current round, and | elμAnd | | represents transformation information of the gradient before and after training.
The above, wherein the overall loss function in the federal learning processWherein f (w, x)li,yli) Is the loss function of the edge user l, where w is the model parameter of the neural network of the edge user l, xliInput vector, y, for edge user l dataliThe edge user l is an output vector of the neural network, and m is the size of each batch of training data of the edge user l.
As above, in the course of performing the federal learning training, the transmission rate is obtained;
the transmission rate is specifically expressed as:
wherein b islμBandwidth allocated for user l in the μ round training, hlμIs the channel gain, p, of the unmanned aerial vehicle and the edge user l during the mu round training in the transmission processlAllocating the transmitting power of uplink communication transmission of user I or the transmitting power of downlink communication of unmanned aerial vehicle to user I, N0Is the noise power spectral density.
A radio map construction and positioning system in a wireless communication network specifically comprises a collection unit, a selection unit, a training unit, an output unit and a positioning unit; wherein the collecting unit is used for collecting the radio frequency spectrum data; the selection unit is used for selecting users participating in training; the training unit is used for responding to the user who selects to participate in the training, and carrying out federal learning training to obtain a trained neural network model; the output unit is used for outputting and storing the neural network model; and the positioning unit is used for positioning the radio map according to the stored neural network model.
The application has the following beneficial effects:
(1) according to the invention, the building speed of the radio map is further accelerated by optimizing the user selection, resource allocation and transmission mechanism in the training process, the airborne energy of the edge user is saved, and the communication delay is reduced, so that the cost for building the radio map is reduced.
(2) According to the method, the privacy of the user is protected, the high-precision radio map capable of meeting the requirements is accurately generalized by using the neural network, and the method is more suitable for intelligent scenes.
(3) According to the method, the accuracy requirement of the given map and the resource condition of the edge users participating in training can be reasonably coordinated according to the intelligent radio map construction mode, and the construction of tasks in a real scene is achieved.
(4) In the application, the user can download the radio map from the APP end or the cloud end on line or off line to perform positioning, advance design of the task track and other applications, so that consumption of real resources is saved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art according to the drawings.
Fig. 1 is a flowchart of a method for constructing and positioning a radio map in a wireless communication network according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of a high-precision radio map construction and positioning system according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application are clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The invention provides a method and a system for constructing and positioning a radio map in a wireless communication network, which enable a plurality of edge data owners to construct an accurate radio map under the condition that data cannot be local, solve the problem of data isolated island and accelerate the construction speed of the radio map. And the construction cost is reduced, the energy consumption of edge users is saved, and the time delay in the communication process is smaller in the process of constructing the radio map. The end user can download the generated radio map online or offline to perform accurate positioning, thereby realizing various applications of the radio map in reality.
According to the method, firstly, on the premise that a plurality of edge user radio data owners cannot share data, a global radio map needs to be constructed jointly, and then accurate positioning is carried out based on the constructed high-precision radio map to complete respective tasks. The radio map is constructed based on federal learning, so that the data of edge users cannot be found out locally, the high-precision radio map can be constructed almost without damage, and the privacy problem of the data is protected. And also proposes algorithms to speed up the radio map construction and energy saving to optimize the radio map construction process.
Scene assumption is as follows: assuming that L edge users can participate in the training of federal learning in the process of generating the radio map through neural network training, due to the limitation of wireless resource blocks, at most L users with the number of N less than or equal to L can participate in the overall training task each time, and the precision of the radio map with given requirements is gamma. The location deployment matrix of the edge users is denoted by u.
As shown in fig. 1, a method for constructing and positioning a radio map in a wireless communication network provided by the present application specifically includes the following steps:
step S110: collection of radio spectrum data is performed.
Before the system is operated, each participant involved in training first collects radio spectrum information of the respective region.
Specifically, the edge users participating in the overall training collect radio spectrum information locally through system equipment, wherein the radio spectrum information mainly comprises position information and channel related information, and the signal-to-interference-and-noise ratio, large-scale channel gain and the like of corresponding geographical position information are collected.
Further, the received radio spectrum information R of a certain position is represented as R ═ { Q (x, y, H), Q }, where Q (x, y, H) in R is certain position information of the area where the edge user is located, x is an abscissa mapped to a horizontal plane, y is an ordinate mapped to the horizontal plane, H is a height of the position, and Q is radio channel information at the position, which may be a signal to interference plus noise ratio, a large-scale channel gain, a small-scale channel gain, or the like.
Step S120: in response to completing the collection of radio-spectrum data, a user participating in training is selected.
After the radio frequency spectrum information is collected, the federal learning aggregation processing end selects users to be trained by using a neural network. If the edge user can accelerate the construction of the radio map, the process of selecting the radio map to be added into the training is carried out for multiple times, otherwise, a small amount of selection is carried out on the user.
Specifically, the edge user stores the collected information in the server, and then the unmanned aerial vehicle aggregation end utilizes gradient change information | | | e generated when the edge user l performs the second round of traininglμThe user is reasonably selected to accelerate the whole training process, wherein | | | elμAnd | l is the change size of the gradient information before and after each round of training of the edge user.
Because the unmanned aerial vehicle needs to utilize gradient information before and after the training of the edge user | | | elμIf the unmanned aerial vehicle is in a training mode, the unmanned aerial vehicle can obtain gradient change information of all edge users. For ease of distinction, | | e is usedlμ| is gradient change information of users participating in the training of the current round, and | el'μAnd | | is gradient change information of the user who does not participate in the training of the current round. According to | | elμI and El'μThe size of | l, the size of the probability that each user is selected in the next round can be derived.
Wherein step S120 specifically includes the following substeps:
step S1201: and initializing parameters.
Wherein, initializing the maximum user selection number N, and determining the users l participating in the training in the whole course*And randomly selecting N-1 edge users from {1,2.. L } users to participate in the training round.
Step S1202: and determining the size of the selected probability of all the edge users participating in the training of the current round and all the edge users not participating in the training of the current round in the next round according to the initialized parameters.
The step S1202 specifically includes the following sub-steps:
step S12021: and training batch data according to the initialization parameters to obtain gradient transformation information before and after the edge user training participating in the training of the current round.
Wherein the gradient change information before and after training comprises gradient change information [ e ] generated by the marginal user l during the mu-th round of traininglμ| | and user l who participates in training in the whole course*Gradient change information of
Specifically, the selected local edge users perform batch data training according to the collected wireless spectrum information, and then gradient change information | e before and after the edge user l performs the μ -th round of training in the current round of training is obtainedlμ| | and user l who participates in training in the whole course*Gradient change information of
The gradient descent method is adopted by the edge user when the neural network model is trained locally, and it is assumed that the edge user trains J groups of data each time, and each group of data (x)j,yj) Comprising inputting xjAnd output yjAnd if the loss function is f (g) and the learning rate is lambda, the update formula of the model parameter alpha of the neural network from the mu training round to the mu +1 training round is expressed as follows:
wherein a isμ+1Representing the neural network model parameters of the [ mu ] +1 th round of training, aμRepresenting the neural network model parameters of the training of the μ round,representing the gradient of the edge user loss function.
By using the above formula, the gradient change information of the model parameter before and after the mu-th round of training of the edge user can be obtainedlμ| | and user l who participates in training in the whole course*Gradient change information of
Step S12022: and the edge users participating in the training of the current round send the obtained gradient change information before and after the training of the edge users participating in the training of the current round to the unmanned aerial vehicle polymerization end through the last data transmission.
Step S12023: the unmanned aerial vehicle polymerization end uses the neural network to utilize the received gradient change information of the edge users participating in the training of the current round | | elμI and user l participating in training in whole course*Gradient change information ofPredicting gradient change information E of the user not participating in the training of the current roundl'μ||。
Specifically, obtained by the above formulaWill be provided withInputting into the neural network via an input layer of the neural network, whereinData representing the neural network needing training comprises the number l of the ith user and the users l participating in training all the time*Gradient change information ofThe edge user's | | | e that will participate in the training of this roundlμAnd | l is trained as a label for neural network training.
After neural network training, any edge user l inputs a vectorBy passingThe input layer of the neural network inputs the neural network, and then the input vector can be learned by hiding the neurons in the layerThe nonlinear relation with the output vector o can output | | | e which does not participate in the training of the current round of usersl'μL. Below is given how the hidden layer learns the input vector through its neuronsNon-linear with the output vector o.
Wherein the state θ of the neuron in the hidden layer is:
wherein vinIs a weight matrix of the connection strength between the input vector and the neuron in the hidden layer,for users l who participate in federal learning training in the whole course*Weight matrix of bθAs a bias vector, a functionIs an activation function of the neural network, wherein exp (g) is an exponential function.
The state θ of a given neuron may result in an output vector o:
o=voutθ+bo
wherein v isoutA weight matrix of the connection strength of the output vector and the neuron in the hidden layer, boIs a bias vector.
Step S12024: according to gradient change information of the edge users not participating in the training of the current round | | | el'μI and gradient change information E of the edge users participating in the training of the current roundlμAnd | l, obtaining the size of the selected probability of the next round of training of each edge user.
For bookI e of edge user with wheel not participating in trainingl'μIf the unmanned aerial vehicle aggregation end predicts the gradient change information, the gradient change information is predicted by the unmanned aerial vehicle aggregation end through a neural network, and the gradient change information is used for accurately predicting the gradient change information of the user which does not participate in the training of the edge userl'μ| |, one user/is selected*The whole course participates in the training. Then according to | | elμI and El'μThe magnitude of | l, the magnitude of the probability that each edge user is selected in the next round of training can be calculated, that is:
wherein l*For the users who always participate in the training process, l' is the user who does not participate in the training of the current round, and N is the maximum number of users allowed to participate in the training of the system.
Step S1203: and determining the edge users participating in the next round of training according to the selection probability of all the edge users participating in the current round of training and all the edge users not participating in the current round of training in the next round of training.
Specifically, a probability interval for each edge user to be selected is set according to the probability of the edge user, and first, the probability of each edge user in the L edge users participating in the current round of training in the μ -th round of training is calculated through the formula in S12024, which is assumed to be P1μ、P2μ、……、PLμ. Then, the probability interval of these L edge users is (0, P)1μ)、(P1μ,P1μ+P2μ)、(P1μ+P2μ,P1μ+P2μ+P3μ)……、(P1μ+P2μ+K+PL-1μ,P1μ+P2μ+K+PLμ). And then generating a random number in an interval (0,1), and if the random number is in the probability interval of the edge user, selecting the edge user to participate in the (mu + 1) th round of training (namely the next round), or else, not participating in the (mu + 1) th round of training.
After determining the edge users participating in the next round of training, the method further comprises the step of allocating resources according to the determined edge users participating in the current round of training.
Specifically, according to the specific position of the edge user, the specific position of the unmanned aerial vehicle, the user selection condition, and the size of the total resource block (such as bandwidth), the technology such as convex optimization (CVX) and interior point method is used to reasonably allocate the resource block, thereby reducing the delay in the communication process.
Step S130: and responding to the edge user selected to participate in the training, and performing federal learning training to obtain a trained neural network model.
After determining that the users participating in the training and the resource allocation are complete, the edge users selected to participate in the current round of training use local radio spectrum data for federally learned training. The essence of the federal learning training is that the edge users only exchange model parameters between the federal learning aggregation end and the edge users to jointly train a neural network model without sharing local data.
Assuming that there are L distant and distant edge users in the edge node network participating in the federal learning training, each edge user has radio spectrum data of its own region, and the data of the respective regions do not intersect. Since the ground base station cannot completely cover the area where the edge user communicates, an Unmanned Aerial Vehicle (UAV) is used as the aggregation processing end of federal learning in this embodiment. The total loss function in the federal learning process is defined in this step as follows:
f(w,xli,yli) Is the loss function of the edge user l, where w is the model parameter of the neural network of the edge user l, xliInput vector, y, for edge user l dataliThe edge user l is an output vector of the neural network, and m is the size of each batch of training data of the edge user l.
Each of which has its own local neural network at the time of training.
Step S140 is performed by performing the following sub-steps of federal training so that the loss function value is less than a specified value. Wherein the substeps comprise:
step S1301: and initializing parameters.
In particular, the accuracy γ of the radio map, the global model parameter w, is initialized0And the local iterative training times Q of each round.
Step S1302: the edge users train the collected radio spectrum data.
Specifically, the edge user trains the collected radio spectrum data R ═ { Q (x, y, H), Q } through its local neural network, where the edge user trains the data with the location information Q (x, y, H) as the input of the local neural network and the channel information Q as the label trained by the local neural network.
And randomly selecting R groups of data each time by the edge users participating in training to perform iterative training of Q times of data.
Step S1303: updating of neural network model parameters is performed in response to completion of the training of the radio frequency spectrum data.
When the edge user trains locally, the gradient descent method is adopted to update the neural network model parameters in the neural network.
The edge user achieves the goal of reducing the loss function in S130 based on this formula until the loss function is reduced to a specified threshold.
WhereinAs a gradient of the edge user loss function, wlμFor the edge user l, the neural network model parameter in the training of the mu round, wl,μ+1For the model parameters of the edge user l in the μ +1 th round of training, k is the learning rate of the edge user l.
Step S1303: and uploading the updated neural network model parameter information by the edge user.
The edge user participating in the training will exercise the second roundModel parameter w oflμAnd transmitting the uplink communication to the unmanned aerial vehicle aggregation end.
And acquiring the transmission rate in the process of transmitting the model parameters. The transmission rate represents the speed of model parameter exchange between the unmanned aerial vehicle and the edge user, the larger the value of the transmission rate is, the faster the data transmission speed between the unmanned aerial vehicle and the edge user is, and the slower the data transmission speed is otherwise.
The smaller the value, the greater the delay in the transmission of the model parameters during communication, resulting in greater delay during communication, which increases the time to construct the radio map. Assuming that the size of the model parameter exchanged between the drone and the local user in each round is W bits, the transmission delay allowed by the system is t, and the transmission rate r islμNeeds to satisfy rlμt is more than or equal to W, otherwise, the transmission delay requirement of the system is not met. Wherein the transmission rate rlμThe concrete expression is as follows:
wherein b islμBandwidth allocated for user l in the μ round training, hlμIs the channel gain, p, of the unmanned aerial vehicle and the edge user l during the mu round training in the transmission processlAllocating the transmitting power of uplink communication transmission of user I or the transmitting power of downlink communication of unmanned aerial vehicle to user I, N0Is the noise power spectral density.
Step S1304: the unmanned aerial vehicle processes the collected model parameters through aggregation weighting processing to obtain an initial value w of the neural network model parameters of the next round of training of the edge userμ+1And communicate w via downlinkμ+1To the respective edge users.
Wherein wμ+1The concrete expression is as follows:
where M is the data size of all edge users participating in the training.
Wherein w is communicated downstreamμ+1The process of transmitting to each edge user also includes obtaining the transmission rate.
Step S1305: each edge user assigns the received neural network model parameters to a respective neural network.
Model parameter w to be obtained by edge userμ+1And assigning values to the local neural networks of the edge users, and repeating the steps S1302-S1304 until the neural networks reach the maximum iteration times Q or the neural network model reaches the specified precision.
Specifically, according to the set accuracy of the radio map, it is determined whether the trained neural network model meets the set requirement of the radio map accuracy, and if the MSE is less than or equal to γ (where γ is a given threshold and has a constant value), the trained neural network reaches the radio map with the generalized given accuracy, and then the federal learning training is completed, and step S140 is executed. Otherwise, the training will be continued and the steps S120-S130 will be repeated until the accuracy requirement of the given radio map is reached.
The MSE is calculated in the following way:
where S is the selected predicted number of radio spectrum data, y is the actual data of the radio spectrum,is the radio spectrum data predicted by the neural network.
Step S140: and outputting and saving the neural network model.
Specifically, the finally obtained neural network model of the radio map meeting the generalization requirements is stored to the APP end or the cloud end, and the user can download and apply the neural network model to various realistic scenes online or offline.
Step S150: and performing radio map positioning according to the stored neural network model.
The step S150 specifically includes the following sub-steps:
step S1501: the user downloads a radio map.
And the user downloads a neural network model capable of accurately performing the radio map from the APP end or the cloud end.
Step S1502: the user inputs the radio frequency spectrum values of the target to be located into the neural network model.
Step S1503: and the neural network outputs the accurate position of the positioning target according to the neural network model.
Step S1504: the user obtains position information of the positioning target.
Example two
The application provides a system for constructing and positioning a radio map in a wireless communication network, which specifically comprises a collecting unit 210, a selecting unit 220, a training unit 230, an output unit 240 and a positioning unit 250.
Wherein the collection unit 210 is used for collecting radio spectrum data.
The selection unit 220 is connected to the collection unit 210 for selecting users participating in the training.
The training unit 230 is connected to the selection unit 220, and is configured to perform federal learning training in response to a user who selects to participate in training, so as to obtain a trained neural network model.
The output unit 240 is connected to the training unit 230, and is configured to output the storage neural network model.
The positioning unit 250 is connected to the output unit 240 for performing a radio map positioning according to the stored neural network model.
The application has the following beneficial effects:
(1) according to the invention, the building speed of the radio map is further accelerated by optimizing the user selection, resource allocation and transmission mechanism in the training process, the airborne energy of the edge user is saved, and the communication delay is reduced, so that the cost for building the radio map is reduced.
(2) According to the method, the privacy of the user is protected, the high-precision radio map capable of meeting the requirements is accurately generalized by using the neural network, and the method is more suitable for intelligent scenes.
(3) According to the method, the accuracy requirement of the given map and the resource condition of the edge users participating in training can be reasonably coordinated according to the intelligent radio map construction mode, and the construction of tasks in a real scene is achieved.
(4) In the application, the user can download the radio map from the APP end or the cloud end on line or off line to perform positioning, advance design of the task track and other applications, so that consumption of real resources is saved.
Although the present application has been described with reference to examples, which are intended to be illustrative only and not to be limiting of the application, changes, additions and/or deletions may be made to the embodiments without departing from the scope of the application.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. A method for constructing and positioning a radio map in a wireless communication network is characterized by comprising the following steps:
performing collection of radio spectrum data;
in response to completing the collection of radio-spectrum data, selecting an edge user to participate in training;
responding to the edge user selected to participate in the training, and performing federal learning training to obtain a trained neural network model;
outputting and saving the neural network model;
and performing radio map positioning according to the stored neural network model.
2. The method as claimed in claim 1, wherein the edge users participating in the overall training collect the radio spectrum information locally through the system equipment, the radio spectrum information mainly comprises the information of the position information and the channel, and the information mainly comprises the signal-to-interference-and-noise ratio and the large-scale channel gain information of the corresponding geographical position information.
3. A method for constructing and positioning a radio map in a wireless communication network as claimed in claim 1, characterized in that the selection of the edge users involved in the training comprises in particular the sub-steps of:
initializing parameters;
determining the size of the selected probability of all edge users participating in the training of the current round and all edge users not participating in the training of the current round in the next round according to the initialized parameters;
and determining the edge users participating in the next round of training according to the selection probability of all the edge users participating in the current round of training and all the edge users not participating in the current round of training in the next round of training.
4. A method of constructing and positioning a radio map in a wireless communication network as claimed in claim 3, characterized in that the initialization parameters comprise initializing a maximum number N of user selections, determining the number of users i participating in the training all the time*And randomly selecting N-1 edge users from {1,2, L, L } users to participate in the training round.
5. A method for constructing and positioning a radio map in a wireless communication network as claimed in claim 4, characterized in that the determination of the size of the probability of each edge user being selected, based on initialized parameters, comprises in particular the sub-steps of:
training batch data according to the initialization parameters to obtain gradient transformation information before and after the edge user training participating in the training of the current round;
the edge users participating in the current round of training send the obtained gradient change information before and after the edge users participating in the current round of training to the unmanned aerial vehicle aggregation terminal through the upper data transmission;
the unmanned aerial vehicle aggregation end uses a neural network to predict the gradient change information of the user not participating in the current round of training by using the received gradient change information before and after training;
and obtaining the magnitude of the selected probability of the next round of training of each edge user according to the gradient change information of the edge users not participating in the current round of training.
6. The method as claimed in claim 5, wherein the gradient change information before and after training comprises gradient change information ie generated by the edge user l during the μ -th training roundlμ| | and user l who participates in training in the whole course*Gradient change information of
7. The method of claim 6, wherein each edge user trains the magnitude P of the probability of being selected in the next round of traininglμ+1The concrete expression is as follows:
wherein l*For the users who always participate in the training process, l' is the user who does not participate in the training of the current round, N is the maximum number of users allowed to participate in the training of the system, | | el'μ| l represents gradient change information of the edge user not participating in the training of the current round, and | elμAnd | | represents transformation information of the gradient before and after training.
8. The method of constructing and locating a radiomap in a wireless communication network of claim 7, wherein a total loss function in the federal learning procedure is definedWherein f (w),xli,yli) Is the loss function of the edge user l, where w is the model parameter of the neural network of the edge user l, xliInput vector, y, for edge user l dataliThe edge user l is an output vector of the neural network, and m is the size of each batch of training data of the edge user l.
9. The method of claim 8, wherein the federal learning training procedure comprises obtaining a transmission rate;
transmission rate rlμThe concrete expression is as follows:
wherein b islμBandwidth allocated for user l in the μ round training, hlμIs the channel gain, p, of the unmanned aerial vehicle and the edge user l during the mu round training in the transmission processlAllocating the transmitting power of uplink communication transmission of user I or the transmitting power of downlink communication of unmanned aerial vehicle to user I, N0Is the noise power spectral density.
10. A radio map construction and positioning system in a wireless communication network is characterized by specifically comprising a collection unit, a selection unit, a training unit, an output unit and a positioning unit;
wherein the collecting unit is used for collecting the radio frequency spectrum data;
the selection unit is used for selecting users participating in training;
the training unit is used for responding to the user who selects to participate in the training, and carrying out federal learning training to obtain a trained neural network model;
the output unit is used for outputting and storing the neural network model;
and the positioning unit is used for positioning the radio map according to the stored neural network model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111114699.0A CN113919483A (en) | 2021-09-23 | 2021-09-23 | Method and system for constructing and positioning radio map in wireless communication network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111114699.0A CN113919483A (en) | 2021-09-23 | 2021-09-23 | Method and system for constructing and positioning radio map in wireless communication network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113919483A true CN113919483A (en) | 2022-01-11 |
Family
ID=79235870
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111114699.0A Pending CN113919483A (en) | 2021-09-23 | 2021-09-23 | Method and system for constructing and positioning radio map in wireless communication network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113919483A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114520954A (en) * | 2022-01-21 | 2022-05-20 | 上海维智卓新信息科技有限公司 | Special equipment positioning method and system, special equipment and server |
CN114679231A (en) * | 2022-03-31 | 2022-06-28 | 中国人民解放军战略支援部队航天工程大学 | Method for acquiring space-based radio frequency map |
CN116405880A (en) * | 2023-05-31 | 2023-07-07 | 湖北国际贸易集团有限公司 | Radio map construction method and system based on federal learning |
-
2021
- 2021-09-23 CN CN202111114699.0A patent/CN113919483A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114520954A (en) * | 2022-01-21 | 2022-05-20 | 上海维智卓新信息科技有限公司 | Special equipment positioning method and system, special equipment and server |
CN114679231A (en) * | 2022-03-31 | 2022-06-28 | 中国人民解放军战略支援部队航天工程大学 | Method for acquiring space-based radio frequency map |
CN116405880A (en) * | 2023-05-31 | 2023-07-07 | 湖北国际贸易集团有限公司 | Radio map construction method and system based on federal learning |
CN116405880B (en) * | 2023-05-31 | 2023-09-12 | 湖北国际贸易集团有限公司 | Radio map construction method and system based on federal learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Liu et al. | Trajectory design and power control for multi-UAV assisted wireless networks: A machine learning approach | |
Bayerlein et al. | Trajectory optimization for autonomous flying base station via reinforcement learning | |
Zhang et al. | Energy-efficient trajectory optimization for UAV-assisted IoT networks | |
CN113919483A (en) | Method and system for constructing and positioning radio map in wireless communication network | |
Saxena et al. | Optimal UAV base station trajectories using flow-level models for reinforcement learning | |
Shi et al. | Drone-cell trajectory planning and resource allocation for highly mobile networks: A hierarchical DRL approach | |
CN111800828B (en) | Mobile edge computing resource allocation method for ultra-dense network | |
CN113359480B (en) | Multi-unmanned aerial vehicle and user cooperative communication optimization method based on MAPPO algorithm | |
CN111526592B (en) | Non-cooperative multi-agent power control method used in wireless interference channel | |
Qi et al. | Energy-efficient resource allocation for UAV-assisted vehicular networks with spectrum sharing | |
Wu et al. | 3D aerial base station position planning based on deep Q-network for capacity enhancement | |
Peng et al. | A predictive on-demand placement of UAV base stations using echo state network | |
CN114980169A (en) | Unmanned aerial vehicle auxiliary ground communication method based on combined optimization of track and phase | |
Chua et al. | Resource allocation for mobile metaverse with the Internet of Vehicles over 6G wireless communications: A deep reinforcement learning approach | |
Nasr-Azadani et al. | Single-and multiagent actor–critic for initial UAV’s deployment and 3-D trajectory design | |
Parvaresh et al. | A continuous actor–critic deep Q-learning-enabled deployment of UAV base stations: Toward 6G small cells in the skies of smart cities | |
CN114020024B (en) | Unmanned aerial vehicle path planning method based on Monte Carlo tree search | |
Gao et al. | Multi-armed bandits scheme for tasks offloading in MEC-enabled maritime communication networks | |
Nasr-Azadani et al. | Distillation and ordinary federated learning actor-critic algorithms in heterogeneous UAV-aided networks | |
Liu et al. | Opportunistic data ferrying in UAV-assisted D2D networks: A dynamic hierarchical game | |
CN116546559B (en) | Distributed multi-target space-ground combined track planning and unloading scheduling method and system | |
Liu et al. | Trajectory design for uav communications with no-fly zones by deep reinforcement learning | |
CN116781788A (en) | Service decision method and service decision device | |
CN116249202A (en) | Combined positioning and computing support method for Internet of things equipment | |
Bhandarkar et al. | User coverage maximization for a uav-mounted base station using reinforcement learning and greedy methods |
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 |