CN113055111A - Channel modeling method and system based on Bayesian optimization - Google Patents
Channel modeling method and system based on Bayesian optimization Download PDFInfo
- Publication number
- CN113055111A CN113055111A CN202110240013.6A CN202110240013A CN113055111A CN 113055111 A CN113055111 A CN 113055111A CN 202110240013 A CN202110240013 A CN 202110240013A CN 113055111 A CN113055111 A CN 113055111A
- Authority
- CN
- China
- Prior art keywords
- hyper
- data
- neural network
- network model
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000005457 optimization Methods 0.000 title claims abstract description 26
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 30
- 238000012360 testing method Methods 0.000 claims description 69
- 238000013527 convolutional neural network Methods 0.000 claims description 62
- 238000002790 cross-validation Methods 0.000 claims description 49
- 238000011156 evaluation Methods 0.000 claims description 39
- 238000012549 training Methods 0.000 claims description 36
- 238000010200 validation analysis Methods 0.000 claims description 30
- 238000004088 simulation Methods 0.000 claims description 21
- 238000005259 measurement Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 8
- 230000001186 cumulative effect Effects 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 2
- 238000012795 verification Methods 0.000 claims description 2
- 238000004891 communication Methods 0.000 abstract description 21
- 238000013528 artificial neural network Methods 0.000 abstract description 12
- 238000002474 experimental method Methods 0.000 abstract description 5
- 230000006870 function Effects 0.000 description 34
- 238000005516 engineering process Methods 0.000 description 10
- 230000008901 benefit Effects 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 6
- 238000010295 mobile communication Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000013179 statistical model Methods 0.000 description 3
- 230000003213 activating effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
- H04B17/3912—Simulation models, e.g. distribution of spectral power density or received signal strength indicator [RSSI] for a given geographic region
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
- H04B17/3913—Predictive models, e.g. based on neural network models
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a channel modeling method and a system based on Bayesian optimization, belonging to the technical field of wireless communication, wherein the modeling method uses a neural network of a back propagation algorithm as a channel modeling model, but is different from other artificial experiments for searching the hyperparameters of the model.
Description
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to a channel modeling method and system based on Bayesian optimization.
Background
Along with the popularization and the use of wireless communication and wired networks, terminals like mobile phones, automobile-mounted systems, smart homes and the like are deeply fused with the internet, the number of the terminals accessing the internet and the wireless data transmission flow are greatly increased, the 4 th generation mobile communication technology (4G) which is being used in a large scale cannot meet the rapid development of mobile communication, internet of vehicles and internet of things, and the 5 th generation mobile communication technology (5G) provides a solution for people. At present, the frequency band used by 5G commercial in China is the middle frequency band below 6GHz, but available and developed frequency band resources below 6GHz are almost used, and frequency spectrum resources below 6GHz cannot meet the requirements of modern data transmission, namely low time delay, high capacity and large connection. As such, the industry is beginning to research the work of higher frequency bands, wherein millimeter wave wireless communication is one of the core technologies to solve the problem, and is the trend of future 5G development.
The main scenes used by the 5G technology are very complex and mainly classified into four categories: continuous wide area coverage, high capacity of hot spots, low power consumption, large connection, low delay and high reliability, the former two use scenes are the main scenes of the current large-scale commercial 4 th generation mobile communication technology (4G), the latter two scenes are the application scenes newly developed by the 5 th generation mobile communication technology (5G), and the characteristic that the 4G technology cannot well support the internet data transmission is solved. Millimeter waves play an important role in these four main scenarios, and the performance thereof in information transmission is affected by the characteristics of the wireless channel. The 5G channel characteristics mainly have some characteristics such as time delay, path loss, arrival angle, etc., and the study of these channel characteristics in a 5G scenario is a necessary step in the present study of millimeter wave wireless communication. In 5G wireless communication technology, channel characteristics may exhibit different characteristics in a usage scenario, and in order to improve system performance in the 5G usage scenario, research on new characteristics of a channel is necessary. Millimeter wave wireless communication still has the characteristics of wavelength length, is favorable to the miniaturization of system transmitting terminal radio frequency hardware, and the wave beam of millimeter wave is more narrow simultaneously, and the directive property is stronger, and anti-interference ability is stronger than 4G technique in the information transmission process, effectively reduces wireless communication system's energy consumption.
In millimeter wave wireless communication, methods for channel simulation and modeling can be mainly classified into three categories: the first category is deterministic modeling methods. The method is characterized in that modeling is carried out under a specific propagation environment, and the method comprises two methods of ray tracing and impulse response; the second category is statistical modeling methods. The method is based on the statistical characteristics of environmental parameters for modeling, and comprises a geometric model based on scatterer distribution, a parameterized model and a modeling method based on spatial correlation. The third type is semi-deterministic modeling, which adopts the characteristics of the first two channel models and makes up the limitations of statistical models and the shortcomings of deterministic model complexity. At present, the propagation characteristic parameters of the wireless channel are learned by utilizing the powerful capability of machine learning, the corresponding relation between the channel parameters and the physical communication scene characteristics is established, the advantages of the traditional deterministic channel model and the random statistical channel model are considered, the complexity is low, the system can better conform to the actual environment, and most wireless channel models can be accurately calculated. The main modeling method mainly comprises a random geometric modeling method and a correlation matrix method.
Researchers have applied many machine learning algorithms to the field of wireless communications. Reference documents: LV L.A novel wireless channel model with multiplex feed-forward network [ C]The wireless communication network is modeled according to a feedforward neural network, a communication channel model based on a multilayer FNN is established, an improved BP algorithm and a modeling method of a simulation result are provided by utilizing the excellent learning characteristic of the neural network, existing problems and solving methods are analyzed, and meanwhile, the simulation result shows that the FNN model can well track the time-varying characteristic of a non-stationary wireless channel. Reference documents: BERHAD N J, IBNKAHLA M,F.Statistical analysis of a two-layer backpropagation algorithm used for modeling nonlinear memoryless channels:The single neuron case[J].IEEE transactions on signal processing,1997,45(3) 747-756. Bershad N J et al trained the network on the BP neural network using zero mean Gaussian data and studied the transient and convergence properties of the network. At the same time, the authors have studied the influence of neural network structure, weights, initial conditions and learning speed step size on the network mean square error. For example, researchers have used radial basis function neural networks (RBF-NN) and Relevance Vector Machines (RVM) to predict path loss and estimate angle of arrival for wireless transmissions, respectively. The following neural network approach is also included: generating a channel impulse response by constructing a cluster kernel by using an Artificial Neural Network (ANN); directly extracting channel impulse response from the measurement data by using a LASSO algorithm; extracting channel characteristics by using a Convolutional Neural Network (CNN) to identify different wireless channels; 11 independent CNN networks were used to learn and predict 11 channel statistical characteristic parameters, respectively. The 11 independent CNN networks share the same set of input features, and the output labels are respectively 11 channel parameters (such as arrival angle mean, delay mean, arrival angle spread, delay spread, and the like), so that the same set of input features are used for learning and predicting a plurality of channel parameters at the same time, and the corresponding relation between the plurality of channel parameters and a communication scene is established.
Statistical modeling relies primarily on measurements of the channel, and derives various important statistical properties of the channel from a large number of actual measurements. Due to the large growth of mobile services and mobile users, microcellular and picocellular systems find application in a wide variety of propagation environments. In these propagation environments, the size of the cell radius is reduced by a lot, so that the statistical similarity between the propagation environments disappears, and the statistical model cannot be applied in this case.
The deterministic modeling needs detailed information of a propagation environment, the accuracy of the information affects the accuracy of channel modeling, and a ray tracing technology and a time domain finite difference method which are mainstream methods for deterministic modeling have respective defects, for example, in the ray tracing technology, rays which cannot reach a receiving point are abandoned at the beginning due to the technical characteristics of a mirror image method, so that the method is very difficult to generate scatterers in a complex environment, can be only applied in a simple geometric environment, and has a narrow application range; the ray transmitting method needs a receiving ball to receive rays, the radius of the receiving ball needs to be accurately set, and the calculated result is influenced when the radius is too large or too small. Although the time-domain finite difference method has high accuracy, the method needs detailed details of a propagation environment, has a large unknown quantity, is a big problem on how to solve, and has a complex solving algorithm.
Semi-deterministic modeling is a modeling method between statistical modeling and deterministic modeling, and at present, channel modeling by using a machine learning algorithm is a trend of semi-deterministic modeling. Neural networks that employ back-propagation algorithms have found many applications in channel modeling as an important algorithm for machine learning. However, in the channel modeling of the neural network, there are many hyper-parameters, such as learning rate, activation function, the number of neurons in a certain layer, batch size, etc., and these hyper-parameters need to artificially set the value of each hyper-parameter. Although the hyper-parameters set by the engineers with great experience can lead the model to obtain better results, the method for setting the hyper-parameters often cannot obtain satisfactory results in the use process of the ordinary people and has no theoretical explanation. Meanwhile, the hyper-parameters set by engineers with abundant experience are only limited to a certain specific scene, and if the set hyper-parameters are separated from the scene, the set hyper-parameters can not be used any more, so that the method has the advantage of strong adaptability, and the hyper-parameters which accord with the scene can only be set again according to the characteristics of the scene.
Disclosure of Invention
Aiming at the defects or improvement requirements in the prior art, the invention provides a channel modeling method and system based on Bayesian optimization, aiming at establishing a channel model based on a convolutional neural network, learning and predicting 6 channel parameters by using 6 input features, establishing a corresponding relation between the channel parameters and a communication scene, searching for optimal hyper-parameters through a Bayesian optimization algorithm, and finally obtaining an optimal model corresponding to the optimal hyper-parameters.
To achieve the above object, according to an aspect of the present invention, there is provided a channel modeling method based on bayesian optimization, including:
building a simulation environment in whichArranging a plurality of receiving antennas and a plurality of transmitting antennas, forming a group of receiving and transmitting antennas by any receiving antenna and any transmitting antenna, obtaining three-dimensional coordinates of the receiving antennas and the transmitting antennas in each group of receiving and transmitting antennas as input parameters, and simulating each group of receiving and transmitting antennas to obtain channel parameters corresponding to each group of receiving and transmitting antennas, wherein the input parameters form a characteristic data set XdataA measurement data set Y consisting of channel parametersdata;
Feature data set XdataRandom division into training sets XtrainCross validation set XvalidationAnd test set XtestMeasuring data set YdataRandom division into training sets YtrainCross validation set YvalidationAnd test set Ytest;
Constructing a convolutional neural network model consisting of an input layer, a convolutional layer, a full-link layer and an output layer, determining a cost function and an evaluation index of the convolutional neural network model, randomly setting an initial hyper-parameter value of the convolutional neural network model, and setting a hyper-parameter search space, an iteration number N and a proxy model of a Bayesian optimization algorithm as a Gaussian process;
feature data set X of training settrainAnd metrology data set YtrainInputting the data into a convolutional neural network model, training the convolutional neural network model by using a back propagation algorithm, and simultaneously, carrying out cross validation on a feature data set X of a setvalidationAnd metrology data set YvalidationInputting the data into a convolutional neural network model, and calculating an evaluation index decision coefficient of a cross validation set and a corresponding hyper-parameter by using a Bayesian optimization algorithm;
performing circulation according to the iteration times N, determining the hyper-parameter of the next circulation according to an EI acquisition function after each circulation until the evaluation index decision coefficient and the corresponding hyper-parameter of the N-circulation cross validation sets are obtained, and selecting the corresponding hyper-parameter combination as the optimal hyper-parameter combination when the evaluation index decision coefficient of the cross validation set is minimum;
feature data set X of test settestAnd metrology data set YtestCorresponding to the optimum hyper-parametric combination obtained after input into the loopAnd in the optimal convolutional neural network model, calculating to obtain an evaluation index decision coefficient of the test set, and verifying the effect of the optimal convolutional neural network model.
In some alternative embodiments, the composition is prepared byDetermining a cost function J of a convolutional neural network modelmseFromDetermining an evaluation index decision coefficient R of a convolutional neural network model2Wherein m represents the number of data groups, yiThe measured actual value of the ith group of data is shown,indicating the measured predicted value of the i-th group of data,the mean value of the measured actual values is shown.
In some alternative embodiments, the composition is prepared byCalculating evaluation index decision coefficient R of cross validation set2 val。
In some alternative embodiments, the composition is prepared byBy adjusting the hyper-parametersSo that is atFinding optimal hyper-parameters inWherein,the optimal hyper-parameter is represented by,representing a hyper-parametric search space,a set of hyper-parameters is represented,representing the optimized objective function.
In some alternative embodiments, the composition is prepared byDetermining a hyper-parameter for a next cycle, wherein,is the acquisition function, v*Is thatThe current value of the optimum function is,is a function of the cumulative density of a standard normal distribution,andmean and standard deviation, respectively, D1:tIs an observed data set.
According to another aspect of the present invention, there is provided a channel modeling system based on bayesian optimization, comprising:
the simulation module is used for building a simulation environment, a plurality of receiving antennas and a plurality of transmitting antennas are arranged in the simulation environment, any receiving antenna and any transmitting antenna form a group of receiving and transmitting antennas, and the receiving and transmitting antennas in each group are obtainedThree-dimensional coordinates of the receiving antenna and the transmitting antenna are used as input parameters, and simulation is carried out on each group of receiving and transmitting antennas to obtain channel parameters corresponding to each group of receiving and transmitting antennas, wherein the input parameters form a characteristic data set XdataA measurement data set Y consisting of channel parametersdata;
A data set dividing module for dividing the characteristic data set XdataRandom division into training sets XtrainCross validation set XvalidationAnd test set XtestMeasuring data set YdataRandom division into training sets YtrainCross validation set YvalidationAnd test set Ytest;
The initialization module is used for constructing a convolutional neural network model composed of an input layer, a convolutional layer, a full-link layer and an output layer, determining a cost function and an evaluation index of the convolutional neural network model, randomly setting an initial hyper-parameter value of the convolutional neural network model, and setting a hyper-parameter search space, iteration times N and a proxy model of a Bayesian optimization algorithm as a Gaussian process;
a training module for converting the feature data set X of the training settrainAnd metrology data set YtrainInputting the data into a convolutional neural network model, training the convolutional neural network model by using a back propagation algorithm, and simultaneously, carrying out cross validation on a feature data set X of a setvalidationAnd metrology data set YvalidationInputting the data into a convolutional neural network model, and calculating an evaluation index decision coefficient of a cross validation set and a corresponding hyper-parameter by using a Bayesian optimization algorithm; performing circulation according to the iteration times N, determining the hyper-parameter of the next circulation according to an EI acquisition function after each circulation until the evaluation index decision coefficient and the corresponding hyper-parameter of the N-circulation cross validation sets are obtained, and selecting the corresponding hyper-parameter combination as the optimal hyper-parameter combination when the evaluation index decision coefficient of the cross validation set is minimum;
a verification module for testing the feature data set X of the test settestAnd metrology data set YtestInputting the data into an optimal convolutional neural network model corresponding to the optimal hyper-parameter combination obtained after circulation, and calculating to obtain an evaluation index decision system of the test setAnd verifying the effect of the optimal convolutional neural network model.
In some alternative embodiments, the composition is prepared byDetermining a cost function J of a convolutional neural network modelmseFromDetermining an evaluation index decision coefficient R of a convolutional neural network model2Wherein m represents the number of data groups, yiThe measured actual value of the ith group of data is shown,indicating the measured predicted value of the i-th group of data,the mean value of the measured actual values is shown.
In some alternative embodiments, the composition is prepared byCalculating evaluation index decision coefficient R of cross validation set2 val。
In some alternative embodiments, the composition is prepared byBy adjusting the hyper-parametersSo that is atFinding optimal hyper-parameters inWherein,the optimal hyper-parameter is represented by,representing a hyper-parametric search space,a set of hyper-parameters is represented,representing the optimized objective function.
In some alternative embodiments, the composition is prepared byDetermining a hyper-parameter for a next cycle, wherein,is the acquisition function, v*Is thatThe current value of the optimum function is,is a function of the cumulative density of a standard normal distribution,andmean and standard deviation, respectively, D1:tIs an observed data set.
According to another aspect of the invention, a computer-readable storage medium is provided, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any of the above.
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
1. the channel modeling method builds a required communication scene based on electromagnetic simulation software Wireless Insite. The Wireless suite (WI) is electromagnetic simulation software based on Ray Tracing, and can simulate electromagnetic propagation and Wireless communication in various complex environments such as cities, indoor areas and suburban areas to directly obtain channel characteristic parameters. Therefore, the channel modeling method does not depend on expensive channel measurement, and various important statistical characteristics of the channel do not need to be obtained from a large number of actual measurements. The statistical model can be applied to other propagation environments mainly by means of statistical similarity among the propagation environments, channel parameters of different propagation environments can be obtained by the modeling method without the statistical similarity among the propagation environments, and the channel modeling method is used in different propagation environments only by building corresponding communication scenes in the simulation module, so that the method has the advantages of universality and expandability.
2. Because the Wireless instruments can simulate electromagnetic propagation and Wireless communication in various environments such as cities, indoor areas and suburban areas, the modeling method can be applied to simple propagation environments and complex propagation environments, is not limited by propagation environment conditions, only needs to build a communication scene and set corresponding communication conditions, and software automatically calculates channel parameters according to a ray tracing method, so that the calculation complexity is low, and the calculation accuracy is high.
3. The modeling method uses a neural network of a back propagation algorithm as a model for channel modeling, but is different from other artificial experiments for searching the hyperparameters of the model, the Bayesian optimization algorithm is adopted for automatically searching the optimal hyperparameters, the defects that the time for searching the hyperparameters through the artificial experiments is long and the precision is low are overcome, meanwhile, the hyperparameters searched through the artificial experiments can only be suitable for a specific propagation environment, the hyperparameters are required to be searched through experiments again when the optimal hyperparameters are changed to another propagation environment, and the modeling method has the advantage of strong adaptability, can be applied to any propagation environment, and has the advantage of strong adaptability.
Drawings
Fig. 1 is a schematic flow chart of a channel modeling method based on bayesian optimization according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Fig. 1 is a schematic flowchart of a channel modeling method based on bayesian optimization according to an embodiment of the present invention, where the method shown in fig. 1 includes the following steps:
s1: building a simulation environment, arranging a plurality of receiving antennas and a plurality of transmitting antennas in the simulation environment, forming a group of receiving and transmitting antennas by any receiving antenna and any transmitting antenna, obtaining three-dimensional coordinates of the receiving antennas and the transmitting antennas in each group of receiving and transmitting antennas as input parameters, and simulating each group of receiving and transmitting antennas to obtain channel parameters corresponding to each group of receiving and transmitting antennas;
in the embodiment of the invention, Wireless InSite software can be used for building a simulation environment, and the frequency can be 100GHz for mainly simulating an outdoor scene according to the requirements of the invention. Through setting the Wireless inite software, for example, setting 300 transmit-receive antennas, 6 input parameters in total of three-dimensional coordinates of the transmit-receive antennas can be obtained, and 6 channel parameters can be output through simulation: the average horizontal arrival angle, the average vertical arrival angle, the average horizontal departure angle, the average vertical departure angle, the path loss, and the received power, and finally 90000 sets of data can be obtained, and the size of the data set is 90000 × 12.
S2: composing feature data sets X from input parametersdataA measurement data set Y consisting of channel parametersdataThe feature data set XdataRandom division into training sets XtrainCross validation set XvalidationAnd test set XtestMeasuring data set YdataRandom division into training sets YtrainCross validation set YvalidationAnd test set Ytest;
In the embodiment of the invention, 90000 groups of data containing 6 input parameters of three-dimensional coordinates of the transmitting and receiving antenna are called a characteristic data set XdataThe 90000 sets of data comprising 6 channel parameters are referred to as a measurement data set YdataRandomly dividing 90000 groups of data into a training set, a cross validation set and a test set, wherein the percentage of the training set, the cross validation set and the test set in 90000 groups of data can be 60%, 20% and 20%, and X istrain、XvalidationAnd XtestSequentially representing the feature data set in the training set, the feature data set in the cross validation set, and the feature data set in the test set, Ytrain、YvalidationAnd YtestThe measurement data sets in the training set, the cross validation set and the test set are sequentially represented.
S3: constructing a convolutional neural network model consisting of an input layer, a convolutional layer, a full-link layer and an output layer, determining a cost function and an evaluation index of the convolutional neural network model, randomly setting an initial hyper-parameter value of the convolutional neural network model, and setting a hyper-parameter search space, an iteration number N and a proxy model of a Bayesian optimization algorithm as a Gaussian process;
in the embodiment of the invention, the network structure of the convolutional neural network model is composed of an input layer, a convolutional layer, a fully-connected layer and an output layer, wherein the convolutional layer can have 3 layers, and the fully-connected layer can have 3 layers, so that the basic structure of the convolutional neural network model is determined, and the convolutional neural network model contains 5 hyper-parameters: there are learning rate, type of activation function, batch size, number of convolutional layer convolutional kernels per layer, and number of fully-connected layer neurons per layer.
In the embodiment of the invention, the cost function J of the convolutional neural network modelmse:
In formula (1): m represents the number of data groups;
yithe measured actual value of the ith group of data is represented;
Evaluation index determination coefficient R2:
In formula (2): m represents the number of data groups;
yithe measured actual value of the ith group of data is represented;
In the embodiment of the invention, the hyperparametric search space of the Bayesian optimization algorithm is setThe iteration number N and the proxy model are Gaussian processes, and in a Bayesian optimization algorithm, the hyper-parameters are usedSo that is atFinding optimal hyper-parameters inNamely:
S4: feature data set X of training settrainAnd metrology data set YtrainInputting into a convolutional neural network, and obtaining a feature data set XtrainAs input to the convolutional neural network, and a data set Y is measuredtrainAs the output of the network, training a convolutional neural network by using a back propagation algorithm; feature data set X of cross validation setvalidationAs input to the network, a metrology data set YvalidationThe output of the network is input into a convolutional neural network, and an evaluation index determination coefficient R of a cross validation set is calculated2 valNamely:
in formula (4): m represents the number of data sets, yiThe measured actual value of the ith group of data is shown,indicating the measured predicted value of the i-th group of data,the mean value of the measured actual values is shown.
S5: manually randomly setting a set of hyper-parametersConstructing an initial convolution network structure, circulating for N times according to the iteration times N, repeating the step S4 in each circulation, and determining the hyper-parameter of the next circulation according to the EI acquisition function, namely:
in formula (5):is the acquisition function, v*Is thatThe current value of the optimum function is,is a function of the cumulative density of a standard normal distribution,andmean and standard deviation, respectively, D1:tIs the observed data set and t represents time.
And obtaining evaluation index decision coefficients and corresponding hyper-parameters of the N-time circulation cross validation sets, and selecting a hyper-parameter combination with the minimum evaluation index decision coefficient of the cross validation sets as an optimal hyper-parameter combination.
S6: feature data set X of test settestAnd metrology data set YtestInput into the optimized hyper-parameter obtained after the loop according to step S5In the corresponding optimal model, calculating to obtain the evaluation index decision coefficient of the test setR2 valAnd verifying the effect of the optimal model.
The present invention is further illustrated by the following detailed description, without limiting the scope of the invention.
(1) Firstly, Wireless InSite software is used for building a simulation environment, the requirements of the invention are met, an outdoor scene is mainly simulated, and the frequency is 100 GHz. Through setting the Wireless InSite software, 300 transceiving antennas are set, 6 input parameters of the three-dimensional coordinates of the transceiving antennas can be obtained, and 6 channel parameters can be output through simulation: the average horizontal arrival angle, the average vertical arrival angle, the average horizontal departure angle, the average vertical departure angle, the path loss, and the received power, and finally 90000 groups of data (see table 1 for details) can be obtained, and the size of the data set is 90000 × 12.
190000 groups of data in Table
(2) The 90000 groups of data containing the three-dimensional coordinates of the transmitting and receiving antennas are referred to as a feature data set XdataAs shown in Table 2, 90000 sets of data containing 6 channel parameters are referred to as a measurement data set YdataAs shown in Table 3, 90000 groups of data are randomly divided into a training set, a cross validation set and a test set, wherein the training set, the cross validation set and the test set account for 60%, 20% and 20% of the 90000 groups of data in sequence, and X istrain、XvalidationAnd XtestSequentially representing the feature data set in the training set, the feature data set in the cross validation set, and the feature data set in the test set, Ytrain、YvalidationAnd YtestThe measurement data sets in the training set, the cross validation set and the test set are sequentially represented.
Table 2 characteristic data set Xdata
TABLE 3 metrology data set Ydata
X in Table 4trainX in Table 5validationAnd X in Table 6testSequentially represent the feature data sets in the training set, the cross-validation set, and the test set, e.g., Y in Table 7trainY in Table 8validationAnd Y in Table 9testThe measurement data sets in the training set, the cross validation set and the test set are sequentially represented.
TABLE 4 feature data set X in training settrain
Table 5 feature data set X in cross-validation setvalidation
TABLE 6 feature data set X in test settest
TABLE 7 measurement data set Y in training settrain
TABLE 8Metrology data set Y in cross-validation setvalidation
TABLE 9 measurement data set Y in test settest
(3) The network structure of the convolutional neural network model is composed of an input layer, a convolutional layer, a fully-connected layer and an output layer, wherein the convolutional layer has 3 layers, the fully-connected layer has 3 layers, so that the basic structure of the convolutional neural network is determined, and the convolutional neural network model contains 5 hyper-parameters: there are learning rate, type of activation function, batch size, number of convolutional layer convolutional kernels per layer, and number of fully-connected layer neurons per layer.
(4) Determining cost function and evaluation index of model, and cost function J of convolutional neural networkmse:Wherein m represents the number of data sets, yiThe measured actual value of the ith group of data is shown,the measured predicted value of the ith data is shown.
Evaluation index determination coefficient R2:Wherein m represents the number of data sets, yiThe measured actual value of the ith group of data is shown,indicating the measured predicted value of the i-th group of data,the mean value of the measured actual values is shown.
(5) Hyperparametric search space for setting Bayesian optimization algorithmAs shown in Table 10, the number of iterations 50 and the proxy model are Gaussian processes, and are optimized by the hyperparameters in the Bayesian optimization algorithmSo that is atFinding optimal hyper-parameters inNamely:wherein,the optimal hyper-parameter is represented by,representing a hyper-parametric search space,a set of hyper-parameters is represented,representing the optimized objective function.
Hyper-parameter | Learning rate | Activating a function | Batch size | Number of convolution kernels per convolution layer | Number of convolution kernels per fully-connected layer | Power |
Range | 55.15 | 91.79 | 235.15 | 90.55 | 0.00 | 115.06 |
(6) Feature data set X of training settrainAnd metrology data set YtrainInputting the data into a convolutional neural network, and training the convolutional neural network by using a back propagation algorithm; feature data set X of cross validation setvalidationAnd metrology data set YvalidationInputting the result into a convolutional neural network, and calculating an evaluation index decision coefficient R of a cross validation set2 valNamely:determining an evaluation index decision coefficient R of a convolutional neural network model2Wherein m represents the number of data groups, yiThe measured actual value of the ith group of data is shown,indicating the measured predicted value of the i-th group of data,the mean value of the measured actual values is shown.
(7) Manually randomly setting a set of hyper-parametersAnd (3) building an initial convolutional network structure, circulating for 50 times according to the iteration times, repeating the step (6) in each circulation, and determining the hyper-parameter of the next circulation according to the EI acquisition function, namely:
determining a hyper-parameter for a next cycle, wherein,is the acquisition function, v*Is thatThe current value of the optimum function is,is a function of the cumulative density of a standard normal distribution,andmean and standard deviation, respectively, D1:tIs an observed data set. The decision coefficients and corresponding hyperparameters of the 50-time cycle cross validation sets are obtained, and the hyperparameter combination with the minimum decision coefficient of the cross validation set is selected as the optimal hyperparameter combination, as shown in table 11.
(8) Feature data set X of test settestAnd metrology data set YtestInputting the optimal hyperparameter obtained after circulation according to the step (7)As shown in table 11, in the corresponding optimal model, the evaluation index determination coefficient R of the test set is calculated2 testAs shown in table 11, the effect of the optimal model was verified.
Hyper-parameter | Learning rate | Activating a function | Batch size | Number of convolution kernels per convolution layer | Number of convolution kernels per fully-connected layer |
0.006890958660007832 | relu | 1000 | 200,52,200 | 200,111,97 |
It should be noted that, according to the implementation requirement, each step/component described in the present application can be divided into more steps/components, and two or more steps/components or partial operations of the steps/components can be combined into new steps/components to achieve the purpose of the present invention.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A channel modeling method based on Bayesian optimization is characterized by comprising the following steps:
building a simulation environment, arranging a plurality of receiving antennas and a plurality of transmitting antennas in the simulation environment, forming a group of receiving and transmitting antennas by any receiving antenna and any transmitting antenna, obtaining three-dimensional coordinates of the receiving antennas and the transmitting antennas in each group of receiving and transmitting antennas as input parameters, and simulating each group of receiving and transmitting antennas to obtain channel parameters corresponding to each group of receiving and transmitting antennas, wherein the input parameters form a characteristic data set XdataA measurement data set Y consisting of channel parametersdata;
Feature data set XdataRandom division into training sets XtrainCross validation set XvalidationAnd test set XtestMeasuring data set YdataRandom division into training sets YtrainCross validation set YvalidationAnd test set Ytest;
Constructing a convolutional neural network model consisting of an input layer, a convolutional layer, a full-link layer and an output layer, determining a cost function and an evaluation index of the convolutional neural network model, randomly setting an initial hyper-parameter value of the convolutional neural network model, and setting a hyper-parameter search space, an iteration number N and a proxy model of a Bayesian optimization algorithm as a Gaussian process;
feature data set X of training settrainAnd metrology data set YtrainInputting the data into a convolutional neural network model, and training convolution by using a back propagation algorithmNeural network model, simultaneously cross-validating the feature data set X of the setvalidationAnd metrology data set YvalidationInputting the data into a convolutional neural network model, and calculating an evaluation index decision coefficient of a cross validation set and a corresponding hyper-parameter by using a Bayesian optimization algorithm;
performing circulation according to the iteration times N, determining the hyper-parameter of the next circulation according to an EI acquisition function after each circulation until the evaluation index decision coefficient and the corresponding hyper-parameter of the N-circulation cross validation sets are obtained, and selecting the corresponding hyper-parameter combination as the optimal hyper-parameter combination when the evaluation index decision coefficient of the cross validation set is minimum;
feature data set X of test settestAnd metrology data set YtestAnd inputting the test result into an optimal convolutional neural network model corresponding to the optimal hyper-parameter combination obtained after circulation, calculating to obtain an evaluation index decision coefficient of the test set, and verifying the effect of the optimal convolutional neural network model.
2. The method of claim 1, wherein the method is performed byDetermining a cost function J of a convolutional neural network modelmseFromDetermining an evaluation index decision coefficient R of a convolutional neural network model2Wherein m represents the number of data groups, yiThe measured actual value of the ith group of data is shown,indicating the measured predicted value of the i-th group of data,the mean value of the measured actual values is shown.
4. The method of claim 3, wherein the method is performed byBy adjusting the hyper-parametersSo that is atFinding optimal hyper-parameters inWherein,the optimal hyper-parameter is represented by,representing a hyper-parametric search space,a set of hyper-parameters is represented,representing the optimized objective function.
5. The method of claim 4, wherein the method is performed by Determining a hyper-parameter for a next cycle, wherein,is the acquisition function, v*Is thatThe current value of the optimum function is,is a function of the cumulative density of a standard normal distribution,andmean and standard deviation, respectively, D1:tIs an observed data set.
6. A channel modeling system based on Bayesian optimization, comprising:
the simulation module is used for building a simulation environment, a plurality of receiving antennas and a plurality of transmitting antennas are arranged in the simulation environment, any receiving antenna and any transmitting antenna form a group of receiving and transmitting antennas, three-dimensional coordinates of the receiving antennas and the transmitting antennas in each group of receiving and transmitting antennas are obtained and used as input parameters, simulation is carried out on each group of receiving and transmitting antennas to obtain channel parameters corresponding to each group of receiving and transmitting antennas, and the input parameters form a characteristic data set XdataA measurement data set Y consisting of channel parametersdata;
A data set dividing module for dividing the characteristic data set XdataRandom division into training sets XtrainCross validation set XvalidationAnd test set XtestMeasuring data set YdataRandom division into training sets YtrainCross validation set YvalidationAnd test set Ytest;
The initialization module is used for constructing a convolutional neural network model composed of an input layer, a convolutional layer, a full-link layer and an output layer, determining a cost function and an evaluation index of the convolutional neural network model, randomly setting an initial hyper-parameter value of the convolutional neural network model, and setting a hyper-parameter search space, iteration times N and a proxy model of a Bayesian optimization algorithm as a Gaussian process;
a training module for converting the feature data set X of the training settrainAnd metrology data set YtrainInputting the data into a convolutional neural network model, training the convolutional neural network model by using a back propagation algorithm, and simultaneously, carrying out cross validation on a feature data set X of a setvalidationAnd metrology data set YvalidationInputting the data into a convolutional neural network model, and calculating an evaluation index decision coefficient of a cross validation set and a corresponding hyper-parameter by using a Bayesian optimization algorithm; performing circulation according to the iteration times N, determining the hyper-parameter of the next circulation according to an EI acquisition function after each circulation until the evaluation index decision coefficient and the corresponding hyper-parameter of the N-circulation cross validation sets are obtained, and selecting the corresponding hyper-parameter combination as the optimal hyper-parameter combination when the evaluation index decision coefficient of the cross validation set is minimum;
a verification module for testing the feature data set X of the test settestAnd metrology data set YtestAnd inputting the test result into an optimal convolutional neural network model corresponding to the optimal hyper-parameter combination obtained after circulation, calculating to obtain an evaluation index decision coefficient of the test set, and verifying the effect of the optimal convolutional neural network model.
7. The system of claim 6, wherein the system is comprised ofDetermining a cost function J of a convolutional neural network modelmseFromDetermining convolution spiritDetermination of coefficient R by evaluation index of network model2Wherein m represents the number of data groups, yiThe measured actual value of the ith group of data is shown,indicating the measured predicted value of the i-th group of data,the mean value of the measured actual values is shown.
8. The system of claim 7, wherein the system is comprised ofAnd calculating an evaluation index decision coefficient of the cross validation set. ByBy adjusting the hyper-parametersSo that is atFinding optimal hyper-parameters inWherein,the optimal hyper-parameter is represented by,representing a hyper-parametric search space,a set of hyper-parameters is represented,representing the optimized objective function.
9. The system of claim 8, wherein the system is comprised of Determining a hyper-parameter for a next cycle, wherein,is the acquisition function, v*Is thatThe current value of the optimum function is,is a function of the cumulative density of a standard normal distribution,andmean and standard deviation, respectively, D1:tIs an observed data set.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110240013.6A CN113055111B (en) | 2021-03-04 | 2021-03-04 | Channel modeling method and system based on Bayesian optimization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110240013.6A CN113055111B (en) | 2021-03-04 | 2021-03-04 | Channel modeling method and system based on Bayesian optimization |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113055111A true CN113055111A (en) | 2021-06-29 |
CN113055111B CN113055111B (en) | 2023-03-24 |
Family
ID=76510040
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110240013.6A Active CN113055111B (en) | 2021-03-04 | 2021-03-04 | Channel modeling method and system based on Bayesian optimization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113055111B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116470975A (en) * | 2023-04-06 | 2023-07-21 | 天津大学 | Modeling method of SVR-based adaptive propagation prediction model |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2013136128A1 (en) * | 2012-03-15 | 2013-09-19 | Nokia Corporation | Generating radio channel models parameter values |
US20170374561A1 (en) * | 2016-06-22 | 2017-12-28 | Beijing University Of Posts And Telecommunications | Big-data-mining-based wireless channel modeling method |
CN108512621A (en) * | 2018-03-02 | 2018-09-07 | 东南大学 | A kind of Wireless Channel Modeling method based on neural network |
CN110113119A (en) * | 2019-04-26 | 2019-08-09 | 国家无线电监测中心 | A kind of Wireless Channel Modeling method based on intelligent algorithm |
CN110289927A (en) * | 2019-07-01 | 2019-09-27 | 上海大学 | The channel simulation implementation method of confrontation network is generated based on condition |
CN110598842A (en) * | 2019-07-17 | 2019-12-20 | 深圳大学 | Deep neural network hyper-parameter optimization method, electronic device and storage medium |
CN111245512A (en) * | 2020-01-15 | 2020-06-05 | 东南大学 | Neural network-based nonlinear channel modeling method for visible light communication system |
WO2021007812A1 (en) * | 2019-07-17 | 2021-01-21 | 深圳大学 | Deep neural network hyperparameter optimization method, electronic device and storage medium |
-
2021
- 2021-03-04 CN CN202110240013.6A patent/CN113055111B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2013136128A1 (en) * | 2012-03-15 | 2013-09-19 | Nokia Corporation | Generating radio channel models parameter values |
US20170374561A1 (en) * | 2016-06-22 | 2017-12-28 | Beijing University Of Posts And Telecommunications | Big-data-mining-based wireless channel modeling method |
CN108512621A (en) * | 2018-03-02 | 2018-09-07 | 东南大学 | A kind of Wireless Channel Modeling method based on neural network |
CN110113119A (en) * | 2019-04-26 | 2019-08-09 | 国家无线电监测中心 | A kind of Wireless Channel Modeling method based on intelligent algorithm |
CN110289927A (en) * | 2019-07-01 | 2019-09-27 | 上海大学 | The channel simulation implementation method of confrontation network is generated based on condition |
CN110598842A (en) * | 2019-07-17 | 2019-12-20 | 深圳大学 | Deep neural network hyper-parameter optimization method, electronic device and storage medium |
WO2021007812A1 (en) * | 2019-07-17 | 2021-01-21 | 深圳大学 | Deep neural network hyperparameter optimization method, electronic device and storage medium |
CN111245512A (en) * | 2020-01-15 | 2020-06-05 | 东南大学 | Neural network-based nonlinear channel modeling method for visible light communication system |
Non-Patent Citations (6)
Title |
---|
HONGJI HUANG等: "Deep Learning for Super-Resolution Channel Estimation and DOA Estimation Based Massive MIMO System", <IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY> * |
孙宁姚: "基于机器学习算法的时变无线信道建模研究", 《中国优秀硕士学位论文电子期刊》 * |
孙湘琛: "基于BP神经网络方法的28GHz毫米波信道建模研究", 《中国优秀硕士学位论文》 * |
赵雄文等: "基于最小二乘支持向量机的时变信道建模", 《北京邮电大学学报》 * |
邓帅: "基于改进贝叶斯优化算法的CNN 超参数优化方法", 《计算机应用研究》 * |
黄杰: "5G无线通信中的多频段毫米波信道测量与建模", 《中国优秀博士学位论文电子期刊》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116470975A (en) * | 2023-04-06 | 2023-07-21 | 天津大学 | Modeling method of SVR-based adaptive propagation prediction model |
Also Published As
Publication number | Publication date |
---|---|
CN113055111B (en) | 2023-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109217955B (en) | Wireless environment electromagnetic parameter fitting method based on machine learning | |
CN112748397B (en) | UWB positioning method based on self-adaptive BP neural network under non-line-of-sight condition | |
CN110213003B (en) | Wireless channel large-scale fading modeling method and device | |
CN113780436B (en) | Complex network key node identification method based on comprehensive degree | |
CN107798383B (en) | Improved positioning method of nuclear extreme learning machine | |
CN113177356B (en) | Target electromagnetic scattering characteristic rapid prediction method based on deep learning | |
CN110247719A (en) | The playback of 5G time varying channel and emulation mode based on machine learning | |
CN115642970A (en) | Self-learning channel modeling method and system | |
CN112949820A (en) | Cognitive anti-interference target detection method based on generation of countermeasure network | |
Saito et al. | Two-step path loss prediction by artificial neural network for wireless service area planning | |
Qiu et al. | Pseudo ray-tracing: Deep leaning assisted outdoor mm-wave path loss prediction | |
CN113055111B (en) | Channel modeling method and system based on Bayesian optimization | |
Wang et al. | Graph Neural Network enabled Propagation Graph Method for Channel Modeling | |
Fu et al. | A Joint-Neural-Network-Based Channel Prediction for Millimeter-Wave Mobile Communications | |
Peng et al. | Hmm-lstm for proactive traffic prediction in 6g wireless networks | |
Wu et al. | Research on RSS based indoor location method | |
CN115567131A (en) | 6G wireless channel characteristic extraction method based on dimensionality reduction complex convolution network | |
Chatelier et al. | Influence of dataset parameters on the performance of direct ue positioning via deep learning | |
US11288856B2 (en) | Method and system for wave propagation prediction | |
Sun et al. | Environment information-based channel prediction method assisted by graph neural network | |
Zhang et al. | An artificial intelligence radio propagation model based on geographical information | |
Zhang et al. | RETRACTED: Indoor 3-D localization based on simulated annealing bat algorithm | |
Yang et al. | Deep-learning-based bouncing-order prediction for propagation channel characterization using graph modeling | |
Deng et al. | A dynamic adaptive positioning method based on differential signal feature map | |
Nagao et al. | A study on path loss modeling using ResNet and pre-training with free space path loss |
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 |