CN112367129B - 5G reference signal received power prediction method based on geographic information - Google Patents

5G reference signal received power prediction method based on geographic information Download PDF

Info

Publication number
CN112367129B
CN112367129B CN202010729363.4A CN202010729363A CN112367129B CN 112367129 B CN112367129 B CN 112367129B CN 202010729363 A CN202010729363 A CN 202010729363A CN 112367129 B CN112367129 B CN 112367129B
Authority
CN
China
Prior art keywords
model
cell
height
neural network
transmitter
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010729363.4A
Other languages
Chinese (zh)
Other versions
CN112367129A (en
Inventor
徐小龙
徐浩严
赵娟
孙雁飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202010729363.4A priority Critical patent/CN112367129B/en
Priority to PCT/CN2020/119194 priority patent/WO2022021602A1/en
Publication of CN112367129A publication Critical patent/CN112367129A/en
Application granted granted Critical
Publication of CN112367129B publication Critical patent/CN112367129B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/309—Measuring or estimating channel quality parameters
    • H04B17/318—Received signal strength
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine 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
    • G06N3/04—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
    • H04B17/373—Predicting channel quality or other radio frequency [RF] parameters
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Electromagnetism (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • Biomedical Technology (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 invention discloses a method for predicting the received power of a 5G reference signal based on geographic information. Belongs to the technical field of artificial intelligence, and comprises the following specific steps: 1. constructing a feature map according to the map information; 2. an error iterative correction model is constructed by using artificial intelligence, and the constructed characteristic map and the actual signal receiving power are trained together to construct the error iterative correction model; the characteristic map takes the building position information of a cell, the height of a transmitter relative to the ground, a machine downward inclination angle, a vertical electrical downward inclination angle, the distance between a grid and the transmitter, the relative height between the grid and a signal line, the loss of a propagation path, a carrier frequency, a user antenna height correction item, the height of the cell transmitter relative to the ground, the transmitting power of the cell transmitter and the signal receiving altitude as the input for constructing the characteristic map, and finally takes the signal receiving power as the output.

Description

5G reference signal received power prediction method based on geographic information
Technical Field
The invention belongs to the technical field of artificial intelligence, and particularly relates to a method for predicting the received power of a 5G reference signal based on geographic information.
Background
With the development of 5G NR technology, the application of 5G in the world is continuously expanding; in the process of deploying a 5G network, an operator needs to reasonably select a base station site in a coverage area, and further, the communication requirement of a user is met by deploying a base station; in the whole wireless network planning process, efficient network estimation has very important significance for accurate 5G network deployment; the wireless propagation model just predicts the radio wave propagation characteristics in the target communication coverage area, so that the estimation of indexes such as cell coverage area, inter-cell network interference, communication rate and the like becomes possible; since the propagation environment of radio waves is complex and affected by various factors on the propagation path, such as plains, mountains, buildings, lakes, oceans, forests, the atmosphere, the curvature of the earth, etc., electromagnetic waves do not propagate in a single mode and path to generate complex transmission, diffraction, scattering, reflection, refraction, etc., it is a very difficult task to establish an accurate model.
The existing wireless propagation models can be distinguished according to research methods, and are generally divided into: an empirical model, a theoretical model, and an improved empirical model; the empirical model is obtained by obtaining a fixed fitting formula from empirical data, and typical models are Cost 231-Hata, Okumura and the like; the theoretical model is a Volcano model which is relatively representative by performing loss calculation according to the electromagnetic wave propagation theory and considering reflection, diffraction, refraction and the like of electromagnetic waves in space; the improved empirical Model is a computational Model, typically a Standard Performance Model (SPM), that can be provided for more detailed classification of scenes by introducing more parameters into the fitting equation.
In the actual propagation model modeling, in order to obtain a propagation model which accords with the actual environment of a target area, a large amount of extra measured data, engineering parameters and an electronic map are required to be collected for correcting the propagation model; in addition, the wireless LTE network is popularized all over the world, and billions of users all over the world generate a large amount of data every moment; how to properly utilize these data to assist wireless network construction becomes an important issue.
In recent years, big data driven AI machine learning technology has advanced significantly and has been used very successfully in the fields of language and image processing; with the development of parallel computing architecture, machine learning technology also has the capability of on-line computation, and its high real-time performance and low complexity make it possible to combine with wireless communication closely.
Disclosure of Invention
In order to solve the problems, the invention provides a method for predicting the received power of a 5G reference signal based on geographic information; the wireless propagation model is established by machine learning, and the model is used for accurately predicting the wireless signal coverage strength in a new environment, so that the network construction cost is greatly reduced, and the network construction efficiency is improved.
The technical scheme of the invention is as follows: the invention relates to a method for predicting the received power of a 5G reference signal based on geographic information, which is characterized by comprising the following steps: the method comprises the following steps:
step (1.1), constructing a feature map according to map information;
step (1.2), an error iterative correction model is constructed by using artificial intelligence, and the constructed characteristic map and the actual signal receiving power are trained together to construct the error iterative correction model;
in step (1.1), the specific construction process of the feature map includes the following steps:
(1.1.1), first, a height profile is constructed: the method comprises the steps of constructing the height characteristic of a transmitter relative to the ground, constructing the relative height characteristic of a grid and a signal line, constructing the height characteristic of a Cell transmitter relative to the ground, constructing the building height characteristic of a grid (Cell X, Cell Y) where a Cell site is located, constructing the altitude characteristic of the grid (Cell X, Cell Y) where the Cell site is located and the altitude characteristic of a grid (X, Y) where a signal receiving end is located;
(1.1.2) secondly, constructing a scene characteristic diagram: positioning the positions of the transmitter and the signal receiving end, specifically: the 5G signal is sent by the transmitter and finally received by the signal receiving end;
transmitting a 5G signal sent by a transmitter to a signal receiving end, and constructing a scene characteristic diagram through statistics of 20 terrains with different altitudes, buildings and scenes;
(1.1.3), and finally, constructing a signal characteristic diagram: acquiring the positions of a transmitter and a signal receiving end through the Cost 231-Hata characteristic, calculating the distance from the transmitter to the signal receiving end, and acquiring the 5G signal transmitting power of the transmitter; constructing a signal characteristic diagram by using the Cost 231-Hata characteristic, the distance from a transmitter to a signal receiving end and the 5G signal transmitting power of the transmitter;
in said step (1.1.2), said topography comprises: the number of grids of oceans, inland lakes, wetlands, suburb open areas, urban open areas, road open areas, vegetation areas, shrub vegetation, forest vegetation, urban super high-rise buildings (>60m), urban high-rise buildings (40m-60m), urban mid-high-rise buildings (20m-40m), urban <20m high-density building groups, urban <20m multi-story buildings, low-density industrial building areas, high-density industrial building areas, suburbs, developed suburb areas, rural areas, CBD business circle terrain;
in step (1.1.3), the Cost 231-Hata characteristic is calculated as follows:
Figure GDA0003318330810000021
wherein d represents the distance from the transmitter to the signal receiving end, and (Cell X, Cell Y) and (X, Y) represent the grid where the Cell site is located and the grid where the signal receiving end is located respectively;
PL＝46.3+33.9 log10 f-13.82 log10 hb-α+(44.9-6.55 log10 hue)log10 d+Cm (2)
where PL denotes propagation path loss, f denotes carrier frequency, and hbRepresenting the effective height of the base station antenna, alpha representing the correction term of the user antenna height, hueIndicating the effective height of the user's antenna, CmRepresents a scene correction constant;
wherein the relation of RSRP to PL: RSRP ═ Pt-PL (3)
RSRP represents the signal received power calculated by the Cost 231-Hata characteristic, PtRepresents the cell transmitter transmit power;
said CmThe scene characteristic graph is obtained through calculation of a convolutional neural network; h isbAnd hueThe system is obtained by calculating the relative height of the grids and the signal lines, the height of a Cell transmitter relative to the ground, the altitude of a grid (Cell X, Cell Y) where a Cell site is located, the altitude of a grid (X, Y) where a signal receiving end is located and the building height of the grid (Cell X, Cell Y) where the Cell site is located through a BP neural network;
the calculation method of the relative height of the grid and the signal wire comprises the following steps:
Δhv＝Height+Cell Altitude-tan(Electrial Downtilt+Mechanical Downtilt) (4)
in the formula (4), Δ hvHeight in Height + Cell availability-tan (electric downlink + Mechanical downlink) represents the Height of the Cell transmitter relative to the ground,
Δhvrespectively representing the Altitude of a grid (Cell X, Cell Y) where a Cell site is located, the Altitude of a grid (X, Y) where a signal receiving end is located, a vertical electrical Downtilt of a Cell transmitter and a vertical Mechanical Downtilt of the Cell transmitter; wherein each grid is a square of 5 meters by 5 meters;
in step (1.2), the specific steps of constructing the error iterative correction model are as follows:
(1.2.1) designing a neural network Model1(x)：pre＝Model1(x) (5)
Wherein x represents a neural network Model1(x) Input values, said input values comprising: a scene feature map, a height feature map and a signal feature map; pre-representation neural network Model1(x) An output of (d);
the neural network model comprises an error back propagation algorithm, a convolutional neural network, an activation layer, a pooling layer and a full-link layer; training with an adam optimizer using the mean square error as a loss function;
aiming at the scene feature map: convolution kernels with different sizes are used for convolution operation, firstly, 0 complementing operation is carried out on the boundary in the input layer, inner product operation is carried out on each convolution kernel in the convolution layer and the input sequence after 0 complementing from the head end of the sequence to the tail end of the sequence, the value of the output layer is obtained, and a new characteristic diagram is formed; then obtaining a scene correction constant through a full connection layer;
for the height profile: calculating the height characteristics by using an error back propagation algorithm; obtaining the effective height of a base station antenna and the effective height of a user antenna; obtaining the signal receiving power calculated by the Cost 231-Hata characteristic according to the formulas (1) to (3);
for a signal profile: RSRP, d, PtForming feature vectors with scene features, and performing convolution by using convolution kernels with different sizesOperation, inner product operation is carried out on each convolution kernel in the convolution layer and the input sequence of the boundary 0 complement in the input layer from the beginning of the sequence to the end of the sequence to obtain the value of the output layer and form a new characteristic diagram f1；
Will f is1Activated by a Relu activation function, which is shown as equation (6):
Ac＝max(0，f1) (6)
where Ac represents the matrix of active layer outputs, max pooling of Ac, ma represents the length of max pooling, and PoiResults representing maximum pooling:
Poi＝max({Aci，Aci+1…Aci+ma-2，Aci+ma-1}) (7)
wherein Ac isiFor the ith element in the matrix, the pooling result Po isiInputting the data into a full connection layer for classification, and mapping the distributed features to a sample mark space; the full-connection layer is formed by connecting each pooling layer into a one-dimensional vector, calculating neurons of the hidden layer, and finally connecting a neuron output; the calculation method of each neuron sum is shown as the formula (8):
pre＝∑i Poi·W (8)
finally; pre is Model1The result of the prediction, W represents a parameter in the network;
(1.2.2) training neural network Model1(W, b) it is desirable to obtain a minimized loss function:
Loss(Model1(W，b)，Label)＝(pre-Label)2 (9)
wherein, Label represents the Label of each data, W and b represent the neural network Model respectively1The parameters and bias of (1);
(1.2.3), calculating an error function:
Error(Model1(x)，Label)＝pre-Label (10)
wherein, Error represents the neural network Model after each data x is trained1(x) The error between pre obtained after mapping and the expected Label;
(1.2.4) designing a Model of neural network Model1(x) Neural network Model with same structure2(x) For correcting neural network Model1(x) Error generated in predicting reference signal power:
pre2＝Model2(x) (11)
therein, pre2Model for representing neural network Model1(x) Predicting an error in signal power generation;
(1.2.5) let Label2Training neural network Model (Model)2(x) It is desirable to obtain a minimized loss function:
Loss(Model2(x)，Label)＝(pre2-Label2)2 (12)
Label2model for representing neural network Model2(x) Predicted value, Error represents neural network Model1(x) Errors in predicting signal power;
(1.2.6) Model Using neural network Model2(x) The result-corrected neural network Model of (1)1(x) The error is calculated as follows:
Pre3＝Model1(x)+Model2(x) (13)；
in step (1.2.2), the Model of the trained neural network Model1(W, b) the specific steps desired to obtain the minimized loss function are as follows: for a given number of iterations, first computing a gradient vector for the input parameter vector W based on a loss function loss (W) found over the entire data set; the parameter W is then updated: subtracting the value of the gradient value multiplied by the learning rate from the parameter W, namely updating the parameter in the direction of the inverse gradient;
wherein the content of the first and second substances,
Figure GDA0003318330810000041
loss (W) represents the gradient descending direction of the parameter, i.e. the partial derivative of loss (W), and eta is the learning rate; label represents the actual value of the sample; when the iteration is completed, the updating of W and the establishment of the model are realized:
Loss(W)＝(Label-(Model1(W)+Model2(W)))2 (14)
Figure GDA0003318330810000051
the invention has the beneficial effects that: the method fully considers all geographical information propagated by the signals, analyzes the geographical information through map features constructed by the geographical information, finally predicts the signal receiving power by using an artificial intelligent model, finally reduces the mean square error to between 40 and 50dBm, and averagely reduces the error of each sample to between 6 and 7 dBm; while using the Cost 231-Hata variance will reach 346.79dBm, the average error per sample is between 18-19 dBm.
Drawings
FIG. 1(a) is a high thermal schematic of a building according to the present invention; 1(b) is an altitude thermodynamic diagram in the present invention; 1(c) ground feature type index map in the invention;
FIG. 2 is a diagram of transmitter parameters in the present invention;
FIG. 3 is a schematic representation of topographical features of the present invention;
FIG. 4 shows a Model according to the present invention1(x) A neural network model schematic diagram;
FIG. 5 is a diagram of the mean square error loss structure of the present invention;
fig. 6 is a graph of weak coverage recognition rate in the present invention.
Detailed Description
In order to more clearly illustrate the technical solution of the present invention, the following detailed description is made with reference to the accompanying drawings:
the main idea of the embodiment is as follows: firstly, constructing a height characteristic diagram, a scene characteristic diagram and a signal characteristic diagram; then, designing a neural network Model according to the characteristic diagrams1(x) And using the data set actually acquired by the satellite to Model the neural network1(x) Optimizing the weight in (1); finally, on the basis of the first model, using the two models with the same structure to correct errors generated by the first model;
specifically, as shown in the figure; A5G reference signal received power prediction method based on geographic information comprises the following steps:
step (1.1), acquiring real map information according to a satellite map, and constructing a feature map through the map information;
step (1.2), an error iterative correction model is constructed by using artificial intelligence, and the constructed characteristic map and the actual signal receiving power are trained together to construct the error iterative correction model; the function is as follows: iteratively correcting the error trained by the data by a model to predict the signal receiving power; the error iterative correction model takes the extracted characteristics of the geographic information of each signal source and the received signal as input and takes the power of the signal sent by the signal source received by the receiving party as output;
the actual signal received power is: in a practical scenario, the 5G signal transmitted by the transmitter receives the actual power at the signal receiving end.
Further, in the step (1.1),
the actually collected map information has 8 fields (as shown in a table I), and the corresponding meanings of the fields are shown in the table I; in consideration of the diversity and complexity of map types, actual ground objects such as urban areas, rural areas, lakes and the like are abstracted into numbers, the numbers are called ground object type name numbers (Clutter Index), and the actual ground object types corresponding to the ground object type name numbers can be seen in the second table;
table one, field meaning of map data:
Figure GDA0003318330810000061
table two, the numbering meaning of the surface feature type name:
Figure GDA0003318330810000062
the map data is also subjected to rasterization processing like the engineering parameter data, each grid represents an area of 5m multiplied by 5m, wherein (X, Y) records the coordinates of the upper left corner of the grid where the map is located;
the construction process of the feature map comprises the following specific steps:
(1.1.1), first, a height profile is constructed: the method comprises the steps of constructing the height characteristic of a transmitter relative to the ground, constructing the relative height characteristic of a grid and a signal line, constructing the height characteristic of a Cell transmitter relative to the ground, constructing the building height characteristic of a grid (Cell X, Cell Y) where a Cell site is located, constructing the altitude characteristic of the grid (Cell X, Cell Y) where the Cell site is located and the altitude characteristic of the upper part of the grid (X, Y) where a signal receiving end is located; the indexes of building height, altitude and ground object type are shown in fig. 1(a), (b), (c), respectively; the relevant parameters for the transmitter are shown in fig. 2;
(1.1.2) secondly, constructing a scene characteristic diagram: the method comprises the following steps of positioning the positions of a transmitter and a signal receiving end (such as a mobile phone, a computer, an intelligent terminal and other equipment with a 5G signal receiving function), and specifically operating the following steps: the 5G signal is sent by the transmitter, sent through a straight line and finally received by the signal receiving end;
transmitting a 5G signal sent by a transmitter to a signal receiving end, and constructing a scene characteristic diagram through statistics of 20 terrains with different altitudes, buildings and scenes; the terrain comprises: the number of grids of oceans, inland lakes, wetlands, suburb open areas, urban open areas, road open areas, vegetation areas, shrub vegetation, forest vegetation, urban super high-rise buildings (>60m), urban high-rise buildings (40m-60m), urban mid-high-rise buildings (20m-40m), urban <20m high-density building groups, urban <20m multi-story buildings, low-density industrial building areas, high-density industrial building areas, suburbs, developed suburb areas, rural areas, CBD business circle terrain; a specific scene feature map is shown in fig. 3;
by inequality
Figure GDA0003318330810000071
The terrain feature coordinates of five grids at each point on a straight line from the transmitter to the signal receiving end can be calculated; wherein (CellX, CellY), (X, Y) and(fx, fy) are the transmitter coordinates, the receiver coordinates and the terrain coordinates of the approach, respectively; counting the number of all (fx, fy) terrain grids to be the characteristic diagram of the constructed terrain;
(1.1.3), and finally, constructing a signal characteristic diagram: acquiring the positions of a transmitter and a signal receiving end through the Cost 231-Hata characteristic, calculating the distance from the transmitter to the signal receiving end, and acquiring the 5G signal transmitting power of the transmitter; constructing a signal characteristic diagram by using the Cost 231-Hata characteristic, the distance from a transmitter to a signal receiving end and the 5G signal transmitting power of the transmitter;
the method for calculating the Cost 231-Hata characteristic comprises the following specific steps:
Figure GDA0003318330810000072
wherein d represents the distance from the transmitter to the signal receiving end, (CellX, CellY) and (X, Y) represent the grid where the cell site is located and the grid where the signal receiving end is located, respectively;
PL＝46.3+33.9 log10 f-13.82 log10 hb-α+(44.9-6.55 log10 hue)log10 d+Cm (2)
wherein; PL denotes propagation path loss (dB), f denotes carrier frequency (MHz), hbRepresents the effective height (m), h) of the base station antennaueRepresenting the effective height (m) of the user antenna, alpha representing the correction term (dB) of the user antenna height, d representing the distance (km) from the transmitter to the signal receiving end, CmRepresents a scene correction constant (dB);
wherein, the relation between RSRP and PL is as follows: RSRP ═ Pt-PL (3)
RSRP represents the signal received power calculated by the Cost 231-Hata characteristic, PtRepresents cell transmitter transmit power (dBm);
said CmThe scene characteristic graph is obtained through calculation of a convolutional neural network; h isbAnd hueThe relative height of the grid and the signal line, the height of the Cell transmitter relative to the ground, and the sea of the grid (Cell X, Cell Y) where the Cell site is locatedThe pulling height, the altitude on the grids (X, Y) where the signal receiving end is located and the building height of the grids (Cell X, Cell Y) where the Cell site is located are obtained by calculation through a BP neural network;
the calculation method of the relative height of the grid and the signal wire comprises the following steps:
Δhv＝Height+Cell Altitude-tan(Electrial Downtilt+Mechanical Downtilt) (4)
in the formula (4), Δ hvHeight in Height + Cell availability-tan (electric downlink + Mechanical downlink) represents the Height of the Cell transmitter relative to the ground,
Δhvrespectively representing the Altitude of a grid (Cell X, Cell Y) where a Cell site is located, the Altitude of a grid (X, Y) where a signal receiving end is located, a vertical electrical Downtilt of a Cell transmitter and a vertical Mechanical Downtilt of the Cell transmitter; wherein each grid is a square of 5 meters by 5 meters;
in step (1.2), the specific steps of constructing the error iterative correction model are as follows:
(1.2.1) designing a neural network Model1(x)：pre＝Model1(x) (5)
The x represents a neural network Model1(x) The input values of (a), the input values comprising: a scene feature map, a height feature map and a signal feature map; pre-representation neural network Model1(x) An output of (d);
the neural network model comprises an error back propagation algorithm, a convolutional neural network, an activation layer, a pooling layer and a full-link layer; training with an adam optimizer using the mean square error as a loss function;
aiming at the scene feature map: convolution kernels with different sizes are used for convolution operation, firstly, 0 complementing operation is carried out on the boundary in the input layer, inner product operation is carried out on each convolution kernel in the convolution layer and the input sequence after 0 complementing from the head end of the sequence to the tail end of the sequence, the value of the output layer is obtained, and a new characteristic diagram is formed; then obtaining a scene correction constant through a full connection layer;
for the height profile: calculating the height characteristics by using an error back propagation algorithm; obtaining the effective height of a base station antenna and the effective height of a user antenna; obtaining the signal receiving power calculated by the Cost 231-Hata characteristic according to the formulas (1) to (3);
for a signal profile: RSRP, d, PtForming feature vectors with scene features, performing convolution operation by using convolution kernels with different sizes, performing inner product operation on each convolution kernel in the convolution layer and an input sequence with 0 being supplemented to the boundary in the input layer from the head end of the sequence to the tail end of the sequence to obtain the value of an output layer, and forming a new feature map f1；
Will f is1Activated by a Relu activation function, which is shown as equation (6):
Ac＝max(0，f1) (6)
ac represents a matrix of active layer outputs, max pooling of Ac is performed, where ma represents the length of max pooling, PoiResults representing maximum pooling:
Poi＝max({Aci，Aci+1...Aci+ma-2，Aci+ma-1}) (7)
wherein Ac isiFor the ith element in the matrix, the pooling result Po isiInputting the data into a full connection layer for classification, and mapping the distributed features to a sample mark space; the full-connection layer is formed by connecting each pooling layer into a one-dimensional vector, calculating neurons of the hidden layer, and finally connecting a neuron output; the calculation method of each neuron sum is shown as the formula (8):
pre＝∑i Poi·W (8)
finally; pre is Model1The result of the prediction, W represents a parameter in the network;
(1.2.2) training neural network Model1(W, b) it is desirable to obtain a minimized loss function:
Loss(Model1(W，b)，Label)＝(pre-Label)2 (9)
wherein, Label represents the Label of each data, W and b are respectively shown in the tableModel of neural network1The parameters and bias of (1);
(1.2.3), calculating an error function:
Error(Model1(x)，Label)＝pre-Label (10)
wherein, Error represents the neural network Model after each data x is trained1(x) The error between pre obtained after mapping and the expected Label;
(1.2.4) designing a Model of neural network Model1(x) Neural network Model with same structure2(x) For correcting neural network Model1(x) Error generated in predicting reference signal power:
pre2＝Model2(x) (11)
therein, pre2Model for representing neural network Model1(x) Predicting an error in signal power generation;
(1.2.5) let Label2Training neural network Model (Model)2(x) It is desirable to obtain a minimized loss function:
Loss(Model2(x)，Label)＝(pre2-Label2)2 (12)
Label2model for representing neural network Model2(x) Predicted value, Error represents neural network Model1(x) Errors in predicting signal power;
(1.2.6) Model Using neural network Model2(x) The result-corrected neural network Model of (1)1(x) The error is calculated as follows:
Pre3＝Model1(x)+Model2(x) (13)；
wherein, training neural network Model1(W, b) the specific steps desired to obtain the minimized loss function are as follows: for a given number of iterations, first computing a gradient vector for the input parameter vector W based on a loss function loss (W) found over the entire data set; the parameter W is then updated: the parameter W is subtracted by the gradient value times the learning rate, i.e. the parameter is updated in the anti-gradient direction；
Wherein the content of the first and second substances,
Figure GDA0003318330810000091
loss (W) represents the gradient descending direction of the parameter, i.e. the partial derivative of loss (W), and eta is the learning rate; label represents the actual value of the sample; when the iteration is finished, the updating of W and the establishment of the model are realized
Loss(W)＝(Label-(Model1(W)+Model2(W)))2 (14)
Figure GDA0003318330810000101
Example (b):
the invention is already applied to a 5G signal receiving power analysis system based on geographic information, and a user needs to select the position of a 5G base station and select the position of a signal receiving end. At the moment, the height characteristic can be constructed according to the building and elevation system of the map; then, the system can calculate according to the position of the 5G base station, the position of the receiving end and the information provided by the satellite map to obtain the topographic features that the signals need to pass from the transmitter to the signal receiving end; then, signal characteristics are constructed according to known base station parameters and map information; finally, these features are input into the Model that has been trained1(x) And a Model2(x) In (2), calculating Pre3 according to formula (13); at this time, Pre3 is the final output result of the system, i.e. the 5G signal power that the signal receiving end can finally receive for the position of the selected base station; the system has guiding function for the construction of the 5G base station, and for the positions needing to receive the 5G signals, the signal power which can be finally received by the 5G base stations at different positions is simulated and calculated, so that the trial and error cost for the construction of the base stations is reduced.
The system tests the real situation and measures the performance of the system by using two performance parameters to represent a neural network Model1(x) And a Model2(x) A specification of performance parameters of;
loss of mean square error: LOSS ═ (label-Y)2/n (16)
Wherein label represents the known signal received power in the training data set, and Y represents the final output result of the model; n represents the number of samples; the loss of the model is finally shown in fig. 5; smaller LOSS values indicate more accurate prediction; according to the method, all geographical information propagated by signals is fully considered, the geographical information is analyzed through map features constructed by the geographical information, finally, an artificial intelligent model is used for verifying a real acquisition data set, the signal receiving power is predicted, finally, the mean square error is reduced to 40-50dBm, and the average error of each sample is 6-7 dBm; while using Cost 231-Hata mean square error will reach 346.79dBm, averaging the error per sample between 18-19 dBm.
Weak coverage recognition rate:
table three definition of TP, FP, FN and TN:
Figure GDA0003318330810000102
the PCRR considers the Precision and Recall targets together, and its calculation formula is as follows:
Figure GDA0003318330810000103
where Precision can be understood as the probability that a grid whose prediction result is weak coverage is actually also weak coverage, it is defined as follows:
Figure GDA0003318330810000111
recall can be understood as the probability that the true result is how many grids of weak coverage are predicted to be weak coverage, which is defined as follows:
Figure GDA0003318330810000112
the weak coverage recognition rate of the final neural network model is shown in fig. 6; in general, the weak coverage recognition rate needs to be at least 20% to be used, and the weak coverage recognition rate of the invention reaches 35% in the present case.
Finally, it should be understood that the embodiments described herein are merely illustrative of the principles of embodiments of the present invention; other variations are possible within the scope of the invention; thus, by way of example, and not limitation, alternative configurations of embodiments of the invention may be considered consistent with the teachings of the present invention; accordingly, the embodiments of the invention are not limited to the embodiments explicitly described and depicted.

Claims (1)

1. A method for predicting received power of a 5G reference signal based on geographic information is characterized in that: the method comprises the following steps:
step (1.1), constructing a feature map according to map information;
step (1.2), an error iterative correction model is constructed by using artificial intelligence, and the constructed characteristic map and the actual signal receiving power are trained together to construct the error iterative correction model;
in step (1.1), the specific construction process of the feature map includes the following steps:
(1.1.1), first, a height profile is constructed: the method comprises the steps of constructing the height characteristic of a transmitter relative to the ground, constructing the relative height characteristic of a grid and a signal line, constructing the height characteristic of a Cell transmitter relative to the ground, constructing the building height characteristic of a grid (Cell X, Cell Y) where a Cell site is located, constructing the altitude characteristic of the grid (Cell X, Cell Y) where the Cell site is located and the altitude characteristic of a grid (X, Y) where a signal receiving end is located;
(1.1.2) secondly, constructing a scene characteristic diagram: positioning the positions of the transmitter and the signal receiving end, specifically: the 5G signal is sent by the transmitter and finally received by the signal receiving end;
transmitting a 5G signal sent by a transmitter to a signal receiving end, and constructing a scene characteristic diagram through statistics of 20 terrains with different altitudes, buildings and scenes;
(1.1.3), and finally, constructing a signal characteristic diagram: acquiring the positions of a transmitter and a signal receiving end through the Cost 231-Hata characteristic, calculating the distance from the transmitter to the signal receiving end, and acquiring the 5G signal transmitting power of the transmitter; constructing a signal characteristic diagram by using the Cost 231-Hata characteristic, the distance from a transmitter to a signal receiving end and the 5G signal transmitting power of the transmitter;
in said step (1.1.2), said topography comprises: the number of grids of oceans, inland lakes, wetlands, suburb open areas, urban open areas, road open areas, vegetation areas, shrub vegetation, forest vegetation, urban super high-rise buildings (>60m), urban high-rise buildings (40m-60m), urban mid-high-rise buildings (20m-40m), urban <20m high-density building groups, urban <20m multi-story buildings, low-density industrial building areas, high-density industrial building areas, suburbs, developed suburb areas, rural areas, CBD business circle terrain;
in step (1.1.3), the Cost 231-Hata characteristic is calculated as follows:
Figure FDA0003318330800000011
wherein d represents the distance from the transmitter to the signal receiving end, and (Cell X, Cell Y) and (X, Y) represent the grid where the Cell site is located and the grid where the signal receiving end is located respectively;
PL＝46.3+33.9log10f-13.82log10hb-α+(44.9-6.55log10hue)log10d+Cm (2)
where PL denotes propagation path loss, f denotes carrier frequency, and hbRepresenting the effective height of the base station antenna, alpha representing the correction term of the user antenna height, hueIndicating the effective height of the user's antenna, CmRepresents a scene correction constant;
wherein the relation of RSRP to PL: RSRP ═ Pt-PL (3)
RSRP represents the signal received power calculated by the Cost 231-Hata characteristic, PtRepresents the cell transmitter transmit power;
said CmThe scene characteristic graph is obtained through calculation of a convolutional neural network; h isbAnd hueThe system is obtained by calculating the relative height of the grids and the signal lines, the height of a Cell transmitter relative to the ground, the altitude of a grid (Cell X, Cell Y) where a Cell site is located, the altitude of a grid (X, Y) where a signal receiving end is located and the building height of the grid (Cell X, Cell Y) where the Cell site is located through a BP neural network;
the calculation method of the relative height of the grid and the signal wire comprises the following steps:
Figure FDA0003318330800000021
in the formula (4), Δ hvHeight in Height + Cell availability-tan (electric downlink + Mechanical downlink) represents the Height of the Cell transmitter relative to the ground,
Δhvrespectively representing the Altitude of a grid (Cell X, Cell Y) where a Cell site is located, the Altitude of a grid (X, Y) where a signal receiving end is located, a vertical electrical Downtilt of a Cell transmitter and a vertical Mechanical Downtilt of the Cell transmitter; wherein each grid is a square of 5 meters by 5 meters;
in step (1.2), the specific steps of constructing the error iterative correction model are as follows:
(1.2.1) designing a neural network Model1(x)：pre＝Model1(x) (5)
Wherein x represents a neural network Model1(x) Input values, said input values comprising: a scene feature map, a height feature map and a signal feature map; pre-representation neural network Model1(x) An output of (d);
the neural network model comprises an error back propagation algorithm, a convolutional neural network, an activation layer, a pooling layer and a full-link layer; training with an adam optimizer using the mean square error as a loss function;
aiming at the scene feature map: convolution kernels with different sizes are used for convolution operation, firstly, 0 complementing operation is carried out on the boundary in the input layer, inner product operation is carried out on each convolution kernel in the convolution layer and the input sequence after 0 complementing from the head end of the sequence to the tail end of the sequence, the value of the output layer is obtained, and a new characteristic diagram is formed; then obtaining a scene correction constant through a full connection layer;
for the height profile: calculating the height characteristics by using an error back propagation algorithm; obtaining the effective height of a base station antenna and the effective height of a user antenna; obtaining the signal receiving power calculated by the Cost 231-Hata characteristic according to the formulas (1) to (3);
for a signal profile: RSRP, d, PtForming feature vectors with scene features, performing convolution operation by using convolution kernels with different sizes, performing inner product operation on each convolution kernel in the convolution layer and an input sequence with 0 being supplemented to the boundary in the input layer from the head end of the sequence to the tail end of the sequence to obtain the value of an output layer, and forming a new feature map f1；
Will f is1Activated by a Relu activation function, which is shown as equation (6):
Ac＝max(0，f1) (6)
where Ac represents the matrix of active layer outputs, max pooling of Ac, ma represents the length of max pooling, and PoiResults representing maximum pooling:
Poi＝max({Aci，Aci+1...Aci+ma-2，Aci+ma-1}) (7)
wherein Ac isiFor the ith element in the matrix, the pooling result Po isiInputting the data into a full connection layer for classification, and mapping the distributed features to a sample mark space; the full-connection layer is formed by connecting each pooling layer into a one-dimensional vector, calculating neurons of the hidden layer, and finally connecting a neuron output; the calculation method of each neuron sum is shown as the formula (8):
pre＝∑iPoi·W (8)
finally; pre is Model1The result of the prediction, W represents a parameter in the network;
(1.2.2) training neural networksModel1(W, b) it is desirable to obtain a minimized loss function:
Loss(Model1(W，b)，Label)＝(pre-Label)2 (9)
wherein, Label represents the Label of each data, W and b represent the neural network Model respectively1The parameters and bias of (1);
(1.2.3), calculating an error function:
Error(Model1(x)，Label)＝pre-Label (10)
wherein, Error represents the neural network Model after each data x is trained1(x) The error between pre obtained after mapping and the expected Label;
(1.2.4) designing a Model of neural network Model1(x) Neural network Model with same structure2(x) For correcting neural network Model1(x) Error generated in predicting reference signal power:
pre2＝Model2(x) (11)
therein, pre2Model for representing neural network Model1(x) Predicting an error in signal power generation;
(1.2.5) let Label2Training neural network Model (Model)2(x) It is desirable to obtain a minimized loss function:
Loss(Model2(x)，Label)＝(pre2-Label2)2 (12)
Label2model for representing neural network Model2(x) Predicted value, Error represents neural network Model1(x) Errors in predicting signal power;
(1.2.6) Model Using neural network Model2(x) The result-corrected neural network Model of (1)1(x) The error is calculated as follows:
Pre3＝Model1(x)+Model2(x) (13)；
in step (1.2.2), the Model of the trained neural network Model1(W, b) the specific steps for which it is desired to obtain a minimized loss function are as follows: for a given number of iterations, first computing a gradient vector for the input parameter vector W based on a loss function loss (W) found over the entire data set; the parameter W is then updated: subtracting the value of the gradient value multiplied by the learning rate from the parameter W, namely updating the parameter in the direction of the inverse gradient;
wherein the content of the first and second substances,
Figure FDA0003318330800000031
represents the gradient descending direction of the parameter, namely the partial derivative of loss (W), and eta is the learning rate; label represents the actual value of the sample; when the iteration is completed, the updating of W and the establishment of the model are realized:
Loss(W)＝(Label-(Model1(W)+Model2(W)))2 (14)
Figure FDA0003318330800000041
CN202010729363.4A 2020-07-27 2020-07-27 5G reference signal received power prediction method based on geographic information Active CN112367129B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202010729363.4A CN112367129B (en) 2020-07-27 2020-07-27 5G reference signal received power prediction method based on geographic information
PCT/CN2020/119194 WO2022021602A1 (en) 2020-07-27 2020-09-30 5g reference signal received power prediction method based on geographic information

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010729363.4A CN112367129B (en) 2020-07-27 2020-07-27 5G reference signal received power prediction method based on geographic information

Publications (2)

Publication Number Publication Date
CN112367129A CN112367129A (en) 2021-02-12
CN112367129B true CN112367129B (en) 2022-02-11

Family

ID=74516890

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010729363.4A Active CN112367129B (en) 2020-07-27 2020-07-27 5G reference signal received power prediction method based on geographic information

Country Status (2)

Country Link
CN (1) CN112367129B (en)
WO (1) WO2022021602A1 (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113644996B (en) * 2021-10-13 2021-12-31 华中师范大学 Cell-level RSRP estimation method based on deep learning
CN114611388B (en) * 2022-03-02 2022-10-25 中国人民解放军战略支援部队航天工程大学 Wireless channel characteristic screening method based on artificial intelligence
CN114826459B (en) * 2022-03-08 2023-02-03 南京航空航天大学 Spectrum map accurate construction method based on cross-domain reasoning
CN115022889B (en) * 2022-05-30 2023-07-11 中国人民解放军空军工程大学 RSRP prediction model of 5G network and establishment method and application thereof
CN115296761B (en) * 2022-10-10 2022-12-02 香港中文大学（深圳） Channel prediction method based on electromagnetic propagation model

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111405585A (en) * 2020-03-19 2020-07-10 北京联合大学 Neighbor relation prediction method based on convolutional neural network

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100439674B1 (en) * 2001-09-27 2004-07-12 (주) 엘지텔레콤 Apparatus and method for failure prediction of radio port using neural network
CN108683463A (en) * 2017-12-15 2018-10-19 南京邮电大学 A kind of propagation properties analysis method and analysis system
CN108513306B (en) * 2018-06-12 2021-12-14 北京中网华通设计咨询有限公司 Network coverage prediction method based on test big data
CN109151866B (en) * 2018-11-27 2019-03-05 南京华苏科技有限公司 The method for realizing antenna azimuth correction based on user data
CA3126954A1 (en) * 2019-01-22 2020-07-30 Trellisware Technologies, Inc. Controlling data communication quality in software-defined heterogenous multi-hop ad hoc networks
CN109842887B (en) * 2019-03-06 2021-10-29 南京邮电大学 Resource mapping method for ultra-dense scene considering user mobility
CN110912627B (en) * 2019-12-24 2021-05-14 电子科技大学 Data-driven cell received power prediction method
CN111405586A (en) * 2020-03-19 2020-07-10 长沙航空职业技术学院 Wireless intelligent propagation method based on Cost231-Hata model

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111405585A (en) * 2020-03-19 2020-07-10 北京联合大学 Neighbor relation prediction method based on convolutional neural network

Also Published As

Publication number Publication date
WO2022021602A1 (en) 2022-02-03
CN112367129A (en) 2021-02-12

Similar Documents

Publication Publication Date Title
CN112367129B (en) 5G reference signal received power prediction method based on geographic information
Ates et al. Path loss exponent and shadowing factor prediction from satellite images using deep learning
EP2659697B1 (en) Method for the prediction of coverage areas of a cellular network
CN102404756B (en) Antenna parameter optimizing system based on mobile phone measurement report
Krijestorac et al. Spatial signal strength prediction using 3D maps and deep learning
CN114501530B (en) Method and device for determining antenna parameters based on deep reinforcement learning
Morocho-Cayamcela et al. Breaking wireless propagation environmental uncertainty with deep learning
JP6696859B2 (en) Quality estimation device and quality estimation method
CN105222787A (en) Based on the location fingerprint base construction method of matrix fill-in
CN113806947B (en) Offshore wind farm layout processing method, device and equipment
CN113644996B (en) Cell-level RSRP estimation method based on deep learning
Gu et al. Localization with incompletely paired data in complex wireless sensor network
Nagao et al. Geographical clustering of path loss modeling for wireless emulation in various environments
CN114513805A (en) Wireless modeling method and system
CN117241215A (en) Wireless sensor network distributed node cooperative positioning method based on graph neural network
CN115913424A (en) Spectrum map accurate construction method for generating confrontation network based on conditions
Inoue et al. On the building map for radio propagation prediction using machine learning
CN115942231A (en) RSS-based 5G outdoor positioning method
Pattanayak A genetically trained neural network for prediction of path loss in outdoor microcell
CN112929818A (en) Indoor positioning method based on Kalman filtering and pan-kriging interpolation
Xu et al. Dubhe: a deep-learning-based B5G coverage analysis method
Ebhota et al. Signal power loss prediction based on artificial neural networks in microcell environment
Tao et al. Feature engineering based intelligent wireless propagation model for RSRP prediction
Haoyan et al. Dubhe: a deep-learning-based B5G coverage analysis method
Hu et al. Study on Wireless Signal Propagation in Residential Outdoor Activity Area Based on Deep Learning

Legal Events

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