CN111122989B - K-clustering intelligent selection microwave signal multipath interference suppression method - Google Patents
K-clustering intelligent selection microwave signal multipath interference suppression method Download PDFInfo
- Publication number
- CN111122989B CN111122989B CN201911398258.0A CN201911398258A CN111122989B CN 111122989 B CN111122989 B CN 111122989B CN 201911398258 A CN201911398258 A CN 201911398258A CN 111122989 B CN111122989 B CN 111122989B
- Authority
- CN
- China
- Prior art keywords
- interference
- antenna
- model
- multipath interference
- multipath
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R29/00—Arrangements for measuring or indicating electric quantities not covered by groups G01R19/00 - G01R27/00
- G01R29/08—Measuring electromagnetic field characteristics
- G01R29/10—Radiation diagrams of antennas
- G01R29/105—Radiation diagrams of antennas using anechoic chambers; Chambers or open field sites used therefor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
Abstract
A K-clustering intelligent selection microwave signal multipath interference suppression method relates to an interference suppression method. The method comprises the following steps: 1) establishing and training a multipath interference suppression model; 2) setting the test condition of the antenna in a microwave darkroom; 3) adding reflecting plate interference sources of different materials or shapes to establish an interference model; 4) dividing the difficulty situation of the interference degree; 5) and correspondingly and respectively selecting proper algorithms according to the classification of the interference degree to finish the multi-path interference suppression of the microwave signals intelligently selected by the K clustering. When the multipath interference mode is simple, a polynomial regression algorithm is applied, a gradient descent method is adopted to train the model until the error is converged, the current optimal multipath interference suppression model is stored, the operation is simple, and the precision is improved. And in the multipath interference mode, the multipath interference suppression under the condition of three interference paths is realized by adopting a random forest. When the multipath interference mode is difficult, the neural network is adopted to realize multipath interference suppression under the condition of five interference paths, and finally the optimization efficiency of the antenna is improved.
Description
Technical Field
The invention relates to an interference suppression method, in particular to a K-clustering intelligent selection microwave signal multipath interference suppression method based on machine learning and a neural network, which can correspondingly and respectively select a proper algorithm according to the classification of interference degrees.
Background
The global informatization is closely related to information transmission when the mobile phone is close to the search of the universe and the mobile phone is close to the mobile phone of one hand. The popularization of information transmission draws academic attention to the efficiency and performance of information transmission. The antenna, as a key hardware for information transmission, plays a role in signal transmission and reception, and the requirements on the performance thereof are becoming stricter.
In order to ensure the performance of the antenna, the performance of the designed and used antenna meets the system requirements, and the precision of the antenna measurement system meets the system requirements. Since the design of the antenna can only be an approximation to a physical antenna, a high-precision measurement system is particularly important, and the high-precision measurement system also comprises high-performance error analysis and correction capability. Currently, antenna testing is used as a professional technology to supplement antenna design. By establishing a proper measuring system and assisting antenna principle knowledge, systematic and professional analysis is carried out on errors which can appear in antenna measurement, and the errors are predicted and eliminated by combining various existing software and hardware tools, so that the antenna far-field measurement precision is ensured. The antenna measurement technologies commonly used at present are classified into a far-field measurement technology, a near-field planar scanning measurement technology, a near-field cylindrical scanning measurement technology, and a near-field spherical scanning measurement technology. The analysis and suppression of possible errors in various antenna systems are continuously researched at home and abroad, and various peculiar and reasonable elimination mechanisms are also proposed based on different mechanisms and different starting angles. In general, the existing method for suppressing the multipath interference error of the antenna is complex, high in cost and problematic in precision.
Multipath interference suppression of antenna radiation patterns is a process of extracting a main signal by removing interference from other signal sources from a synthesized signal, and the interference of the type is very common in a microwave hybrid measurement darkroom. Because the near-field measurement antenna system is adopted, only the field on the antenna aperture surface needs to be measured, and a plurality of defects in the far-field measurement technology can be avoided. However, the equipment in the near-field measurement system is prone to cause multipath interference in the far-field. In recent years, artificial intelligence such as machine learning, neural networks and the like have become very powerful tools in various fields, and are widely applied to recognition and optimization algorithms. Although the traditional antenna field analysis method can accurately calculate the performance of the antenna, the method has the disadvantages of large computation amount, high memory occupation, concentration in a special antenna software, and difficulty in adding an optimization program. The method of artificial intelligence and the like is used in the antenna optimization program, although the efficiency and the precision of the antenna optimization can be obviously improved by using the characteristics of the method, the algorithm is single, and the requirement for the diversification of the algorithm cannot be met. The K clustering selective suppression method based on machine learning and the neural network is applied, so that the flexibility of algorithm selection is improved, the superiority of machine learning and the generalization capability of the neural network can be fully exerted, and the optimization efficiency of the antenna is finally improved.
Disclosure of Invention
The invention aims to provide a K-clustering intelligent selection microwave signal multipath interference suppression method which can select different algorithms according to different interference conditions, can fully play the superiority of machine learning and the generalization capability of a neural network and finally improve the antenna optimization efficiency.
The invention comprises several steps:
1) establishing and training a multipath interference suppression model;
2) setting the test condition of the antenna in a microwave darkroom;
3) adding reflecting plate interference sources of different materials or shapes to establish an interference model;
4) dividing the difficulty situation of the interference degree;
5) and correspondingly and respectively selecting proper algorithms according to the classification of the interference degree to finish the multi-path interference suppression of the microwave signals intelligently selected by the K clustering.
In step 1), the specific steps of establishing and training the multipath interference suppression model may be: taking a monopole antenna as an experimental antenna, taking an x axis on a plane where vertical waves are transmitted as sampling points, taking an original point as a starting point and 0.1mm as a moving unit to take 10 sampling points in the positive direction of the x axis, and respectively collecting data with interference and data without interference; training data by taking a synthesized signal of an interference signal and a direct signal as an input and taking a direct signal of the same sampling point as an output; the synthetic signal of any point on the matrix surface is used as input, the obtained output value is compared with the known direct signal at the point, the error sum of the output value and the known direct signal is further defined, the gradient of the error is solved, the weight and the offset are updated by adopting a gradient descent method, iteration is continuously carried out until the output value is infinitely close to the known direct signal, the training is finished, and the model is stored.
In step 2), the specific step of setting the test condition of the antenna in the microwave darkroom is as follows: characterizing multipath interference in a microwave mixing darkroom as a three-dimensional MPM process, and assuming a measuring system of an antenna far-field directional pattern as: an antenna AUT to be measured in a radiation field is used as a transmission antenna TX, a receiving antenna RX moves along a rectangular plane area, and measurement data are collected at each sampling point, wherein the rectangular plane area is taken from one part of a limited scanning surface of a plane near-field measurement system;
setting the transmission antenna to be positioned in a far field area of the receiving antenna so as to meet the far field measurement condition of the antenna; similarly, multipath signals from other signal sources or diffraction can be regarded as far-field signals, and are incident to a receiving area as approximate area plane waves; in the measuring process, the rectangular area is divided into L multiplied by M rectangular lattices to collect sampling signals, so that the effect of simple and convenient calculation is achieved.
In step 3), the interference source can adopt PEC, aluminum and copper reflecting plates, and the reflecting plates can adopt a plane, a concave surface and a convex surface; the reflecting plate is selected from different materials or shapes to be trained for multiple times, so that the model can generalize different interference sources. And finally, obtaining the antenna radiation value under the interference environment through simulation operation, and deriving the antenna radiation value under the interference environment from software for use.
In step 4), the difficulty situation of dividing the interference degree intelligently divides the multipath interference signals into three types of simple mode, medium mode and difficult mode according to the difference of the multipath interference degree, and the multipath interference signals respectively correspond to the multipath interference sources of the N reflecting plates (N is 1, 3 and 5), and the N reflecting plates under each material are subjected to comparative analysis of average errors, so that the diversity of the interference sources in the real communication process is simulated; and correspondingly and respectively selecting proper algorithms according to the classification of the interference degree to finish the multi-path interference suppression of the microwave signals intelligently selected by the K clustering.
In step 5), correspondingly and respectively selecting proper algorithms according to the classification of the interference degrees, wherein a simple mode corresponds to 1 reflecting plate, and a polynomial regression algorithm is selected; selecting a random forest algorithm according to the 3 reflectors in the medium mode; selecting a BP neural network algorithm when the difficult mode corresponds to 5 reflecting plates;
the polynomial regression algorithm may specifically include the following steps: when an interference source is a microwave mixing darkroom multipath interference signal generated under a simple mode of a reflecting plate (N is 1, materials can be PEC, aluminum and copper, and shapes can be plane, concave and convex), a multipath interference suppression method based on polynomial regression is adopted, simulation data are divided into a training set and a test set, the training set is used for carrying out data mining work on a machine learning algorithm, the learning rate is set to be 0.01, a loss function adopts a mean square error, so that a reliable characteristic is obtained to separate a standard signal and an interference signal, the test set is used for verifying the reliability of a model, cross verification is adopted to ensure that each sampling point is tested, and finally, an average error value is taken as a model evaluation index;
the random forest algorithm comprises the following specific steps: when the interference source is a microwave mixed darkroom multipath interference signal generated under a medium mode of three reflecting plates (N is 3, the materials can be PEC, aluminum and copper, and the shapes can be plane, concave and convex), a random forest network is formed by adopting a multipath interference suppression method based on random forest and jointly deciding the output of the whole model through a plurality of CARTs; under the fixed communication environment such as a darkroom, the material, the position and the like of an interference source are basically fixed, and the interference caused by the material, the position and the like can be fitted to the interference condition of each sampling point in the whole receiving area through a large amount of signal data; fitting a multipath interference suppression model representing the whole sampling grid by the random forest through the data of the sampling grid; when a new synthesized signal is sent into the random forest model, the corresponding optimal grid point is selected through the characteristics of the synthesized signal, and a corresponding direct signal is directly predicted; and fitting interference signals and other interference factors through interference antenna directional diagrams of 10 sampling points and corresponding interference-free antenna directional diagrams, adopting 10-fold cross validation to ensure that each sampling point can be tested, and finally taking the average value of errors of the sampling points as the performance index of the random forest.
The BP neural network algorithm may specifically include the steps of: when the interference source is a microwave hybrid darkroom multipath interference signal generated under a difficult mode of five reflecting plates (N is 5, the materials can be PEC, aluminum and copper, and the shapes can be plane, concave and convex), a multipath interference suppression method based on a BP neural network is adopted, wherein the BP neural network is composed of an input layer, a hidden layer and an output layer, and the input and the output of the BP neural network are set to be 720 multiplied by 10 vectors; under the condition that the dimensionality of an input vector is too high and overflow is easy to occur, determining that the number of hidden layers is 1-3 and the number of hidden layer neurons is 5-7; in the multipath interference suppression method based on the BP neural network, analog measurement data is divided into five arrays; wherein, the first array is a training array which is divided into a training input vector and a training output vector which respectively correspond to the input layer and the output layer, and the calculation of the gradient and the deviation is carried out in the hidden layer, so that the numerical value of the connection weight is continuously changed until the requirement is met; the second array is a test array, is divided into a test input vector and a test output vector in the same way, is used for verifying the calculation result, avoids the condition of unsuccessful fitting or overfitting, and is also used for testing the fitting effect of the training neural network; the fifth vector array is a target input vector and is used for solving a target output vector; in the multipath interference suppression method based on the BP neural network, because a rotation measurement method is adopted in the multipath interference simulation experiment process of a microwave mixing darkroom, the dimension of each input vector is 360 degrees; since the BP neural network can only fit continuous functions in a limited region, the real part and the imaginary part of the input vector are respectively taken out and combined into an effective input vector with the dimension of 720 degrees;
collecting data by sampling points, and carrying out a plurality of tests by changing the number of hidden layers and the number of hidden layer neurons to determine the number of the suitable hidden layers; performing normalization processing on all data by adopting a Mapminmax function, measuring a loss function by adopting a least square method, and adopting a neural network library function system initialization scheme for other network parameters such as weight, bias and the like; the higher the iteration frequency is, the smoother the mean square error curve is, and the iteration frequency needs to prevent the curve from being over-fitted under the condition of ensuring the fitting precision;
training a mapping relation model from an antenna radiation value under an interference environment to an antenna radiation value under an interference-free environment according to preset initial parameters; inputting an antenna radiation value in an interference environment into a currently trained model for prediction to obtain a predicted value, and measuring the total difference between the predicted value and a real value by adopting a mean square error mode between the real values; and continuously repeating the operation steps and adjusting parameters, minimizing the square sum of errors by using a least square method, searching the optimal function matching of data, minimizing the square sum of the errors between the model output and the antenna radiation value under the non-interference environment, setting iteration times, and recording all weights and offsets in the neural network when the errors are minimized when the errors are close to converging to the minimum value.
The number of iterations is preferably 200.
The invention relates to a K clustering selective inhibition method by using machine learning and a neural network, namely, respectively selecting proper algorithms according to the classification of interference degrees. When the multipath interference mode is simple (one reflecting plate), a polynomial regression algorithm is applied, a gradient descent method is adopted to train the model until the error is converged, the current optimal multipath interference suppression model is stored, the operation is simple, and the precision is improved. In the multipath interference mode, the multipath interference suppression under the condition of three interference paths is realized by adopting a random forest (three reflecting plates). When the multipath interference mode is difficult (five reflecting plates), the neural network is adopted to realize multipath interference suppression under the condition of five interference paths, and finally, the optimization efficiency of the antenna is improved.
Compared with the existing design method for multipath interference suppression, the design method has the following outstanding advantages:
1. the K clustering selective inhibition method based on the machine learning and the neural network increases the flexibility of algorithm selection, different algorithms can be selected according to different interference conditions, the superiority of the machine learning and the generalization capability of the neural network can be fully exerted, and the optimization efficiency of the antenna is finally improved;
2. the multipath interference suppression model is trained through polynomial regression, the optimal model obtained through training is made to be persistent, the multipath interference suppression is carried out through directly calling the model, and the calculation amount of the algorithm can be effectively reduced;
3. the random forest gets rid of the limitation that polynomial regression must model expressions, is suitable for signal suppression of three interference paths, and can increase the suppression precision of multipath interference;
4. the neural network can be used for the optimization process of other antennas after the trained model is duralized, so that the optimization efficiency of other antennas can be improved, and the cost required by redesigning the multipath interference suppression model is reduced.
Drawings
Fig. 1 is a 3D model diagram of multipath interference in a microwave mixing darkroom according to an embodiment of the present invention.
Fig. 2 is a flowchart of a machine learning algorithm based on microwave hybrid darkroom multipath interference suppression model training according to an embodiment of the present invention.
Fig. 3 is a diagram of a neural network structure according to an embodiment of the present invention.
Fig. 4 is a flowchart of a neural network algorithm based on microwave hybrid darkroom multipath interference suppression model training according to an embodiment of the present invention.
Fig. 5 is a graph of the average error of the polynomial regression algorithm corresponding to 1-N reflecting plate in the embodiment of the present invention.
Fig. 6 is an average error graph of a random forest algorithm corresponding to 3 reflectors in an embodiment of the present invention.
Fig. 7 is a graph of average error of the neural network algorithm corresponding to 5 reflectors in the embodiment of the present invention.
Detailed Description
The following examples will further illustrate the present invention with reference to the accompanying drawings.
The design steps of the embodiment of the invention are as follows:
step 1: establishing and training a multipath interference suppression model;
the specific steps for establishing and training the multipath interference suppression model may be: taking a monopole antenna as an experimental antenna, taking an x axis on a plane where vertical waves are transmitted as sampling points, taking an original point as a starting point and 0.1mm as a moving unit to take 10 sampling points in the positive direction of the x axis, and respectively collecting data with interference and data without interference; training data by taking a synthesized signal of an interference signal and a direct signal as an input and taking a direct signal of the same sampling point as an output; the synthetic signal of any point on the matrix surface is used as input, the obtained output value is compared with the known direct signal at the point, the error sum of the output value and the known direct signal is further defined, the gradient of the error is solved, the weight and the offset are updated by adopting a gradient descent method, iteration is continuously carried out until the output value is infinitely close to the known direct signal, the training is finished, and the model is stored.
Step 2: setting the test condition of the antenna in a microwave darkroom;
system modeling of microwave hybrid darkroom measurement antenna pattern as shown in fig. 1, multipath interference in a microwave hybrid darkroom is characterized as a three-dimensional MPM process. The measurement system of the antenna far field pattern is assumed as: an antenna AUT to be measured in a radiation field is used as a transmission antenna TX, a receiving antenna RX moves along a rectangular plane area, and measurement data are collected at each sampling point, wherein the rectangular plane area is taken from one part of a limited scanning surface of a plane near-field measurement system;
the transmission antenna is set to be positioned in a far field area of the receiving antenna so as to meet far field measurement conditions of the antenna. Similarly, multipath signals from other sources or diffractions may be considered far-field signals, incident on the receiving area as approximately area plane waves. In the measuring process, the rectangular area is divided into L multiplied by M rectangular lattices to collect sampling signals, so that the effect of simple and convenient calculation is achieved; the non-interference simulation model adopts a monopole antenna, a transmission antenna TX is positioned at an origin point and has coordinates of (0,0,0), a reception antenna RX is positioned on a y axis and has coordinates of (0,2.5, 0). Because the electromagnetic simulation software is in an interference-free environment, interference cannot exist, and the establishment of an interference-free simulation model is met. The distance between the transmitting antenna (TX) and the receiving antenna (RX) is 2.5m, and the data is measured by moving the receiving antenna in dx and dy directions, respectively. The antenna at each position is rotated by 1 degree, and the 360-degree rotation process is repeated for 360 times and measured, so that the whole directional diagram is determined.
And step 3: adding reflecting plate interference sources of different materials or shapes to establish an interference model;
the interference source can adopt PEC, aluminum and copper reflecting plates, and the reflecting plates can adopt a plane, a concave surface and a convex surface; the reflecting plate is selected from different materials or shapes to be trained for multiple times, so that the model can generalize different interference sources. And finally, obtaining the antenna radiation value under the interference environment through simulation operation, and deriving the antenna radiation value under the interference environment from software for use.
And 4, step 4: dividing the difficulty situation of the interference degree;
the method comprises the following steps of intelligently dividing multipath interference signals into a simple mode, a medium mode and a difficult mode according to different multipath interference degrees, respectively corresponding to N reflecting plates (N is 1, 3 and 5) multipath interference sources, and carrying out comparative analysis on average errors of the N reflecting plates under each material, so as to simulate the diversity of the interference sources in the real communication process; and correspondingly and respectively selecting proper algorithms according to the classification of the interference degree to finish the multi-path interference suppression of the microwave signals intelligently selected by the K clustering.
In step 5: respectively selecting proper algorithms according to the classification of the interference degree to finish the multi-path interference suppression of the microwave signals intelligently selected by the K clustering;
the simple mode corresponds to 1 reflecting plate, and a polynomial regression algorithm is selected; selecting a random forest algorithm according to the 3 reflectors in the medium mode; the difficult mode corresponds to 5 reflecting plates, a BP neural network algorithm is selected, and the method specifically comprises the following steps:
(1) when an interference source is a microwave mixing darkroom multipath interference signal generated under a simple mode of a reflecting plate (N is 1, materials can be PEC, aluminum and copper, and shapes can be plane, concave and convex), a multipath interference suppression method based on polynomial regression is adopted, simulation data are divided into a training set and a test set, the training set is used for carrying out data mining work on a machine learning algorithm, the learning rate is set to be 0.01, a loss function adopts a mean square error, so that a reliable characteristic is obtained to separate a standard signal and an interference signal, the test set is used for verifying the reliability of a model, cross verification is adopted to ensure that each sampling point is tested, and finally, an average error value is taken as a model evaluation index;
and training a mapping relation model from the antenna radiation value under the interference environment to the antenna radiation value under the interference-free environment according to the preset initial parameters. And inputting the antenna radiation value in the interference environment into a currently trained model for prediction to obtain a predicted value, and measuring the total difference between the predicted value and the actual value by adopting a mean square error mode between the actual values. And continuously repeating the operation steps and adjusting parameters, setting different propagation path numbers, and when the number is 5, setting the corresponding error to be minimum, so that a multipath interference signal propagation process of the whole darkroom is modeled by 5 paths of interference signals, minimizing a loss function until convergence by a gradient descent method, and storing the model.
(2) When the interference source is a microwave mixed darkroom multipath interference signal generated under a medium mode of three reflecting plates (N is 3, the materials can be PEC, aluminum and copper, and the shapes can be plane, concave and convex), a random forest network is formed by adopting a multipath interference suppression method based on random forest and jointly deciding the output of the whole model through a plurality of CARTs. Under the fixed communication environment such as darkroom, the material and the position of the interference source are basically fixed, and the interference caused by the interference can be fit to the interference condition of each sampling point in the whole receiving area through a large amount of signal data. Fitting a multipath interference suppression model representing the whole sampling grid by the random forest through the data of the sampling grid; when a new synthesized signal is sent into the random forest model, the corresponding optimal grid point is selected through the characteristics of the synthesized signal, and the corresponding direct signal is directly predicted.
In the multipath interference suppression method based on polynomial regression and random forest algorithm (1) and (2), interference signals and other interference factors are fitted through interference antenna directional diagrams of 10 sampling points and corresponding interference-free antenna directional diagrams under interference source simulation. By adopting 10-fold cross validation, the validity of a simulation experiment result can be ensured, each sampling point can be tested, and finally the average value of errors of 10 sampling points is taken as a performance index of polynomial regression and random forests;
(3) the microwave hybrid darkroom multipath interference signal generated in the difficult mode when the interference source is five reflecting plates (N ═ 5, PEC, aluminum and copper can be taken as materials and plane, concave and convex can be taken as shapes), adopts the multipath interference suppression method based on the BP neural network, wherein the BP neural network is composed of an input layer, a hidden layer and an output layer, and the input and the output of the BP neural network are set to be 720 multiplied by 10 vectors. Under the condition that the dimensionality of an input vector is too high and overflow is easy to occur, determining that the number of hidden layers is 1-3 and the number of hidden layer neurons is 5-7; in the multipath interference suppression method based on the BP neural network, analog measurement data is divided into five arrays. Wherein, the first pair of arrays is a training array, which is divided into a training input vector and a training output vector, which respectively correspond to the input layer and the output layer shown in fig. 3, and the calculation of the gradient and the deviation is performed in the hidden layer, so as to continuously change the numerical value of the connection weight to meet the requirements; the second array is a test array, is divided into a test input vector and a test output vector in the same way, is used for verifying the calculation result, avoids the condition of unsuccessful fitting or overfitting, and is also used for testing the fitting effect of the training neural network; the fifth vector array is a target input vector and is used for solving a target output vector; in the multipath interference suppression method based on the BP neural network, because a rotation measurement method is adopted in the multipath interference simulation experiment process of a microwave mixing darkroom, the dimension of each input vector is 360 degrees. And because the BP neural network can only fit continuous functions in a limited region, the real part and the imaginary part of the input vector are respectively taken out and combined into an effective input vector with the dimension of 720 degrees.
Data is collected at each time taking 10 samples, so the input and output of the network are 720 x 10 vectors. Considering the condition that the dimensionality of an input vector is too high and overflow is easy, the hidden layer number and the hidden layer neuron number are changed, and multiple times of experiments are carried out, so that the hidden layer number determined to be suitable for by the method is 1 and the hidden layer neuron number is 6 finally. The structure of the neural network is shown in fig. 4. All data were normalized using the Mapminmax function. The loss function is measured by a least square method, and other network parameters such as weight, bias and the like are measured by a neural network library function system initialization scheme. The higher the number of iterations, the smoother the mean square error curve, and the curve tends to stabilize at 130 iterations. In an actual training process, the number of iterations is determined as a result of multiple attempts. The iteration number needs to prevent the curve from being over-fitted under the condition of ensuring the fitting precision. By comparison, the number of iterations was chosen to be 200.
And training a mapping relation model from the antenna radiation value under the interference environment to the antenna radiation value under the interference-free environment according to the preset initial parameters. And inputting the antenna radiation value in the interference environment into a currently trained model for prediction to obtain a predicted value, and measuring the total difference between the predicted value and the actual value by adopting a mean square error mode between the actual values. And continuously repeating the operation steps and adjusting parameters, minimizing the square sum of errors by using a least square method, searching the optimal function matching of data, minimizing the square sum of the errors between the model output and the antenna radiation value under the non-interference environment, setting iteration times, and recording all weights and offsets in the neural network when the errors are minimized when the errors are close to converging to the minimum value.
The system modeling of the microwave hybrid darkroom measurement antenna pattern is shown in fig. 1.
A flow chart of the microwave hybrid darkroom multipath interference suppression algorithm combined with the polynomial regression algorithm is shown in fig. 2.
The structure of the neural network is shown in fig. 3.
A microwave hybrid darkroom multipath interference suppression algorithm flow chart incorporating a neural network algorithm is shown in fig. 4.
Fig. 5, 6, and 7 are graphs of average errors of polynomial regression, random forest, and neural network algorithms corresponding to N reflectors (N ═ 1, 3, and 5). The abscissa is the number of iterations and the ordinate is the error in dB. Because the random forest has no model training iteration times in a strict sense, iteration is mainly carried out according to the splitting condition of each subtree, and quantization is difficult, the average error of the random forest is directly drawn in a graph by a constant value and does not change along with the iteration times, so that the errors of the three algorithms can be intuitively felt conveniently.
The error is an average value under simulation of different interference sources, and comprises simulation of the interference sources with different materials, shapes and numbers. In a simple mode where the source of the interference is a reflecting plate (N ═ 1, PEC, aluminum and copper can be used as materials and plane, concave and convex shapes can be used as materials), a polynomial regression algorithm is used, the signal propagation expression modeling is relatively intuitive and easy to understand, and converges at about 40 iterations with an error of 0.212 dB. The random forest algorithm is used in medium mode with three reflectors (N-3, PEC, aluminum and copper as materials and flat, concave and convex as shapes) as the source of interference, without modeling signal propagation expressions, with a final error of 0.173 dB. The neural network algorithm is used in a difficult mode where the interference source is five reflecting plates (N-5, PEC, aluminum and copper are taken as materials and plane, concave and convex shapes are taken), the neural network does not need to model signal propagation expressions like random forest, and converges at about 15 iterations with an error of 0.100 dB. When the interference of a difficult mode is suppressed by polynomial regression and random forests, the error is larger, the neural network is well represented, the large interference can be effectively suppressed, and the overall representation is more stable. Under the inhibition method of K cluster selection based on machine learning and neural network, the interference signal is almost completely eliminated, and the error is as low as about 0.2 dB. Compared with the traditional single algorithm, the K clustering selective suppression method based on the machine learning and the neural network is obviously more effective and accurate, and has high suppression sensitivity on large interference signals and small interference signals. After the K clustering selection suppression model based on machine learning and neural network is successfully trained, the K clustering selection suppression model is applied to other fields, such as multipath interference suppression in satellite communication, so that the superiority of the K clustering selection suppression model can be fully exerted, model parameters suitable for multipath interference suppression in satellite communication are reset and trained, the efficiency of redesigning the multipath interference suppression model when the multipath interference suppression model is applied to another field can be improved, and the K clustering selection suppression model has higher suppression accuracy compared with the traditional method. And the optimization efficiency of the antenna can be improved by correspondingly and respectively selecting proper algorithms according to the interference degree classification. If the most serious interference situation such as five aluminum or copper reflecting plates and more than five interference sources occurs, a weighting comprehensive optimization scheme of more than two algorithms can be adopted, so that a relatively ideal interference suppression effect is achieved, and the weighting comprehensive optimization scheme of more than two algorithms expands a new direction for the future research of the multipath interference work of a microwave anechoic chamber.
Claims (6)
- A microwave signal multipath interference suppression method for K clustering intelligent selection is characterized by comprising the following steps:1) the method comprises the following steps of establishing and training a multipath interference suppression model: taking a monopole antenna as an experimental antenna, taking an x axis on a plane where vertical waves are transmitted as sampling points, taking an original point as a starting point and 0.1mm as a moving unit to take 10 sampling points in the positive direction of the x axis, and respectively collecting data with interference and data without interference; training data by taking a synthesized signal of an interference signal and a direct signal as an input and taking a direct signal of the same sampling point as an output; taking a synthetic signal of any point on a matrix surface as input, comparing an obtained output value with a known direct signal at the point, further defining the error sum of the output value and the known direct signal, calculating the gradient of the error, updating a weight and an offset by adopting a gradient descent method, continuously iterating until the output value is infinitely close to the known direct signal, finishing training and storing a model;2) setting the test condition of the antenna in a microwave darkroom, and specifically comprising the following steps: characterizing multipath interference in a microwave mixing darkroom as a three-dimensional MPM process, and assuming a measuring system of an antenna far-field directional pattern as: an antenna AUT to be measured in a radiation field is used as a transmission antenna TX, a receiving antenna RX moves along a rectangular plane area, and measurement data are collected at each sampling point, wherein the rectangular plane area is taken from one part of a limited scanning surface of a plane near-field measurement system;setting the transmission antenna to be positioned in a far field area of the receiving antenna so as to meet the far field measurement condition of the antenna; similarly, multipath signals from other signal sources or diffraction can be regarded as far-field signals, and are incident to a receiving area as approximate area plane waves; during the measurement, the rectangular area is divided intoL×MSampling signals are collected by the rectangular lattices, so that the effect of simple and convenient calculation is achieved;3) adding reflecting plate interference sources of different materials or shapes to establish an interference model;4) dividing the difficulty situation of the interference degree; the difficulty situation of dividing the interference degree intelligently divides the multipath interference signals into three types of simple mode, medium mode and difficult mode according to the difference of the multipath interference degree, and the three types respectively correspond to the simple mode, the medium mode and the difficult modeNThe block reflecting plate is a multi-path interference source,N= 1, 3, 5; for each materialNThe block reflecting plate carries out comparison analysis on average errors so as to simulate the diversity of interference sources in the real communication process; respectively selecting proper algorithms according to the classification of the interference degree to finish the multi-path interference suppression of the microwave signals intelligently selected by the K clustering;5) respectively selecting proper algorithms according to the interference degree classification to finish K-clustering intelligent selective microwave signal multipath interference suppression, respectively selecting proper algorithms according to the interference degree classification, corresponding to 1 reflecting plate in a simple mode, and selecting a polynomial regression algorithm; selecting a random forest algorithm according to the 3 reflectors in the medium mode; the difficult mode corresponds to 5 reflecting plates, and a BP neural network algorithm is selected.
- 2. The method for suppressing multi-path interference of microwave signals intelligently chosen through K-clustering according to claim 1, wherein in step 3), the interference source is a PEC, aluminum, copper reflector plate, and the reflector plate is flat, concave, convex; the reflecting plate is selected from different materials or shapes to be trained for multiple times, so that the model can be generalized into different interference sources; and finally, obtaining an antenna radiation value under the interference environment through simulation operation, and deriving the antenna radiation value under the interference environment for use.
- 3. The method for suppressing multi-path interference of microwave signals intelligently chosen through K-clustering according to claim 1, wherein in step 5), the polynomial regression algorithm specifically comprises the steps of: when an interference source is a microwave mixed darkroom multipath interference signal generated under a simple mode of a reflecting plate, a multipath interference suppression method based on polynomial regression is adopted, simulation data are divided into a training set and a testing set, the training set is used for carrying out data mining work on a machine learning algorithm, the learning rate is set to be 0.01, a loss function adopts a mean square error, so that reliable characteristics are obtained to separate a standard signal and an interference signal, the testing set is used for verifying the reliability of a model, cross verification is adopted to ensure that each sampling point is tested, and finally, an average error value is taken as a model evaluation index;training a mapping relation model from an antenna radiation value under an interference environment to an antenna radiation value under an interference-free environment according to preset initial parameters; inputting an antenna radiation value in an interference environment into a currently trained model for prediction to obtain a predicted value, and measuring the total difference between the predicted value and a real value by adopting a mean square error mode between the real values; and continuously repeating the operation steps and adjusting parameters, setting different propagation path numbers, and when the number is 5, setting the corresponding error to be minimum, so that a multipath interference signal propagation process of the whole darkroom is modeled by 5 paths of interference signals, minimizing a loss function until convergence by a gradient descent method, and storing the model.
- 4. The method for suppressing multi-path interference of microwave signals intelligently chosen by K-clustering according to claim 1, wherein in step 1), the specific steps of the random forest algorithm are as follows: when the interference source is a microwave mixed darkroom multipath interference signal generated under a medium mode of three reflecting plates, a random forest-based multipath interference suppression method is adopted, and the output of the whole model is jointly decided by a plurality of CARTs to form a random forest network; under the fixed communication environment of a darkroom, the material and the position of an interference source are fixed, and the interference caused by the fixed material and the fixed position is fitted to the interference condition of each sampling point in the whole receiving area through a large amount of signal data; fitting a multipath interference suppression model representing the whole sampling grid by the random forest through the data of the sampling grid; when a new synthesized signal is sent into the random forest model, the corresponding optimal grid point is selected through the characteristics of the synthesized signal, and a corresponding direct signal is directly predicted; and cross validation is adopted to ensure that each sampling point can be tested, and the average value of errors of the sampling points is used as a performance index of the random forest.
- 5. The method for suppressing multi-path interference of microwave signals intelligently chosen by K-clustering according to claim 1, wherein in step 3), the BP neural network algorithm specifically comprises the steps of: when the interference source is a microwave mixed darkroom multipath interference signal generated under the difficult mode of five reflecting plates, a multipath interference suppression method based on a BP neural network is adopted, wherein the BP neural network consists of an input layer, a hidden layer and an output layer, and the input and the output of the BP neural network are set as vectors of 720 multiplied by 10; under the condition that the dimensionality of an input vector is too high and overflow is easy to occur, determining that the number of hidden layers is 1-3 and the number of hidden layer neurons is 5-7; in the multipath interference suppression method based on the BP neural network, analog measurement data is divided into five arrays; wherein, the first array is a training array which is divided into a training input vector and a training output vector which respectively correspond to the input layer and the output layer, and the calculation of the gradient and the deviation is carried out in the hidden layer, so that the numerical value of the connection weight is continuously changed until the requirement is met; the second array is a test array, is divided into a test input vector and a test output vector in the same way, is used for verifying the calculation result, avoids the condition of unsuccessful fitting or overfitting, and is also used for testing the fitting effect of the training neural network; the fifth vector array is a target input vector and is used for solving a target output vector; in the multipath interference suppression method based on the BP neural network, because a rotation measurement method is adopted in the multipath interference simulation experiment process of a microwave mixing darkroom, the dimension of each input vector is 360 degrees; since the BP neural network can only fit continuous functions in a limited region, the real part and the imaginary part of the input vector are respectively taken out and combined into an effective input vector with the dimension of 720 degrees;collecting data by sampling points, and carrying out a plurality of tests by changing the number of hidden layers and the number of hidden layer neurons to determine the number of the suitable hidden layers; normalization processing is carried out on all data by adopting a Mapminmax function, a loss function is measured by adopting a least square method, and a neural network library function system initialization scheme is adopted for network parameter weight and bias; the higher the iteration frequency is, the smoother the mean square error curve is, and the iteration frequency needs to prevent the curve from being over-fitted under the condition of ensuring the fitting precision;training a mapping relation model from an antenna radiation value under an interference environment to an antenna radiation value under an interference-free environment according to preset initial parameters; inputting an antenna radiation value in an interference environment into a currently trained model for prediction to obtain a predicted value, and measuring the total difference between the predicted value and a real value by adopting a mean square error mode between the real values; and continuously repeating the operation steps and adjusting parameters, minimizing the square sum of errors by using a least square method, searching the optimal function matching of data, minimizing the square sum of the errors between the model output and the antenna radiation value under the non-interference environment, setting iteration times, and recording all weights and offsets in the neural network when the errors are minimized when the errors are close to converging to the minimum value.
- 6. The method for K-cluster intelligent-picking microwave signal multipath interference mitigation of claim 5, wherein the number of iterations is 200.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911398258.0A CN111122989B (en) | 2019-12-30 | 2019-12-30 | K-clustering intelligent selection microwave signal multipath interference suppression method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911398258.0A CN111122989B (en) | 2019-12-30 | 2019-12-30 | K-clustering intelligent selection microwave signal multipath interference suppression method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111122989A CN111122989A (en) | 2020-05-08 |
| CN111122989B true CN111122989B (en) | 2020-12-01 |
Family
ID=70505454
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911398258.0A Active CN111122989B (en) | 2019-12-30 | 2019-12-30 | K-clustering intelligent selection microwave signal multipath interference suppression method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111122989B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111596159B (en) * | 2020-06-11 | 2021-06-15 | 青岛大学 | Electronic system EMI detection and positioning method based on six-axis mechanical arm |
| CN112118531B (en) * | 2020-09-12 | 2021-06-29 | 上海大学 | Privacy protection method of crowd sensing application based on position |
| CN113452379B (en) * | 2021-07-16 | 2022-08-02 | 燕山大学 | Section contour dimension reduction model training method and system and data compression method and system |
| JP7382430B2 (en) | 2022-02-18 | 2023-11-16 | アンリツ株式会社 | Antenna device and performance test method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6795424B1 (en) * | 1998-06-30 | 2004-09-21 | Tellabs Operations, Inc. | Method and apparatus for interference suppression in orthogonal frequency division multiplexed (OFDM) wireless communication systems |
| CN102063642A (en) * | 2010-12-30 | 2011-05-18 | 上海电机学院 | Selection method for fuzzy neural network sample on basis of active learning |
| US9312895B1 (en) * | 2008-08-07 | 2016-04-12 | Hypres, Inc. | Two stage radio frequency interference cancellation system and method |
| CN106899326A (en) * | 2017-03-10 | 2017-06-27 | 重庆邮电大学 | A kind of method of raising IDMA system baseband signal estimated accuracies |
| CN108551430A (en) * | 2018-03-13 | 2018-09-18 | 上海交通大学 | A kind of equal balance system of adaptive frequency domain based on multipath self-interference channel |
| CN109738856A (en) * | 2019-01-29 | 2019-05-10 | 西南交通大学 | A kind of noise suppressed Capon active target DOA evaluation method based on time reversal |
| CN109766942A (en) * | 2019-01-07 | 2019-05-17 | 西南交通大学 | A kind of small-sample learning image-recognizing method based on attention neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8284819B2 (en) * | 2009-10-21 | 2012-10-09 | Broadcom Corporation | Method and system for interference suppression in WCDMA systems |
-
2019
- 2019-12-30 CN CN201911398258.0A patent/CN111122989B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6795424B1 (en) * | 1998-06-30 | 2004-09-21 | Tellabs Operations, Inc. | Method and apparatus for interference suppression in orthogonal frequency division multiplexed (OFDM) wireless communication systems |
| US9312895B1 (en) * | 2008-08-07 | 2016-04-12 | Hypres, Inc. | Two stage radio frequency interference cancellation system and method |
| CN102063642A (en) * | 2010-12-30 | 2011-05-18 | 上海电机学院 | Selection method for fuzzy neural network sample on basis of active learning |
| CN106899326A (en) * | 2017-03-10 | 2017-06-27 | 重庆邮电大学 | A kind of method of raising IDMA system baseband signal estimated accuracies |
| CN108551430A (en) * | 2018-03-13 | 2018-09-18 | 上海交通大学 | A kind of equal balance system of adaptive frequency domain based on multipath self-interference channel |
| CN109766942A (en) * | 2019-01-07 | 2019-05-17 | 西南交通大学 | A kind of small-sample learning image-recognizing method based on attention neural network |
| CN109738856A (en) * | 2019-01-29 | 2019-05-10 | 西南交通大学 | A kind of noise suppressed Capon active target DOA evaluation method based on time reversal |
Non-Patent Citations (2)
| Title |
|---|
| Multipath interference suppression based on a stable and effective neural network;Jianhua Zhou 等;《2018 Progress in electromagnetics research symposium》;20180804;第2066-2071页 * |
| 基于矩阵束和神经网络两种方法的多径干扰抑制研究;唐瑜;《中国优秀硕士学位论文全文数据库 工程科技II辑》;20190715(第7期);论文摘要、正文第25-27页、第42页，附图3.1-3.2 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111122989A (en) | 2020-05-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111122989B (en) | K-clustering intelligent selection microwave signal multipath interference suppression method | |
| CN108696932B (en) | Outdoor fingerprint positioning method using CSI multipath and machine learning | |
| CN101349742B (en) | Method for optimizing space between broad band phased array elements and measuring frequency and direction of frequency domain multiple targets | |
| Xiang et al. | Improved de-multipath neural network models with self-paced feature-to-feature learning for DOA estimation in multipath environment | |
| CN105224743B (en) | A kind of antenna arrangement optimization on full frequency band based on particle cluster algorithm | |
| Zhang et al. | Multiple sources localization by the WSN using the direction-of-arrivals classified by the genetic algorithm | |
| Türkyilmaz et al. | Environment-aware location estimation in cellular networks | |
| Kutty et al. | An improved numerical optimization method for efficient beam search in 60 GHz indoor millimeter wave wireless networks | |
| CN110830939B (en) | Positioning method based on improved CPN-WLAN fingerprint positioning database | |
| Liu et al. | Location-based visible region recognition in extra-large massive MIMO systems | |
| CN116306199A (en) | Efficient optimization method based on multivariate vector control cross eye interference technology | |
| Wu et al. | Enhanced path loss model by image-based environmental characterization | |
| WO2023187447A1 (en) | Method for determining the correct placement of an antenna with radiation pattern prediction | |
| Quezada-Gaibor et al. | Lightweight hybrid CNN-ELM model for multi-building and multi-floor classification | |
| Timoteo et al. | A hybrid machine learning approach for mobile user positioning in cellular networks | |
| Labbaf et al. | Robust DoA Estimation in a Uniform Circular Array Antenna with Errors and Unknown Parameters Using Deep Learning | |
| Chen et al. | Neural network for WGDOP approximation and mobile location | |
| Faye et al. | Two-dimensional DOA estimation based on a single uniform linear array | |
| Wu et al. | A Weighted Random Forest Based Positioning Algorithm for 6G Indoor Communications | |
| Vilovic et al. | Design of an indoor wireless network with neural prediction model | |
| CN114742081A (en) | Phased array antenna optimized deployment method suitable for ultrahigh frequency RFID positioning system | |
| Zheng et al. | An intelligent wireless communication model based on multi-feature fusion and quantile regression neural network | |
| Calist et al. | Large Scale Channel Parameters Estimation by Utilization of Reflected Rays Information for an Urban Environment | |
| Chen et al. | General Improvements of Heuristic Algorithms for Low Complexity DOA Estimation | |
| Zheng et al. | DoA and ToA Estimation Method of OFDM Signal Based on Cascaded Deep Neural Network. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |