CN106355245B - A kind of Pattern Synthesis of Antenna Array method based on neural network algorithm - Google Patents
A kind of Pattern Synthesis of Antenna Array method based on neural network algorithm Download PDFInfo
- Publication number
- CN106355245B CN106355245B CN201610817682.4A CN201610817682A CN106355245B CN 106355245 B CN106355245 B CN 106355245B CN 201610817682 A CN201610817682 A CN 201610817682A CN 106355245 B CN106355245 B CN 106355245B
- Authority
- CN
- China
- Prior art keywords
- array
- antenna array
- network
- antenna
- sample
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- 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
Abstract
The Pattern Synthesis of Antenna Array method based on neural network algorithm that the present invention provides a kind of, computationally intensive to solve traditional array Antenna measuring table technology, the calculating time is long, and best initial weights find extremely difficult problem.The method includes antenna array model foundation step, best initial weights obtaining step and antenna array pattern generation step.The present invention uses neural network algorithm in Pattern Synthesis of Antenna Array technology, trained RBF neural is effectively used for Pattern Synthesis of Antenna Array, neural network algorithm is utilized to realize Pattern Synthesis of Antenna Array, aggregate velocity is especially fast, null level is low, main lobe protrudes, and pattern shapes are stablized.
Description
Technical field
The present invention relates to mobile communication antenna technical fields.
Background technology
Pattern Synthesis of Antenna Array technology is applied to any antenna array with sophisticated signal processor, it can be adjusted
Or its adaptive beam pattern, it is therefore an objective to enhance interested signal and reduce interference signal, additionally it is possible to mitigate multipath effect
Adverse effect, it is one of research field of smart antenna.With social information exchange sharply increase, personal mobile communications
Rapid proliferation, frequency spectrum has become more and more valuable resource, and smart antenna uses space division multiple access technique, is propagated using signal
Difference on direction changes array aerial direction figure, to inhibit by adjusting weighted amplitude and the phase of each array element signals
Interference, improve signal-to-noise ratio, power system capacity and permission signal bandwidth, the resources such as effectively save frequency spectrum and power.
The directional diagram of antenna array system is realized by the superposition of the directional diagram of each oscillator, if each oscillator
The amplitude and phase of voltage drive are varied from, then the intelligent antenna system wants corresponding directional diagram that can also change.
In mobile communication application, many times base station is relatively-stationary, and terminal is mobile；In order to which maintenance is good between them
Wireless channel, ensure the normal work of link, it is desirable to which the main lobe of smart antenna is capable of the track terminal at moment, that is to say, that
The directional diagram of intelligent antenna system will do the adjustment of self with the change in location of user.The present invention is using in the shortest time
Find a vector so that the direction of the main lobe alignment user of the antenna radiation pattern corresponding to it.
The present invention mainly utilizes intelligent algorithm to realize that linear array antenna Pattern Synthesis, neural network algorithm have very strong appearance
Wrong and Function approximation capabilities, using multigroup ideal directional diagram sample data train RBF Neural Network, when training network, is non-
The independent variable and dependent variable of Linear Mapping are the snapshot of signal direction of arrival and the optimum weight coefficient value of array element, signal snapshot respectively
It is randomly selecting, a large amount of.After network training is good, when inputting the direction of arrival at different angles, the adaptive life of antenna array system
At directional diagram.
Invention content
The shortcomings that it is an object of the invention to overcome the prior art and deficiency, provide a kind of battle array based on neural network algorithm
Array antenna Pattern Synthesis method, this method aggregate velocity is especially fast, and null level is low, and main lobe protrudes, and pattern shapes are steady
It is fixed.
The purpose of the present invention is achieved through the following technical solutions：A kind of array aerial direction figure based on neural network algorithm
Integrated approach, including：
Antenna array model foundation step：N number of COCO antennas form array, and the array includes N2A monopole antenna, day
It is 1/2 medium wavelength that linear array, which uses uniform straight line array form, bay spacing d≤λ, the unit section length of side,：
According to the operating center frequency f=1.8GHz of antenna, the permittivity ε of substrate materialr=2.56, to make positive and negative
Micro-strip section is in picturesque disorder, generates suitable transmission mode and radiation mode, needs a ≈ b；Dielectric-slab length L is that dielectric-slab is wide simultaneously
Spend W 6.5 times, wherein b are chip unit interval；
Best initial weights obtaining step：Including information determining step, network parameter selection step, learning training algorithms selection step
Suddenly, adjustment and obtaining step；
Information determining step：It converts problem to the mode that network can express and the form that can be handled, determines each node
Input and output；Each center of a sample is calculated with K mean cluster method, N is center of a sample's number, i.e.,：It first clusters, then by value
Method asks the center of a sample of every class, is implicit layer unit with required center of a sample；
Network parameter selects step：Determine input, the number of output neuron, the number of plies of multitiered network and hidden neuron
Number, snapshot number is K, and training sample number is A, and wherein A is plural number, is divided into real and imaginary parts calculating, desired output is
The real and imaginary parts of the plural weight coefficient of array element, training sample input is by taking identical as number of training group of wave to reach at random
What angle obtained, plural weight coefficient is obtained using conjugate gradient algorithms；The selection of anticipation error is by missing two different expectations
The network of difference is trained, and therefrom determines a network；
Learning training algorithms selection step：It is first initialized, then radial basis function network is utilized to complete function approximation, determined
Input sample and desired output, wherein the snapshot A of antenna array K × N matrix is set as input sample, the weight vector ω of each array element
It is set as desired output；Antenna array array number is configured, determines arrival bearing, input vector signal is obtained after normalization, it is real
Mapping of the existing input vector signal to weight vector；
Adjustment and obtaining step：To clustering method, Hidden nodes, hidden node data center learning coefficient, hidden node extension
Constant learning coefficient and hidden node output weights learning coefficient are adjusted, and network connection, which is weighed, after completion training is determined, can
The best initial weights of desired signal are directly obtained using network；
Antenna array pattern generation step：Antenna array model is verified using the best initial weights of acquisition, to raw
At antenna array pattern.
The snapshot number K is set as 200, and training sample number A is set as 800, and gained input sample data are 1600,
800 training sample inputs obtain 800 groups of plural numbers by taking 800 groups of direction of arrival to obtain at random, using conjugate gradient algorithms
Weight coefficient trains network with these data.
Hidden node data center learning coefficient is 0.001, and it is 0.002 that hidden node, which extends constant learning coefficient, hidden section
Point output weights learning coefficient is 0.005.
The antenna array array number is set as 8, and arrival bearing is arranged between [- π, π].
Every group of direction of arrival includes a useful signal and three interference signals, and Gauss is added in each signal
Noise.
The present invention can be easy to find out best initial weights, and aggregate velocity is especially fast, and null level is low, and main lobe protrudes, and directional diagram
Dimensionally stable.
Description of the drawings
Fig. 1 is that the robust detection value of simulation result three times compares.
Fig. 2 is radial base neural net structural schematic diagram.
Fig. 3 is the array antenna figure general flowchart of neural network algorithm.
Fig. 4 is antenna model schematic diagram.
Fig. 5 is collinear array schematic diagram.
Specific implementation mode
Specific implementation mode one：A kind of Pattern Synthesis of Antenna Array method based on neural network algorithm, specifically includes
Following steps,
Antenna array model foundation step：N number of COCO antennas form array, and the array includes N2A monopole antenna, day
It is 1/2 medium wavelength that linear array, which uses uniform straight line array form, bay spacing d≤λ, the unit section length of side,：
According to the operating center frequency f=1.8GHz of antenna, the permittivity ε of substrate materialr=2.56, to make positive and negative
Micro-strip section is in picturesque disorder, generates suitable transmission mode and radiation mode, needs a ≈ b；Dielectric-slab length L is that dielectric-slab is wide simultaneously
Spend W 6.5 times, wherein b are chip unit interval；
Best initial weights obtaining step：Including information determining step, network parameter selection step, learning training algorithms selection step
Suddenly, adjustment and obtaining step；
Information determining step：It converts problem to the mode that network can express and the form that can be handled, determines each node
Input and output；Each center of a sample is calculated with K mean cluster method, N is center of a sample's number, i.e.,：It first clusters, then by value
Method asks the center of a sample of every class, is implicit layer unit with required center of a sample；
Network parameter selects step：Determine input, the number of output neuron, the number of plies of multitiered network and hidden neuron
Number, snapshot number is K, and training sample number is A, and wherein A is plural number, is divided into real and imaginary parts calculating, desired output is
The real and imaginary parts of the plural weight coefficient of array element, training sample input is by taking identical as number of training group of wave to reach at random
What angle obtained, plural weight coefficient is obtained using conjugate gradient algorithms；The selection of anticipation error is by missing two different expectations
The network of difference is trained, and therefrom determines a network；
Learning training algorithms selection step：It is first initialized, then radial basis function network is utilized to complete function approximation, determined
Input sample and desired output, wherein the snapshot A of antenna array K × N matrix is set as input sample, the weight vector ω of each array element
It is set as desired output；Antenna array array number is configured, determines arrival bearing, input vector signal is obtained after normalization, it is real
Mapping of the existing input vector signal to weight vector；
Adjustment and obtaining step：To clustering method, Hidden nodes, hidden node data center learning coefficient, hidden node extension
Constant learning coefficient and hidden node output weights learning coefficient are adjusted, and network connection, which is weighed, after completion training is determined, can
The best initial weights of desired signal are directly obtained using network；
Antenna array pattern generation step：Antenna array model is verified using the best initial weights of acquisition, to raw
At antenna array pattern.
Specific implementation mode two：Specific implementation mode two and the main distinction of specific implementation mode one be, first snapshot
Number K is set as 200, and training sample number A is set as 800, since A is plural number, so real and imaginary parts calculating need to be separated into, in this way
Input sample data 1600, desired output are the real and imaginary parts of the plural weight coefficient of array element.This 800 training sample inputs
By taking 800 groups of direction of arrival to obtain at random.800 groups of plural number weight coefficients are obtained using conjugate gradient algorithms, with these data
Training network.
Maximum frequency of training is 5000 times in a network.Under normal circumstances, the selection of learning rate is tended to choose smaller
Learning rate to ensure the stability of system, the selection range of learning rate is between 0.01~0.8.Too big learning rate
Lead to the unstable of study, too small value leads to the training time extremely grown again.Learning rate is adjusted with frequency of training, but each
The learning rate of whole network is identical in training.Specific there are two types of schemes, first, being trained with larger learning rate when starting, so
Learning rate is adjusted according to training afterwards.Second is that being trained with smaller learning rate when starting, is then adjusted and learned according to training
Habit rate.Final hidden node data center learning coefficient is 0.001, and it is 0.002 that hidden node, which extends constant learning coefficient, and hidden node is defeated
It is 0.005 to go out weights learning coefficient.
The selection of anticipation error：Under normal circumstances as a comparison, can simultaneously the expected error value different to two net
Network is trained, and finally by one of network is determined the considerations of composite factor, determines that expectation target error is 0.09.
Secondly, antenna array array number is set as 8.Arrival bearing is between [- π, π], the 800 groups of direction of arrival randomly selected,
Every group of direction of arrival includes a useful signal and three interference signals, also adds Gaussian noise in each signal.
Operation principle：
Radial basis function neural network is by up of three layers, structure chart such as Fig. 2, and input layer only transmits input signal to hidden
Layer, hidden node are made of various transmission functions, and most common function radial as Gaussian function is constituted, and
It is typically simple linear function to export node layer.
When n is 0, radial basis function exports maximum value 1, i.e. when distance reduces between weight vector ω and input vector p,
Output will increase.That is, radial basis function is locally generating response to input signal.The input signal n of function is close
When the central range of function, hidden node will generate larger output.It can be seen that this network has partial approximation energy
Power, so radial primary function network also becomes local sensing field network.Threshold value b is used to adjust the susceptibility of radial base neuron.
Assuming that input vector signal after being normalized is F, when data by input layer close to hidden layer when, hidden layer RBF quilts
Activation, it is the weight vector of time-varying that output end, which needs the amount of nonlinearity approached, to realize mappings of the F to weight vector ω.It can indicate
For：
Wherein, L, M are hidden layer and output layer number of nodes, H respectivelyi,lIt is first of node of hidden layer and i-th of node of output layer
Connection weight, ClIt is Gaussian function center, | | | | it is Euclidean Norm.The corresponding output of first of unit, which can be obtained, is：
In formula, zi(k) it is the output of first of hidden unit, xn(t) it is n-th of component of t-th of input sample vector, ct,n
For n-th of component of first of transform center vector in hidden layer, σlIt is the control parameter of corresponding first of center vector.
Omnidirectional antenna is emulated：
The great advantages that array antenna is made of COCO antennas are exactly：The array being made of N number of COCO antennas is suitable
In containing N2A monopole antenna (assuming that each COCO antennas are also to be made of N sections of microstrip line sections), but it only has N number of feedback
Electric, in comparison, if an array is made of N number of monopole antenna array, it must have N2A distributing point is (false
If each monopole subarray is also to be made of N number of monopole).Certainly, since COCO antenna structures are simple, have price advantage and
Performance advantage.
The unit section length of side is 1/2 medium wavelength：
According to the operating center frequency f=1.8GHz of antenna, dielectric constant=2.56 of substrate material, to make positive and negative
Micro-strip section is in picturesque disorder, generates suitable transmission mode and radiation mode, needs a ≈ b；Dielectric-slab length L is that dielectric-slab is wide simultaneously
Spend W 6.5 times or so, can obtain a=52mm, b=58.5mm, L=520mm, W=80mm, Wa=69.5mm, Wb=
11.4mm, h=2.5mm are punched on patch of the antenna far from feed end, and hole is connected to upper and lower surface, and the radius in hole is 3mm,
At away from feed end 480mm.Wherein h is substrate thickness, and b is chip unit interval.Antenna schematic diagram is as shown in Figure 4.Antenna
The input impedance of distributing point requires to be 50 Ω.Observe the input impedance Z parameter of the feed port under centre frequency.Adjust patch
Size and all changeable antenna of feed position input impedance value, purpose so that the real part of input impedance is 50 Ω and void
Portion is 0 Ω.
It is emulated using FEKO softwares, model is created according to data above and adjusts mesh-density, by adjusting medium
Control card, output control card, Electromagnetic Calculation card consider to become to taking effect, calculate far field, calculate standing wave etc..The feed electricity of antenna
The amplitude of pressure is 1V, and phase is 0 °.Simulation establishes 8 array element linear array antennas, in known direction of arrival, by using
The weights that algorithm obtains control the amplitude and phase of each antenna element voltage, in such a way that each array element is individually fed.It is depositing
In the case of coupling mutual inductance between smaller antenna array element, the reliability and stability of the directional diagram of generation are examined.As Fig. 5 is
The schematic diagram that 8 array element collinear arrays are placed.Linear array design uses uniform straight line array form, prevents from occurring on the faces H of antenna
Graing lobe needs to ensure bay center spacing d≤λ, and as bay spacing d≤λ/2, it may appear that Mutual Inductance Coupling is existing
As so 0.6 λ of selection d ≈.Each bay is enjoyed there are one feed voltage source.
The advantageous effect of actual moving process：
The present invention obtains each group of data mean value by many experiments emulation, it can be seen that the mean value of null level and side everywhere
The mean value of valve level all meets desired indicator and requires (referring to table 1).
The average value of 1 each group partial parameters of table
Averagely there are 3.5~3.8 nulls for being less than -30dB, illustrates that neural network algorithm is easier in every group of emulation
Low level null is generated, but in these more low level nulls, some are appeared on unexpected direction, but certainly not
It appears in useful signal principal direction, the probability of appearance accounts for deep null occurs 20%~50%, illustrates side occurring
There is null in the angle of valve.
Although there is unexpected null, it is comprehensive to have no effect on the adaptive solution array aerial direction figure of neural network algorithm
Conjunction problem, it still has practicability.For different array antennas, as long as choosing appropriate neural network algorithm, provide enough
Training sample, according to actual requirement adjust network parameter, so that it may to obtain accurate network weight.Neural network algorithm is realized
Pattern Synthesis of Antenna Array, aggregate velocity is especially fast, and null level is low, and main lobe protrudes, and pattern shapes are stablized.
Under same group, the interference signal effect in three directions, the directional diagram generated every time is very identical, tentatively illustrates to calculate
The robustness of method is good.Emulation the data obtained is calculated into robust detection value, acquired results are shown in Table 2.
2 neural network algorithm robust detection value of table
It is drawn in block diagram, such as Fig. 1.Can more intuitively it find out, the detected value variation of each robustness is little, so through
Interference signal direction of arrival direction is varied multiple times, the directional diagram obtained by the algorithm remains to adaptive transformation therewith, obtains ideal
Antenna reception.
In terms of the depth of null, null level is relatively low so that main lobe is very full.Directional diagram is formed using neural network
When, efficiency is very high, and robustness is good, and gained directional diagram is stablized.
Network connection power determines that after completing training, and the optimal power of desired signal can be directly obtained using network
Value, flow chart such as Fig. 3 of whole process utilize multigroup ideal directional diagram sample data train RBF Neural Network, training
The independent variable and dependent variable of network non-linear hour mapping, are the snapshot of signal direction of arrival and the optimum weight coefficient value of array element respectively,
Signal snapshot is randomly selecting, a large amount of.After network training is good, when inputting the direction of arrival at different angles, antenna array system is certainly
The generation directional diagram of adaptation.
Above to a kind of Pattern Synthesis of Antenna Array method based on neural network algorithm provided by the present invention, carry out
It is discussed in detail, principle and implementation of the present invention are described for specific case used herein, above example
Explanation be merely used to help understand the present invention method and its core concept；Meanwhile for those of ordinary skill in the art,
According to the thought of the present invention, there will be changes in the specific implementation manner and application range, in conclusion in this specification
Appearance should not be construed as limiting the invention.
Claims (5)
1. a kind of Pattern Synthesis of Antenna Array method based on neural network algorithm, which is characterized in that
Antenna array model foundation step：N number of COCO antennas form array, and the array includes N2A monopole antenna, aerial array
Using uniform straight line array form, bay spacing d≤λ, the unit section length of side is 1/2 medium wavelength：
According to the operating center frequency f=1.8GHz of antenna, the permittivity ε of substrate materialr=2.56, to make positive and negative micro-strip
Section is in picturesque disorder, generates suitable transmission mode and radiation mode, needs a ≈ b；Dielectric-slab length L is dielectric-slab width W simultaneously
6.5 times, wherein b is chip unit interval；
Best initial weights obtaining step：Including information determining step, network parameter selection step, learning training algorithms selection step,
Adjustment and obtaining step；
Information determining step：It converts problem to the mode that network can express and the form that can be handled, determines the defeated of each node
Enter output；Each center of a sample is calculated with K mean cluster method, N is center of a sample's number, i.e.,：It first clusters, then by the method for value
The center of a sample for seeking every class is implicit layer unit with required center of a sample；
Network parameter selects step：Determine input, the number of output neuron, the number of plies of multitiered network and hidden neuron number
Mesh, snapshot number are K, and training sample number is A, and wherein A is plural number, is divided into real and imaginary parts calculating, desired output is array element
Plural weight coefficient real and imaginary parts, training sample input be by taking identical as number of training group of direction of arrival to obtain at random
It arrives, plural weight coefficient is obtained using conjugate gradient algorithms；The selection of anticipation error passes through to two different expected error values
Network be trained, therefrom determine a network；
Learning training algorithms selection step：It is first initialized, then radial basis function network is utilized to complete function approximation, determine input
Sample and desired output, wherein the snapshot A of antenna array K × N matrix is set as input sample, and the weight vector ω of each array element is set as
Desired output；Antenna array array number is configured, determines arrival bearing, input vector signal is obtained after normalization, realizes letter
The mapping of number input vector to weight vector；
Adjustment and obtaining step：Constant is extended to clustering method, Hidden nodes, hidden node data center learning coefficient, hidden node
Learning coefficient and hidden node output weights learning coefficient are adjusted, and network connection, which is weighed, after completion training is determined, can be direct
The best initial weights of desired signal are obtained using network；
Antenna array pattern generation step：Antenna array model is verified using the best initial weights of acquisition, to generate day
Linear array directional diagram.
2. the method as described in claim 1, which is characterized in that snapshot number K is set as 200, and training sample number A is set as 800,
Gained input sample data are 1600, and 800 training sample inputs are utilized by taking 800 groups of direction of arrival to obtain at random
Conjugate gradient algorithms obtain 800 groups of plural number weight coefficients, and network is trained with these data.
3. method as claimed in claim 2, which is characterized in that hidden node data center learning coefficient is 0.001, hidden section
Point extension constant learning coefficient is 0.002, and it is 0.005 that hidden node, which exports weights learning coefficient,.
4. method as claimed in claim 3, which is characterized in that the antenna array array number is set as 8, and arrival bearing's setting exists
Between [- π, π].
5. method as claimed in claim 4, which is characterized in that every group of direction of arrival includes a useful signal and three interference
Signal, and Gaussian noise is added in each signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610817682.4A CN106355245B (en) | 2016-09-12 | 2016-09-12 | A kind of Pattern Synthesis of Antenna Array method based on neural network algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610817682.4A CN106355245B (en) | 2016-09-12 | 2016-09-12 | A kind of Pattern Synthesis of Antenna Array method based on neural network algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106355245A CN106355245A (en) | 2017-01-25 |
| CN106355245B true CN106355245B (en) | 2018-09-21 |
Family
ID=57859704
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610817682.4A Active CN106355245B (en) | 2016-09-12 | 2016-09-12 | A kind of Pattern Synthesis of Antenna Array method based on neural network algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106355245B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107729627B (en) * | 2017-09-26 | 2020-12-08 | 西安电子科技大学 | Quick selection method for unit of opportunistic array antenna |
| CN107623691A (en) * | 2017-09-29 | 2018-01-23 | 长沙市智为信息技术有限公司 | A kind of ddos attack detecting system and method based on reverse transmittance nerve network algorithm |
| CN107729694B (en) * | 2017-11-17 | 2020-09-25 | 电子科技大学 | Multi-parameter electromagnetic field modeling simulation method based on neural network |
| CN108957406A (en) * | 2018-05-18 | 2018-12-07 | 电子科技大学 | A kind of radar main lobe interference suppression method neural network based |
| CN109271695B (en) * | 2018-09-07 | 2022-12-23 | 中南大学 | Multi-target antenna design method based on neural network |
| CN109631997B (en) * | 2018-12-20 | 2021-07-09 | 树根互联股份有限公司 | Engineering equipment health detection method and device and electronic equipment |
| CN110427669B (en) * | 2019-07-20 | 2023-04-07 | 中国船舶重工集团公司第七二四研究所 | Neural network model calculation method for phased array scanning radiation beams |
| CN110427590B (en) * | 2019-08-01 | 2022-03-25 | 电子科技大学 | Large sparse array antenna efficient comprehensive method based on adaptive probability learning |
| CN110554352A (en) * | 2019-09-11 | 2019-12-10 | 哈尔滨工业大学 | Method for estimating direction of arrival of interference source of aerospace measurement and control system based on VGG16 neural network |
| CN110596668B (en) * | 2019-09-20 | 2021-06-04 | 中国人民解放军国防科技大学 | Target external radiation source passive positioning method based on reciprocal deep neural network |
| CN111276822B (en) * | 2020-02-24 | 2021-02-19 | 西安电子科技大学 | Antenna pattern main lobe controllable antenna array design method |
| CN112013833B (en) * | 2020-07-23 | 2022-08-05 | 中国人民解放军战略支援部队信息工程大学 | VICTS antenna satellite-to-satellite tracking method based on deep neural network |
| CN112054869B (en) * | 2020-08-05 | 2021-07-23 | 西安电子科技大学 | Wavelength allocation method based on continuous Hopfield neural network |
| CN115622597A (en) * | 2021-07-14 | 2023-01-17 | 中兴通讯股份有限公司 | Beam forming method, device, base station and storage medium |
| CN113777571B (en) * | 2021-08-04 | 2023-08-11 | 中山大学 | Unmanned aerial vehicle cluster dynamic pattern synthesis method based on deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104485513A (en) * | 2014-12-17 | 2015-04-01 | 镇江市高等专科学校 | Real-time integrating method for wideband antenna array |
| CN104517035A (en) * | 2014-12-22 | 2015-04-15 | 中国舰船研究设计中心 | Planar array antenna active scattering directional diagram predication method |
| CN104569907A (en) * | 2014-09-04 | 2015-04-29 | 深圳市金溢科技股份有限公司 | Wireless positioning method and system based on neural network and road side unit |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9818136B1 (en) * | 2003-02-05 | 2017-11-14 | Steven M. Hoffberg | System and method for determining contingent relevance |
-
2016
- 2016-09-12 CN CN201610817682.4A patent/CN106355245B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104569907A (en) * | 2014-09-04 | 2015-04-29 | 深圳市金溢科技股份有限公司 | Wireless positioning method and system based on neural network and road side unit |
| CN104485513A (en) * | 2014-12-17 | 2015-04-01 | 镇江市高等专科学校 | Real-time integrating method for wideband antenna array |
| CN104517035A (en) * | 2014-12-22 | 2015-04-15 | 中国舰船研究设计中心 | Planar array antenna active scattering directional diagram predication method |
Non-Patent Citations (2)
| Title |
|---|
| BP神经网络天线方向图获取方法研究;姜正茂 等;《电子科技大学学报》;20100131;第39卷(第1期);第37-40页 * |
| 基于RBF神经网络的天线阵方向图建模;刘建霞;《太原理工大学学报》;20080131;第39卷(第1期);第36-38、42页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106355245A (en) | 2017-01-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106355245B (en) | A kind of Pattern Synthesis of Antenna Array method based on neural network algorithm | |
| CN106407549B (en) | A kind of Pattern Synthesis of Antenna Array optimization method based on genetic algorithm | |
| Bai et al. | A hybrid IWO/PSO algorithm for pattern synthesis of conformal phased arrays | |
| US8094074B2 (en) | Dynamic radiation pattern antenna system | |
| Zeng et al. | Intelligent omni-surfaces: Reflection-refraction circuit model, full-dimensional beamforming, and system implementation | |
| Liang et al. | Design of license plate RFID tag antenna using characteristic mode pattern synthesis | |
| CN113093094B (en) | Intelligent incident wave direction detection method based on phase regulation and control super surface | |
| Li et al. | Reconfigurable intelligent surfaces 2.0: Beyond diagonal phase shift matrices | |
| Roshna et al. | Compact UWB MIMO antenna for tridirectional pattern diversity characteristics | |
| CN104102775B (en) | A kind of electricity based on Sidelobe Suppression adjusts passive antenna beam optimization method | |
| Janapala et al. | Metasurface based pattern reconfigurable antenna for 2.45 GHz ISM band applications | |
| Chen et al. | Channel customization for joint Tx-RISs-Rx design in hybrid mmWave systems | |
| CN105510872B (en) | A kind of assay method of two-dimentional direction of arrival suitable for extensive mimo system | |
| Tang et al. | Design of 5G dual-antenna passive repeater based on machine learning | |
| CN106301632A (en) | A kind of method of rapid evaluation multiaerial system channel capacity stability | |
| Kao et al. | AI-aided 3-D beamforming for millimeter wave communications | |
| CN108429003A (en) | A kind of conformal paster antenna of taper of producible vortex electromagnetic wave | |
| Ma et al. | Data-driven beam tracking for mobile millimeter-wave communication systems without channel estimation | |
| Chen et al. | Razor-sharp Narrowbeam Communications and Management for Terahertz Wireless Networks | |
| Liu et al. | Low‐cost intelligent antenna with low profile and broad bandwidth | |
| CN110532631A (en) | Method is determined based on the 6G communication antenna element position tolerance of channel capacity sensitivity | |
| Zhang et al. | Multi-domain extended ELM-DOA subarray localization method based on distributed coherent aperture radar | |
| Patnaik et al. | Modeling frequency reconfigurable antenna array using neural networks | |
| Rastegari et al. | Analogue beamforming solution for beam‐alignment problem | |
| Wang et al. | Antenna Evolution for Massive MIMO |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |