CN110427669A - A kind of neural network model calculation method of phase-array scanning radiation beam - Google Patents

A kind of neural network model calculation method of phase-array scanning radiation beam Download PDF

Info

Publication number
CN110427669A
CN110427669A CN201910657700.0A CN201910657700A CN110427669A CN 110427669 A CN110427669 A CN 110427669A CN 201910657700 A CN201910657700 A CN 201910657700A CN 110427669 A CN110427669 A CN 110427669A
Authority
CN
China
Prior art keywords
neural network
network model
array
output
radiation beam
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910657700.0A
Other languages
Chinese (zh)
Other versions
CN110427669B (en
Inventor
温亚庆
周雷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
724th Research Institute of CSIC
Original Assignee
724th Research Institute of CSIC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 724th Research Institute of CSIC filed Critical 724th Research Institute of CSIC
Priority to CN201910657700.0A priority Critical patent/CN110427669B/en
Publication of CN110427669A publication Critical patent/CN110427669A/en
Application granted granted Critical
Publication of CN110427669B publication Critical patent/CN110427669B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Computational Mathematics (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Radar Systems Or Details Thereof (AREA)
  • Variable-Direction Aerials And Aerial Arrays (AREA)

Abstract

The present invention relates to a kind of neural network model calculation methods of phase-array scanning radiation beam, it mainly include the following contents: building feedback neural network model, neural network model training is participated in using the array beams calculated result obtained as fixed input value, the neural network model of training is completed as phased array radiation beam calculation, is verified with model of the similarity function to building.There is accuracy and a wide range of spreadability to the calculating of phased array radiation beam in the present invention, improve computational efficiency and accuracy in phased array beam analysis.

Description

A kind of neural network model calculation method of phase-array scanning radiation beam
Technical field
The present invention relates to a kind of calculating of phased array beam, more particularly to the calculation method of airspace scanning radiation beam.
Background technique
The beam quality of Planar Phased Array Antenna can be gradually reduced with the increase of scanning angle, be mainly reflected in wave beam In the stability of gain and the size of secondary lobe.Accurately calculating with phase-array scanning beam pattern can be in design exposure early period These problems, convenient for being effectively prevented from and solving in engineering practice.It is main in the calculating of phase-array scanning beam pattern Difficult point the considerations of being between array element stiffness of coupling and the different scattering environments of each array element, these conditions are difficult in array factor It is embodied in a manner of parsing in the calculating of product theorem, to affect the accuracy of calculating.It is big especially for having The Planar Phased Array Antenna of angle scanning range, above-mentioned factor will lead to array radiation beam pattern side penetrate region have compared with High-accuracy, and computational accuracy can sharp fall in low elevation coverage.
Present invention is generally directed to the Planar Phased Array Antennas with wide-angle scan characteristic, to sweeping in entire scanning area Wave beam introducing Neural Network model predictive method is retouched accurately to be calculated.
Summary of the invention
The purpose of the present invention is to provide the methods that a kind of radiation beam of adaptation planar big-angle scanning phased array calculates.
Realize technical solution of the invention are as follows: choose the input value substitution array all-wave mould that K group feeds amplitude at random In type, phase difference selects step increments according to practical phase shifter, corresponds in the case of calculating fixed skew with full-wave simulation Precise irradiation output.Elman formula feedback neural network model is established, initial input value is added in input layer, mainly includes Port current feed phase and amplitude information.Network model includes k hidden layer and final output layer, establishes neuronal function letter in layer It counts, is Sigmoid function in usual hidden layer, be linear function in output layer.Input information and output letter in hidden layer Feedback information and feedback operator are added between breath, construct monocycle feedback system.It will complete the feed amplitude input value of all-wave calculating It is substituted into neural network model with radiation beam output valve, is trained using conjugate gradient algorithms, takes precision of prediction to or reach Training is terminated after to preset times.After completing Establishment of Neural Model, stochastic inputs I (N) is chosen again, using full wave model Emulation mode and neural net model method calculate separately array radiation patterns output value matrix.The two similitude is calculated, if Precision cutoff value is set, neural network model accuracy in computations is verified.
The present invention is characterized in that:
(1) scanning beam of phased array is calculated using neural network model, without additional addition array element in calculating Coupling influence, it is more convenient accurate that the planar phased array scanning beam of wide-angle is calculated；
(2) after the I/O data base of full wave model is established, universal computer model is generated by neural network model, Computational accuracy is similar with full-wave simulation, while calculating the time and being greatly reduced；
(3) using the input range distribution of quantization in calculating process, phase value is set for the stepping of practical phase shifter It sets, easily facilitates practical engineering application.
Detailed description of the invention
Fig. 1 is Elman neural network model schematic diagram.
Fig. 2 is neural network model optimized flow chart.
Fig. 3 is that same magnitude is distributed the phase array method of full-wave simulation acquisition to radiation beam pattern.
Fig. 4 is that same magnitude is distributed lower neural network model calculating phase array method to radiation beam pattern.
Fig. 5 is the low elevation angle direction radiation beam pattern of phased array that same magnitude is distributed that lower full-wave simulation obtains.
Fig. 6 is that same magnitude is distributed the low elevation angle direction radiation beam pattern of lower neural network model calculating phased array.
Specific embodiment
The present invention is illustrated using eight yuan of phased array antenna as embodiment.Array element uses doublet unit, array number totally 8 A, array element spacing is 0.5 λ.Detailed process is as follows:
Step 1: establishing neuronal function function, and when handling neuron, the information from other neurons is xi, it Connection weight between neuron processed be wi, then the input of processed neuron are as follows:
The output of neuron processed are as follows:
Wherein, f is that neuronal function function is also referred to as activation primitive, it determines the output of neuron.θ represents hidden layer The threshold value of interior neurode.Sigmoid function (tanh is increased continuously function) is used in hidden layer:
Linear function is used in output layer:
F (x)=x；
Step 2: establishing the type of attachment between neuron, using monocycle feedback system, inputs information xi(n), feedback letter Cease x 'i(n), output information yi(n).Assuming that system is linearly, to be made of forward direction operator f and feedback operator g, output relation are as follows:
yi(n)=f (x 'i(n))；
x′i(n)=xi(n)+g(yi(n))；
To obtain output information are as follows:
Wherein, f is time delay operator z using a fixed weight w, g-1, therefore this monocycle operator can be write At:
Last is available with Maclaurin expansion:
Wherein,
z-k(xi(n))=xi(n-k)；
Step 3: in the neural network model of eight yuan of microstrip dipole phased arrays, the feed Amplitude Ratio of each port is counted as I (N)=[i1, i2 ..., i8].Cheng Qian is crossed in Establishment of Neural Model, the feed Amplitude Ratio that shared K group randomly selects is as solid Determine the training that input value participates in neural network.In every group of feed, phase difference chooses the different fixations of six bit digital phase shifters Value makes array radiation wave beam achieve the effect that scanning.Its fixed skew is followed successively by 0 °, and ± 11.25 °, ± 22.5 °, ± 45 °, ± 90 °, ± 123.75 °, ± 151.875 ° and ± 174.375 °.By K group different port feed Amplitude Ratio under out of phase difference Array radiation patterns carry out full-wave simulation calculating, obtain output valve corresponding to every group of input.
In Elman neural network model, input layer is the feed Amplitude Ratio matrix of eight feed ports, network model In include three hidden layers and final output layer.Activation primitive in hidden layer is Sigmoid function, the activation in output layer Function is linear function.Conjugate gradient algorithms (quantitative conjugate gradient) is used in training method, it will It outputs and inputs and is substituted into neural metwork training known to above-mentioned K group, training rate is el.It is neural after M circuit training The precision of prediction of network model reaches er；
Step 4: the input value I of the new port feed Amplitude Ratio of Kn group is regeneratedn(N), full-wave simulation method is respectively adopted Corresponding Kn group phased array beam antenna pattern output valve Ep and Eq are calculated with neural network model:
The similarity of the two output value matrix is calculated, and sets similarity threshold ri, the value is set as 0.85 in this example.
If every group of similarity calculation result is all larger than default threshold, neural network model constructs achievement；
If at least one group similarity calculation result is less than default threshold, return step three re-starts instruction to model Practice.
With the neural network model built to eight yuan of phase array methods to radiation beam pattern and low elevation radiation wave Beam directional diagram is calculated, and result compares almost the same with full-wave simulation.
Therefore the present invention can accurately calculate the radiation beam pattern in phased array wide-angle scanning range.

Claims (1)

1. a kind of neural network model calculation method of phase-array scanning radiation beam, it is characterised in that:
Step 1: establishing neural network model neuronal function function, and the information from other neurons is xi, with mind processed It is w through the connection weight between memberi, then the input of processed neuron are as follows:
The output of neuron processed are as follows:
Step 2: establishing the type of attachment between neuron, using monocycle feedback system, inputs information xi(n), feedback information x 'i (n), output information yi(n)；Assuming that system is linearly, to be made of forward direction operator f and feedback operator g, output relation are as follows:
yi(n)=f (x 'i(n))；
x’i(n)=xi(n)+g(yi(n))；
To obtain output information are as follows:
Step 3: neural network model training process is established: firstly, it is random defeated to randomly select K group to the Array Model pre-seted Enter I (N), the corresponding array radiation patterns output valve of every group of input is calculated using full-wave simulation mode；Then, by known K Group input and output value is substituted into neural network model using conjugate gradient algorithms, and predetermined training rate is arranged, it is contemplated that precision and is followed Ring number completes Establishment of Neural Model；
Step 4: matrix similarity function is established:
To newly generated stochastic inputs I (N), array is calculated separately using full wave model emulation mode and neural net model method Antenna pattern output valve, the matrix form for being calculated as identical scale calculate the two similitude with similarity function, precision are arranged Cutoff value verifies neural network model accuracy in computations.
CN201910657700.0A 2019-07-20 2019-07-20 Neural network model calculation method for phased array scanning radiation beams Active CN110427669B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910657700.0A CN110427669B (en) 2019-07-20 2019-07-20 Neural network model calculation method for phased array scanning radiation beams

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910657700.0A CN110427669B (en) 2019-07-20 2019-07-20 Neural network model calculation method for phased array scanning radiation beams

Publications (2)

Publication Number Publication Date
CN110427669A true CN110427669A (en) 2019-11-08
CN110427669B CN110427669B (en) 2023-04-07

Family

ID=68411381

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910657700.0A Active CN110427669B (en) 2019-07-20 2019-07-20 Neural network model calculation method for phased array scanning radiation beams

Country Status (1)

Country Link
CN (1) CN110427669B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112964941A (en) * 2021-03-24 2021-06-15 中山大学 Phased array antenna test method, device, equipment and medium
WO2024060132A1 (en) * 2022-09-22 2024-03-28 北京京东方技术开发有限公司 Tunable antenna control method and apparatus and tunable antenna system

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106355245A (en) * 2016-09-12 2017-01-25 哈尔滨工业大学 Method for integrating array antenna directional images on basis of neural network algorithms
CN106597391A (en) * 2016-11-25 2017-04-26 中国电子科技集团公司第三十八研究所 All-weather monitoring method for phased array antenna array plane deformation
CN107067077A (en) * 2017-04-18 2017-08-18 武汉大学 A kind of weighting algorithm of convolutional neural networks
CN107609185A (en) * 2017-09-30 2018-01-19 百度在线网络技术（北京）有限公司 Method, apparatus, equipment and computer-readable recording medium for POI Similarity Measure
CN109670580A (en) * 2018-12-21 2019-04-23 浙江工业大学 A kind of data recovery method based on time series

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106355245A (en) * 2016-09-12 2017-01-25 哈尔滨工业大学 Method for integrating array antenna directional images on basis of neural network algorithms
CN106597391A (en) * 2016-11-25 2017-04-26 中国电子科技集团公司第三十八研究所 All-weather monitoring method for phased array antenna array plane deformation
CN107067077A (en) * 2017-04-18 2017-08-18 武汉大学 A kind of weighting algorithm of convolutional neural networks
CN107609185A (en) * 2017-09-30 2018-01-19 百度在线网络技术（北京）有限公司 Method, apparatus, equipment and computer-readable recording medium for POI Similarity Measure
CN109670580A (en) * 2018-12-21 2019-04-23 浙江工业大学 A kind of data recovery method based on time series

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112964941A (en) * 2021-03-24 2021-06-15 中山大学 Phased array antenna test method, device, equipment and medium
CN112964941B (en) * 2021-03-24 2021-12-21 中山大学 Phased array antenna test method, device, equipment and medium
WO2024060132A1 (en) * 2022-09-22 2024-03-28 北京京东方技术开发有限公司 Tunable antenna control method and apparatus and tunable antenna system

Also Published As

Publication number Publication date
CN110427669B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
CN108919199B (en) Side lobe suppression method of multi-beam imaging sonar sparse array and array sparse method
CN105426578B (en) A kind of MIMO-SAR planar array element position optimization methods based on genetic algorithm
CN107944133B (en) Annular antenna array sparse method based on multi-target quantum spider swarm evolution mechanism
CN112511209B (en) Beam control method, system and controller of array antenna based on frequency diversity
CN106850016B (en) Phase-only weighted array antenna beam forming optimization method based on MIFT and CP mixed algorithm
CN111313158B (en) Method for thinning circular array
CN111160556B (en) Array sparse optimization method based on adaptive genetic algorithm
CN109885872A (en) A kind of uniform surface battle array sparse optimization method based on differential evolution algorithm
CN104392034A (en) L1/2-norm-based sparse linear array optimization method
CN106295122A (en) A kind of sane zero falls into broadening Adaptive beamformer method
CN109992845B (en) Beam control and array pattern optimization method based on MATLAB and HFSS
CN107658573B (en) Circular array directional diagram synthesis method based on quantum searcher search mechanism
CN110427669A (en) A kind of neural network model calculation method of phase-array scanning radiation beam
CN113311397B (en) Large array rapid self-adaptive anti-interference method based on convolutional neural network
CN104020448A (en) Optimized formation method of radar subarray-level sum/difference beams constrained by equal array elements
CN105842666B (en) Radar Subarray partition optimization method based on difference algorithm
CN111896929B (en) DOD/DOA estimation algorithm of non-uniform MIMO radar
CN111262612B (en) Array shaped beam analytic synthesis method based on pencil beam
CN104950290A (en) Large-scale phased-array antenna sub array division method based on weighted K average value clustering
CN116882147A (en) Array antenna synthesis method based on mixed sparrow search algorithm
CN115146544A (en) Array antenna design method adopting knowledge and data hybrid driving
CN112711852B (en) Antenna array amplitude-phase distribution design method of plane wave generator
Senthilkumar et al. Accuracy of Perceptron based beamforming for embedded smart and MIMO antennas
CN110705680B (en) Subarray position optimization method and system of distributed array
CN112114289B (en) Optimization method of minimum array element number sparse array based on frame theory

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