CN108287941A - A kind of RF power amplification temperature characterisitic modeling method based on DNN - Google Patents
A kind of RF power amplification temperature characterisitic modeling method based on DNN Download PDFInfo
- Publication number
- CN108287941A CN108287941A CN201711409314.7A CN201711409314A CN108287941A CN 108287941 A CN108287941 A CN 108287941A CN 201711409314 A CN201711409314 A CN 201711409314A CN 108287941 A CN108287941 A CN 108287941A
- Authority
- CN
- China
- Prior art keywords
- dnn
- dnn models
- models
- test
- training
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/36—Circuit design at the analogue level
- G06F30/367—Design verification, e.g. using simulation, simulation program with integrated circuit emphasis [SPICE], direct methods or relaxation methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/36—Circuit design at the analogue level
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/10—Interfaces, programming languages or software development kits, e.g. for simulating neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
Landscapes
- Engineering & Computer Science (AREA)
- Computer Hardware Design (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Geometry (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Amplifiers (AREA)
Abstract
The RF power amplification temperature characterisitic modeling method based on DNN that the invention discloses a kind of：The training data and test data of DNN models are chosen according to the measured data of radio-frequency power amplifier temperature characterisitic；Determine the input variable and output variable of DNN models；Training data is imported in DNN models, DNN models are trained；Test data is imported in trained DNN models, DNN models output result and test result and the error for calculating the two are compared；Compare the size of test error MSE and DNN the model accuracy desired value of DNN models, if MSE is less than precision desired value, training is completed；If MSE is more than precision desired value, adjusting parameter re -training, until MSE is less than precision desired value, training terminates.The present invention establishes correlation model of the radio-frequency power amplifier performance indicator about temperature, realizes the prediction to radio-frequency power amplifier performance indicator situation of change in given temperature section.
Description
Technical field
The invention belongs to radio-frequency power amplifier temperature characterisitics to model field, and more specifically, it relates to a kind of bases
In the RF power amplification temperature characterisitic modeling method of DNN.
Background technology
Radio-frequency power amplifier is the indispensable core circuit module of radio-frequency front-end, and the quality of performance directly determines
The good and the bad of performance in wireless communication systems.The performance of radio-frequency power amplifier is influenced by temperature huge, and the defect is in actual work
It can not be avoided during work.Literature survey the result shows that：Not yet establish complete device temperature model at present, cause it is designed go out
The radio-frequency power amplifier come can only be met the requirements within the scope of a certain specific temperature.When the work temperature of radio-frequency power amplifier
Degree change (especially temperature generation wide variation) when, the characteristic of radio-frequency power amplifier will become therewith
Change, and this will greatly influence the performance and working condition of entire communication system.Therefore, it is necessary to what is completed for a design to penetrate
Frequency power amplifier characterized about the behavior of temperature characterisitic, and to realize, radio-frequency power is put under a wide range of temperature variations
The prediction of big device characteristic.
Predict the characteristic of the radio-frequency power amplifier under a wide range of temperature variations, this just needs to carry out a large amount of
Actual test, and actual test process is extremely time-consuming, and also actual test point is unlikely to be complete continuous covering again
All temperature.Therefore, in the case where predicting a wide range of temperature variations, the characteristic of radio-frequency power amplifier just faces one urgently
Demand：The characteristic of the radio-frequency power amplifier in entire temperature range is characterized according to a small number of crucial test points.
Invention content
Purpose of the invention is to overcome the shortcomings in the prior art, provides a kind of RF power amplification temperature based on DNN
Characteristics modeling method is spent, based on the measured data of radio-frequency power amplifier temperature characterisitic, using Deep Neural Network
(DNN), establish radio-frequency power amplifier performance indicator (such as：S parameter, output power, PAE etc.) relevant mode about temperature
Type realizes the prediction to radio-frequency power amplifier performance indicator situation of change in given temperature section.
The purpose of the present invention is what is be achieved through the following technical solutions.
The RF power amplification temperature characterisitic modeling method based on DNN of the present invention, includes the following steps：
Step 1 chooses the training of DNN models according to the measured data of the radio-frequency power amplifier temperature characterisitic obtained
Data and test data；
Step 2, determine DNN models input variable (input power, frequency and temperature) and output variable (output power,
S parameter and PAE)；
Training data is imported in DNN models, is trained to DNN models by step 3；
Step 4 imports test data in trained DNN models, compares DNN models output result and test
As a result and the error that both calculates, i.e. the test error MSE of DNN models：
Wherein, miIndicate the test result of DNN models, i.e., the ideal output of DNN models is as a result, oiIndicate the defeated of DNN models
Go out as a result, i.e. the reality output of DNN models is as a result, n indicates sample size；
Step 5 compares the size of test error MSE and DNN the model accuracy desired value of DNN models, if test error
MSE is less than precision desired value, then DNN model trainings are completed；If test error MSE is more than precision desired value, need to pass through
Adjusting parameter (excitation function F (￡), hiding number of plies L and hidden neuron number N) carries out re -training, until MSE is less than essence
Desired value is spent, training terminates.
Compared with prior art, advantageous effect caused by technical scheme of the present invention is：
DNN models are applied to the modeling of radio-frequency power amplifier temperature characterisitic by the present invention for the first time, can be met and be closed by minority
The test point of key characterizes the current demand of radio-frequency power amplifier characteristic in entire temperature range, has filled up radio-frequency power amplification
Device temperature characterisitic models the blank in field.Compared to traditional SNN, DNN due to more levels as a result, thus to things
Modeling or abstraction ability are stronger, thus energy analog radio frequency power amplifier has compared with strong nonlinearity characteristic and multi input in this way
The complex model of multi output variable.The model will first do unsupervised learning before doing supervised learning, then by unsupervised learning
The weights acquired are trained as the initial value of supervised learning, contribute to the risk for reducing over-fitting in this way.
In addition, the model and its modeling procedure may be equally applied to other RF/Microwave Integrated Circuits temperature characterisitics analysis and
Prediction has very strong practical guided significance to the design of actual radio frequency/microwave circuit.
Description of the drawings
Fig. 1 is the modeling procedure schematic diagram based on DNN；
Fig. 2 is radio-frequency power amplifier temperature property test circuit diagram；
Fig. 3 is radio-frequency power amplifier output power and the modeling result schematic diagram of PAE；
Fig. 4 is the modeling result schematic diagram of radio-frequency power amplifier S21 and S22.
Specific implementation mode
The invention will be further described below in conjunction with the accompanying drawings.
The RF power amplification temperature characterisitic modeling method based on DNN of the present invention, as shown in Figure 1, including the following steps：
Step 1 chooses the training of DNN models according to the measured data of the radio-frequency power amplifier temperature characterisitic obtained
Data and test data；
Step 2, determine DNN models input variable (input power, frequency and temperature) and output variable (output power,
S parameter and PAE)；
Training data is imported in DNN models, is trained to DNN models by step 3；
Step 4 imports test data in trained DNN models, compares DNN models output result and test
As a result and the error that both calculates, i.e. the test error MSE of DNN models：
Wherein, miIndicate the test result of DNN models, i.e., the ideal output of DNN models is as a result, oiIndicate the defeated of DNN models
Go out as a result, i.e. the reality output of DNN models is as a result, n indicates sample size；
Step 5 compares the size of test error MSE and DNN the model accuracy desired value of DNN models, if test error
MSE is less than precision desired value, then DNN model trainings are completed；If test error MSE is more than precision desired value, need to pass through
Adjusting parameter (excitation function F (￡), hiding number of plies L and hidden neuron number N) carries out re -training, until MSE is less than essence
Desired value is spent, training terminates.
Embodiment：0.1-1.2GHzCMOS radio-frequency power amplifiers
Step 1 is carried out using environment testing case and is tested, and Range of measuring temp is -25~125 DEG C, and supply voltage is
3.3V, scan frequency are 100MHz to 1.2GHz, and measured data, test circuit are recorded using vector network analyzer and frequency spectrograph
Figure is as shown in Figure 2.
According to the measured data of the radio-frequency power amplifier temperature characterisitic obtained, 2 groups are splitted data into, respectively as
The training data and test data of DNN models.
Step 2, determine DNN models input variable (input power, frequency and temperature) and output variable (output power,
S parameter and PAE).
Selected training data is imported in DNN models, is trained to DNN models by step 3.
Step 4 imports selected test data in trained DNN models, compares DNN models output knot
Fruit and test result and the error for calculating the two, i.e. the test error MSE of DNN models are calculated, herein n=156 by formula (1).
Step 5 compares the test error MSE and DNN model accuracy desired value (10 of DNN models-3) size, if survey
It tries error MSE and is less than precision desired value (10-3), then DNN model trainings are completed；If test error MSE is more than precision desired value
(10-3), then it needs to be instructed again by adjusting parameter (excitation function F (￡), hiding number of plies L and hidden neuron number N)
Practice, until MSE is less than precision desired value (10-3), training terminates.
In the present embodiment, selected DNN models, it is 5 to hide layer parameter, and hidden neuron number is 25, and test misses
Poor MSE=4.449 × 10-4.The RF power amplification temperatures model based on DNN is drawn, as shown in Figures 3 and 4.It can from Fig. 3 and 4
To find out, DNN models and measured result substantially completely coincide, and illustrate that the DNN models can effectively represent 0.1-completely
1.2GHz CMOS radio-frequency power amplifiers output and input between non-linear relation, may be implemented completely to given temperature area
The prediction of interior radio-frequency power amplifier performance indicator situation of change.
In conclusion the data based on a small number of key testpoints, and DNN models are applied, it is entire that characterization may be implemented completely
The characteristic of radio-frequency power amplifier in temperature range, and can greatly shorten the testing time, to realize to a wide range of temperature
The quick analysis and prediction of radio-frequency power amplifier temperature characterisitic under situation of change.
Although the function and the course of work of the present invention are described above in conjunction with attached drawing, the invention is not limited in
Above-mentioned concrete function and the course of work, the above mentioned embodiment is only schematical, rather than restrictive, ability
The those of ordinary skill in domain under the inspiration of the present invention, is not departing from present inventive concept and scope of the claimed protection situation
Under, many forms can also be made, all of these belong to the protection of the present invention.
Claims (1)
1. a kind of RF power amplification temperature characterisitic modeling method based on DNN, which is characterized in that include the following steps：
Step 1 chooses the training data of DNN models according to the measured data of the radio-frequency power amplifier temperature characterisitic obtained
And test data；
Step 2 determines the input variable (input power, frequency and temperature) and output variable (output power, S ginsengs of DNN models
Number and PAE)；
Training data is imported in DNN models, is trained to DNN models by step 3；
Step 4 imports test data in trained DNN models, compares DNN models output result and test result
And calculate the error of the two, i.e. the test error MSE of DNN models：
Wherein, miIndicate the test result of DNN models, i.e., the ideal output of DNN models is as a result, oiIndicate the output knot of DNN models
Fruit, the i.e. reality output of DNN models are as a result, n indicates sample size；
Step 5 compares the size of test error MSE and DNN the model accuracy desired value of DNN models, if test error MSE
Less than precision desired value, then DNN model trainings completion；If test error MSE be more than precision desired value, need by adjusting
Parameter (excitation function F (￡), hiding number of plies L and hidden neuron number N) carries out re -training, until MSE is less than the precision phase
Prestige value, training terminate.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711409314.7A CN108287941A (en) | 2017-12-22 | 2017-12-22 | A kind of RF power amplification temperature characterisitic modeling method based on DNN |
| LU100964A LU100964B1 (en) | 2017-12-22 | 2018-10-10 | A DNN-based method for modeling temperature characteristics of radio frequency power amplifier |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711409314.7A CN108287941A (en) | 2017-12-22 | 2017-12-22 | A kind of RF power amplification temperature characterisitic modeling method based on DNN |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108287941A true CN108287941A (en) | 2018-07-17 |
Family
ID=62832388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711409314.7A Pending CN108287941A (en) | 2017-12-22 | 2017-12-22 | A kind of RF power amplification temperature characterisitic modeling method based on DNN |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN108287941A (en) |
| LU (1) | LU100964B1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110907704A (en) * | 2018-09-14 | 2020-03-24 | 天津大学青岛海洋技术研究院 | Method for extracting unique values of microwave complex dielectric constant and complex permeability of material |
-
2017
- 2017-12-22 CN CN201711409314.7A patent/CN108287941A/en active Pending
-
2018
- 2018-10-10 LU LU100964A patent/LU100964B1/en active IP Right Grant
Non-Patent Citations (4)
| Title |
|---|
| CHENG-YU ZHANG等: "Extreme learning machine for the behavioral modeling of RF power amplifiers", 《2017 IEEE MTT-S INTERNATIONAL MICROWAVE SYMPOSIUM (IMS)》 * |
| QIAN LIN等: "Interconnect Reliability Analysis for Power Amplifier Based on Artificial Neural Networks", 《JOURNAL OF ELECTRONIC TESTING》 * |
| 林倩等: "宽带功率放大器温度可靠性研究", 《天津理工大学学报》 * |
| 王祯祥等: "GaN HEMT高效功率放大器电路温度特性研究", 《南开大学学报（自然科学版）》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110907704A (en) * | 2018-09-14 | 2020-03-24 | 天津大学青岛海洋技术研究院 | Method for extracting unique values of microwave complex dielectric constant and complex permeability of material |
Also Published As
| Publication number | Publication date |
|---|---|
| LU100964B1 (en) | 2019-06-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107977535A (en) | A kind of RF power amplification temperature characterisitic modeling method based on ELM | |
| Li et al. | MagNet: An open-source database for data-driven magnetic core loss modeling | |
| CN105184026B (en) | A kind of the Building of Simulation Model method and emulation mode of MOS variable capacitances | |
| Cai et al. | Padé-approximation-based behavioral modeling | |
| CN101256214A (en) | Self-adapting testing system and method for radio-frequency power amplifier | |
| CN109142865B (en) | Frequency domain spectrum identification method considering polarization equivalent circuit parameters of oiled paper insulation interface | |
| CN108549767B (en) | Neural network space mapping modeling method for large-signal power transistor | |
| Andraud et al. | One-shot calibration of RF circuits based on non-intrusive sensors | |
| CN105426921B (en) | A kind of RFID label antenna optimization method | |
| CN110163430A (en) | Concrete material Prediction of compressive strength method based on AdaBoost algorithm | |
| CN108734268A (en) | Compression method and device, terminal, the storage medium of deep neural network model | |
| CN108768550A (en) | Wide-band transmitter nonlinear modeling method based on dynamic multi-core bandwidth generalized regression neural network algorithm | |
| CN108268700A (en) | A kind of RF power amplification temperature characterisitic modeling method based on BPNN | |
| Karra et al. | Prediction of received signal power in mobile communications using different machine learning algorithms: A comparative study | |
| CN102867104B (en) | A kind of quantitative analysis device of electromagnetic environmental impact factor of electronic equipment and method | |
| CN108287941A (en) | A kind of RF power amplification temperature characterisitic modeling method based on DNN | |
| Xiao et al. | Artificial neural network with data mining techniques for antenna design | |
| Cai et al. | A new nonlinear behavioral modeling technique for RF power transistors based on Bayesian inference | |
| CN111898320A (en) | Radio frequency power amplifier scattering parameter extraction method and device based on neural network | |
| Geng et al. | Accurate and effective nonlinear behavioral modeling of a 10-W GaN HEMT based on LSTM neural networks | |
| CN104008292B (en) | Broad-band antenna super-broadband electromagnetic impulse response prediction method | |
| CN108446451A (en) | A kind of low-noise amplifier temperature characterisitic modeling method based on ELM | |
| CN111880003B (en) | Method for extracting harmonic balance parameters of radio frequency power amplifier based on neural network | |
| Wang et al. | A comparison of second‐order derivative based models for time domain reflectometry waveform analysis | |
| Lin et al. | X-parameter modeling investigation for microwave power devices |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180717 |
|
| RJ01 | Rejection of invention patent application after publication |