CN108768550A - Wide-band transmitter nonlinear modeling method based on dynamic multi-core bandwidth generalized regression neural network algorithm - Google Patents
Wide-band transmitter nonlinear modeling method based on dynamic multi-core bandwidth generalized regression neural network algorithm Download PDFInfo
- Publication number
- CN108768550A CN108768550A CN201810648237.9A CN201810648237A CN108768550A CN 108768550 A CN108768550 A CN 108768550A CN 201810648237 A CN201810648237 A CN 201810648237A CN 108768550 A CN108768550 A CN 108768550A
- Authority
- CN
- China
- Prior art keywords
- amplitude
- signal
- fingerprint
- frequency
- dynamic
- 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
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/10—Monitoring; Testing of transmitters
- H04B17/11—Monitoring; Testing of transmitters for calibration
- H04B17/13—Monitoring; Testing of transmitters for calibration of power amplifiers, e.g. gain or non-linearity
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/0082—Monitoring; Testing using service channels; using auxiliary channels
- H04B17/0087—Monitoring; Testing using service channels; using auxiliary channels using auxiliary channels or channel simulators
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/10—Monitoring; Testing of transmitters
- H04B17/11—Monitoring; Testing of transmitters for calibration
- H04B17/12—Monitoring; Testing of transmitters for calibration of transmit antennas, e.g. of the amplitude or phase
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/02—Transmitters
- H04B1/04—Circuits
- H04B2001/0408—Circuits with power amplifiers
- H04B2001/0425—Circuits with power amplifiers with linearisation using predistortion
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Electromagnetism (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Nonlinear Science (AREA)
- Transmitters (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a wide-band transmitter nonlinear modeling method based on a dynamic multi-core bandwidth generalized regression neural network algorithm, which comprises the following steps: building a test platform, measuring and recording the nonlinear characteristics of the full working frequency band of the wide-frequency-band transmitter, and collecting the input and output amplitudes and phase values of a test signal and the corresponding carrier frequency; carrying out segmentation processing on the signal sample according to carrier frequency, quantizing the amplitude of an input signal by using an amplitude non-uniform quantization technology, and realizing the construction of a dynamic non-linear fingerprint sample data set by using a dynamic memory fingerprint technology; and training the generalized regression neural network by using the dynamic nonlinear fingerprint sample data set, and combining an optimization algorithm to obtain a dynamic multi-core bandwidth generalized regression neural network model so as to realize modeling of full-band nonlinear characteristics of the wideband transmitter.
Description
Technical field
The present invention relates to sender design fields in wireless communication, especially sender modeling nonlinear distortions and school
Positive technology.By measuring signal of the wide-band sender when full frequency band works, feature mode composition training sample therein is extracted
Notebook data is fitted the nonlinear characteristic of sender using learning algorithm and neural network model, is sender predistortion correction
Equal linearization techniques provide model basis.
Background technology
The non-linear distortion of wide-band sender is mainly caused by the nonlinear characteristic of RF power amplification.It is mainly shown as signal
Amplitude distortion and phase distortion:After signal is by nonlinear power amplifier, the gain of amplitude output signal is with input signal amplitude
And change, while the knots modification of phase also changes with input signal amplitude.Non-linear distortion not only results in amplified signal
Compression, also will produce new frequency content, leads to the extension of frequency spectrum, not only seriously affected the noise of signal in working frequency range
Than also generating interference to neighbouring channel, especially having larger harm to the communication system of networking type.
The non-linear of sender can be usually distorted by AM-AM and AM-PM be distorted to indicate.Wherein AM-AM indicate by
The amplitude distortion generated after signal amplification, AM-PM indicate the phase distortion generated after being amplified by signal.
Other than AM-AM distortions and AM-PM distortions, there is also memory effects for RF power amplification.In terms of time domain angle, memory
Effect is that the current output signal of RF power amplification depends not only on current input signal, but also has with past input signal
It closes, i.e. the distortion of device depends on the radio frequency power output of 10 to 20 nanoseconds before.It is substantially instantaneous in device channel
Temperature can influence to be distorted;From the angle of frequency domain, memory effect can be defined as RF power amplification amplitude and phase characteristic with defeated
The phenomenon that entering the variation of signal envelope frequency and changing.The maximum source of memory effect is since the variation of signal envelope frequency is led
Matching network, the variation of biasing networks node impedance is caused to bring the fluctuation of output amplitude and phase.So with communication bandwidth
Increase, the memory effect of RF power amplification can be more and more significant, needs to add on the basis of memoryless RF power amplification nonlinear model
Enter the description to memory effect to have established memory RF power amplification model, complexity is directly proportional to the depth of memory effect.
Since the working frequency range of wide-band sender is very wide, sometimes across multiple octaves.By it is practical measure it is found that
Under the pattern, the nonlinear characteristic of sender can be significantly changed as frequency changes, and the size of variable quantity not with frequently
The variable quantity of rate is at monotone variation relationship, and wideband operation can also bring serious memory effect, so being difficult to simple
Mathematic(al) representation describes the nonlinear characteristic of wide-band sender.
Generalized regression nerve networks (General Regression Neural Network GRNN) model is a kind of base
In the kernel regression mathematical model of density estimation.Classical general regression neural network by input layer, mode layer, summation layer and
Output layer forms.Its kernel function obeys multivariate Gaussian distribution, has good nonlinear fitting performance and noise resisting ability.Its core
Estimated value such as following formula
Wherein σ is the bandwidth of kernel function, and Xi is observation sample value, and yi is output valve, and X is input value.Estimated value can recognize
To be the weighted average of all observation sample values.
Invention content
Goal of the invention:In order to overcome the deficiencies in the prior art, the present invention to provide a kind of for being transmitted to wide-band
The modeling method of machine nonlinear characteristic, thought of this method based on machine learning, passes through the side of discovery learning and incremental learning
Method is extracted and is screened the nonlinear characteristic pattern of wide-band sender, utilizes neural network algorithm training pattern, model structure
Flexibly, need trained parameter less, fitting performance is good, and noise resisting ability is strong, can accurately indicate sender in wide-band
Nonlinear characteristic.
It is calculated based on dynamic multinuclear bandwidth generalized regression nerve networks in view of the above and other objects, the present invention provides one kind
The wide-band sender non-linear modeling method of method, includes the following steps:Test platform is built, measures and records wide-band and transmit
The nonlinear characteristic of the full working frequency range of machine collects input, output amplitude and the phase value of test signal and corresponding carrier frequency
Rate;Segment processing is carried out to sample of signal according to carrier frequency, using amplitude non-uniform quantizing technology to the amplitude of input signal
Quantified, and realizes the structure of kinematic nonlinearity sample fingerprint data set by dynamic memory fingerprint technique;Use dynamic
Non-linear sample fingerprint data set trains generalized regression nerve networks, and obtaining dynamic multinuclear bandwidth broad sense in conjunction with optimization algorithm returns
Return neural network model, realizes and wide-band sender full frequency band nonlinear characteristic is modeled.
The test platform is made of equipment such as vector signal generator, Vector Signal Analyzer, attenuator and PC machine.
Test process is completed by instrument automatic measurement software program, by PC machine by cable or GPIB line traffic controls, test data preserves
Into PC machine.
The signal tested has:The amplitude A of digital baseband signal is inputted in vector signal generatorin, phase thetain, and
Radiofrequency signal carrier frequency F after up-conversionc;The amplitude A of digital baseband signal after being demodulated in Vector Signal Analyzerfb, phase thetafb。
Test program flow is:
Step 1:Bandwidth, the sampling speed of digital baseband signal in vector network generator and Vector Signal Analyzer are set
Rate, modulation system, rolloff-factor etc., wherein the sampling rate f of digital signalsample100Mps must be more than;
Step 2:Setting vector network generator radio-frequency carrier signal frequency is Fi, it includes baseband signal that signal source, which is one section,
The code of amplitude A and phase theta full dynamic range;
Step 3:The data and test result for extracting vector network generator and Vector Signal Analyzer, are sent to PC machine
In database;
Step 4:Increase radio-frequency carrier signal frequency in vector network generator according to the frequency stepping Δ F set in advance
Value repeats step 2~3, until the test completed to sender whole working frequency points (sets sender bandwidth of operation as B, shares n
=mod (B/ Δ F)+1 Frequency point).
After completing testing process, the signal data obtained after measurement is handled, kinematic nonlinearity sample fingerprint is obtained
Data set:
Step 1:The frequency segmentation technology can be non-linear to working band classification processing according to unit frequency-is transmitted, really
Determine numerical value and the stepping of frequency fingerprint point, control dynamic multinuclear bandwidth generalized regression nerve networks mode layer neuron quantity and
Bandwidth.
Step 1:Determine sender full frequency band transmitting gain value G0, usually by sender full working frequency range minimum transmitting
Yield value determines;
Step 2:According to the data of measurement, the gain vector of sender under each carrier frequency is calculated, and normalizing is made to G0
Change is handled, and obtains { GI, k, i ∈ (0, max (A)], k ∈ [1, n] };
Step 3:Calculate the Euclidean distance d between each frequency gain vectorI, j, obtain Distance matrix D:
Step 4:By clustering algorithm, such as K mean cluster algorithm, DBSCAN density clustering algorithms or AGNES levels are poly-
Class algorithm carries out classification processing to frequency vector, working frequency range is divided into p continuous frequency sub-band, and obtain by each frequency range
Initial frequency value composition frequency fingerprint collection { fi, (i ∈ [1, p]) }.
The amplitude non-uniform quantizing technology can be according to signal amplitude gain compression ratio, i.e. AM-AM amount distortions, to input
The amplitude of signal carries out segment processing, determines numerical value and the stepping of amplitude fingerprint point, control dynamic multinuclear bandwidth generalized regression god
Quantity and bandwidth through network mode layer neuron:
Step 1:Calculate the mean value { MG of the gain vector in each frequency rangeI, k, i ∈ (0, max (A)], k ∈ [1, p] };
Step 2:Following processing is carried out to each mean vector obtained in step 1 using line rectification function (RELU):
w(MGI, k)=RELU (G0-MGI, k)=max (0, G0-MGI, k) (6)
Wherein, gain is more than the point w=0 of G0, it is believed that the corresponding point of these amplitudes it is non-linear smaller, therefore by its amplitude
Fingerprint value is classified as 0.It, can be with by the AM-AM characteristics of power amplifier it is found that gain compression degree is directly proportional with the size of input signal amplitude
Approximation regards monotone variation as.In order to exclude noise jamming, to data acquisition system { wI, k, i ∈ (0, max (A)], k ∈ [1, p] } into
The processing of row fitting of a polynomial, obtains continuously smooth gain compression curve g (w).
Step 3:Threshold value threshold delta G is set, and threshold value is bigger, and piecewise interval length is longer, and segments is fewer, model accuracy
It is lower;Threshold value is smaller, and piecewise interval length is shorter, and segments is more, and model accuracy is higher;
Step 4:Segment processing is carried out to gained gain compression curve g (w) in step 2 according to threshold value threshold delta G, obtains width
Spend fingerprint value:
If the starting point of current amplitude fingerprint corresponds to z (initial value 1) a range value Ak, j=z+1 is enabled,
If g (wj)-g(wz) > Δ G, then z=j+1, and by AkIt is set as next amplitude fingerprint value, then repeatedly step 4;
If g (wj)-g(wz)≤Δ G, then j=j+1, repeats step 4;
Step 5:After completing step 4 to entire data set, amplitude fingerprint collection { α is obtainedi, i ∈ (0, max (A)] };
Step 6:Step 4 and step 5 are repeated by the data set { MG } obtained by formula (6) to all p frequency ranges, obtain full range
Section amplitude finger print data collection { αI, k, i ∈ (0, max (A)], k ∈ [1, p] }.
In [0,2 π] interior uniform quantization phase intervals Δ θ is arranged, then phase fingerprint collection length in the phase fingerprint value of signal
For:R=INT (2 π/Δ θ).R is smaller, and model accuracy is lower;Conversely, model accuracy is higher.Phase fingerprint collection after quantization
Close { θj, j ∈ [1, r] }
The dynamic memory fingerprint technique can be by changing input signal amplitude and load in non-linear sample fingerprint data
The memory depth m of wave frequency rate controls the quantity and mode layer of dynamic multinuclear bandwidth generalized regression nerve networks input layer
The dimension of kernel function.
The kinematic nonlinearity sample fingerprint structure:
{ [α (t), α (t-1) ..., α (t-m)], [f (t), f (t-1) ..., f (t-m)], θ }
The dynamic multinuclear bandwidth generalized regression nerve networks topology is by input layer, mode layer, summation layer and output layer
Composition, with reference to figure 3.
The neuronal quantity of the input layer is v=(2 × m+1), corresponds to the amplitude that the length in finger print data is m respectively
M frequency vector of vector sum and 1 phase data;
The quantity of the mode layer neuron is by frequency fingerprint collection length p, amplitude fingerprint collection length, phase fingerprint collection length
R and memory depth m are codetermined;
The summation layer includes 4 neurons;
The output layer includes 2 neurons, corresponds to the range value and phase value that model estimates respectively.
Using the kinematic nonlinearity sample fingerprint data set training generalized regression nerve networks, trained target component is
The corresponding Gaussian kernel bandwidth aggregation { σ of mode layer neuroni, i ∈ [1, v] }.
The Gaussian kernel bandwidth parameter can be by optimization algorithm, such as gradient descent method, Newton method, particle cluster algorithm, heredity
The mathematical methods such as algorithm, annealing algorithm acquire.It realizes and wide-band sender full frequency band nonlinear characteristic is modeled.
The dynamic multinuclear bandwidth general regression neural network uses incremental learning pattern, can be by inputting newly
Training sample further increases model accuracy, and can the nonlinear variation of tracking power amplifier in time, corrected parameter improves mould
The robustness of type.
The accurate modeling to wide band station sender nonlinear characteristic may be implemented in the present invention, the think of based on machine learning
Think, by the method for discovery learning and incremental learning, extract and screen the nonlinear characteristic pattern of wide-band sender, utilizes
Neural network algorithm training pattern, model structure is flexible, needs trained parameter less, and fitting performance is good, noise resisting ability
By force, can accurately indicate sender wide-band nonlinear characteristic, and can tracking power amplifier nonlinear variation in time, repair
Positive parameter improves the robustness of model.
Description of the drawings
Fig. 1 is the wide-band sender nonlinear characteristic modeling procedure figure of the present invention.
Fig. 2 is that the wide-band sender nonlinear characteristic test platform of the present invention constitutes schematic diagram.
Fig. 3 is that kinematic nonlinearity sample fingerprint data set builds schematic diagram in the present invention.
Fig. 4 is dynamic multinuclear bandwidth general regression neural network structural schematic diagram in the present invention.
Specific implementation mode
Below in conjunction with description of the drawings embodiments of the present invention, those skilled in the art can be by the content of this specification announcement
The further advantage and effect of the present invention are understood easily.The present invention also can be implemented or be answered by other different specific examples
With details in this specification can also be based on different perspectives and applications, carries out without departing from the spirit of the present invention various
Modification and change.
With reference to figure 1, this method comprises the following steps:It is complete by building test platform, measuring and recording wide-band sender
The nonlinear characteristic of working frequency range collects input, output amplitude and the phase value of test signal and corresponding carrier frequency,
Segment processing is carried out to sample of signal according to carrier frequency, non-uniform quantizing processing then is carried out to the amplitude of input signal, and
The structure that kinematic nonlinearity sample fingerprint data set is realized using dynamic memory fingerprint technique is instructed using the sample fingerprint data
Practice generalized regression nerve networks, dynamic multinuclear bandwidth general regression neural network, realization pair are obtained in conjunction with optimization algorithm
Wide-band sender full frequency band nonlinear characteristic models.
With reference to figure 2, the test platform is by vector signal generator, Vector Signal Analyzer, attenuator and PC machine etc.
Equipment forms.Test process is completed by instrument automatic measurement software program, cable or GPIB line traffic controls, test are passed through by PC machine
Data are preserved into PC machine.
The signal tested has:The amplitude A of digital baseband signal is inputted in vector signal generatorin, phase thetain, and
Radiofrequency signal carrier frequency F after up-conversionc;The amplitude A of digital baseband signal after being demodulated in Vector Signal Analyzerfb, phase thetafb。
Test program flow is:
Step 1:Bandwidth, the sampling speed of digital baseband signal in vector network generator and Vector Signal Analyzer are set
Rate, modulation system, rolloff-factor etc., wherein the sampling rate f of digital signalsample100Mps must be more than;
Step 2:Setting vector network generator radio-frequency carrier signal frequency is Fi, it includes baseband signal that signal source, which is one section,
The code of amplitude A and phase theta full dynamic range;
Step 3:The data and test result for extracting vector network generator and Vector Signal Analyzer, are sent to PC machine
In database;
Step 4:Increase radio-frequency carrier signal frequency in vector network generator according to the frequency stepping Δ F set in advance
Value repeats step 2~3, until the test completed to sender whole working frequency points (sets sender bandwidth of operation as B, shares n
=mod (B/ Δ F)+1 Frequency point).
The signal data obtained after measurement is handled, kinematic nonlinearity is obtained after completing testing process with reference to figure 3
Sample fingerprint data set:
Step 1:The frequency segmentation technology can be non-linear to working band classification processing according to unit frequency-is transmitted, really
Determine numerical value and the stepping of frequency fingerprint point, control dynamic multinuclear bandwidth generalized regression nerve networks mode layer neuron quantity and
Bandwidth.
Step 1:Determine sender full frequency band transmitting gain value G0, usually by sender full working frequency range minimum transmitting
Yield value determines;
Walk mule 2:According to the data of measurement, the gain vector of sender under each carrier frequency is calculated, and normalizing is made to G0
Change is handled, and obtains { GI, k, i ∈ (0, max (A)], k ∈ [1, n] };
Step 3:Calculate the Euclidean distance d between each frequency gain vectorI, j, obtain Distance matrix D:
Step 4:By clustering algorithm, such as K mean cluster algorithm, DBSCAN density clustering algorithms or AGNES levels are poly-
Class algorithm carries out classification processing to frequency vector, working frequency range is divided into p continuous frequency sub-band, and obtain by each frequency range
Initial frequency value composition frequency fingerprint collection { fi, (i ∈ [1, p]) }.
The amplitude non-uniform quantizing technology can be according to signal amplitude gain compression ratio, i.e. AM-AM amount distortions, to input
The amplitude of signal carries out segment processing, determines numerical value and the stepping of amplitude fingerprint point, control dynamic multinuclear bandwidth generalized regression god
Quantity and bandwidth through network mode layer neuron:
Step 1:Calculate the mean value { MG of the gain vector in each frequency rangeI, k, i ∈ (0, max (A)], k ∈ [1, p] };
Step 2:Following processing is carried out to each mean vector obtained in step 1 using line rectification function (RELU):
w(MGI, k)=RELU (G0-MGI, k)=max (0, G0-MGI, k) (6)
Wherein, gain is more than the point w=0 of G0, it is believed that the corresponding point of these amplitudes it is non-linear smaller, therefore by its amplitude
Fingerprint value is classified as 0.It, can be with by the AM-AM characteristics of power amplifier it is found that gain compression degree is directly proportional with the size of input signal amplitude
Approximation regards monotone variation as.In order to exclude noise jamming, to data acquisition system { wI, k, i ∈ (0, max (A)], k ∈ [1, p] } into
The processing of row fitting of a polynomial, obtains continuously smooth gain compression curve g (w).
Step 3:Threshold value threshold delta G is set, and threshold value is bigger, and piecewise interval length is longer, and segments is fewer, model accuracy
It is lower;Threshold value is smaller, and piecewise interval length is shorter, and segments is more, and model accuracy is higher;
Step 4:Segment processing is carried out to gained gain compression curve g (w) in step 2 according to threshold value threshold delta G, obtains width
Spend fingerprint value:
If the starting point of current amplitude fingerprint corresponds to z (initial value 1) a range value Ak, j=z+1 is enabled,
If g (wj)-g(wz) > Δ G, then z=j+1, and by AkIt is set as next amplitude fingerprint value, then repeatedly step 4;
If g (wj)-g(wz)≤Δ G, then j=j+1, repeats step 4;
Step 5:After completing step 4 to entire data set, amplitude fingerprint collection { α is obtainedi, i ∈ (0, max (A)] };
Step 6:Step 4 and step 5 are repeated by the data set { MG } obtained by formula (6) to all p frequency ranges, obtain full range
Section amplitude finger print data collection { αI, k, i ∈ (0, max (A)], k ∈ [1, p] }.
In [0,2 π] interior uniform quantization phase intervals Δ θ is arranged, then phase fingerprint collection length in the phase fingerprint value of signal
For:R=INT (2 π/Δ θ).
R is smaller, and model accuracy is lower;Conversely, model accuracy is higher.Phase fingerprint set { θ after quantizationj, j ∈
[1, r] }
The dynamic memory fingerprint technique can be by changing input signal amplitude and load in non-linear sample fingerprint data
The memory depth m of wave frequency rate controls the quantity and mode layer of dynamic multinuclear bandwidth generalized regression nerve networks input layer
The dimension of kernel function.
The kinematic nonlinearity sample fingerprint structure:
{ [α (t), α (t-1) ..., α (t-m)], [f (t), f (t-1) ..., f (t-m)], θ }
With reference to figure 4, the dynamic multinuclear bandwidth generalized regression nerve networks topology by input layer, mode layer, summation layer with
And output layer composition, with reference to figure 3.
The neuronal quantity of the input layer is v=(2 × m+1), corresponds to the amplitude that the length in finger print data is m respectively
M frequency vector of vector sum and 1 phase data;
The quantity of the mode layer neuron is by frequency fingerprint collection length p, amplitude fingerprint collection length, phase fingerprint collection length
R and memory depth m are codetermined;
The summation layer includes 4 neurons;
The output layer includes 2 neurons, corresponds to the range value and phase value that model estimates respectively.
Using the kinematic nonlinearity sample fingerprint data set training generalized regression nerve networks, trained target component is
The corresponding Gaussian kernel bandwidth aggregation { σ of mode layer neuroni, i ∈ [1, v] }.
The Gaussian kernel bandwidth parameter can be by optimization algorithm, such as gradient descent method, Newton method, particle cluster algorithm, heredity
The mathematical methods such as algorithm, annealing algorithm acquire.It realizes and wide-band sender full frequency band nonlinear characteristic is modeled.
The dynamic multinuclear bandwidth general regression neural network uses incremental learning pattern, can be by inputting newly
Training sample further increases model accuracy, and can the nonlinear variation of tracking power amplifier in time, corrected parameter improves mould
The robustness of type.
The accurate modeling to wide band station sender nonlinear characteristic may be implemented in the present invention, the think of based on machine learning
Think, by the method for discovery learning and incremental learning, extract and screen the nonlinear characteristic pattern of wide-band sender, utilizes
Neural network algorithm training pattern, model structure is flexible, needs trained parameter less, and fitting performance is good, noise resisting ability
By force, can accurately indicate sender wide-band nonlinear characteristic, and can tracking power amplifier nonlinear variation in time, repair
Positive parameter improves the robustness of model.
The above, best specific implementation mode only of the invention, but scope of protection of the present invention is not limited thereto,
Any one skilled in the art in the technical scope disclosed by the present invention, the change or replacement that can be readily occurred in,
It should be covered by the protection scope of the present invention.
Claims (10)
1. a kind of wide-band sender non-linear modeling method based on dynamic multinuclear bandwidth generalized regression nerve networks algorithm, packet
Include following steps:Test platform is built, the nonlinear characteristic of the full working frequency range of wide-band sender is measured and record, collects test
Input, output amplitude and the phase value of signal and corresponding carrier frequency;Sample of signal is segmented according to carrier frequency
Processing, the amplitude of input signal is quantified using amplitude non-uniform quantizing technology, and by dynamic memory fingerprint technique come
Realize the structure of kinematic nonlinearity sample fingerprint data set;Use kinematic nonlinearity sample fingerprint data set training generalized regression god
Through network, dynamic multinuclear bandwidth general regression neural network is obtained in conjunction with optimization algorithm, is realized to wide-band sender
Full frequency band nonlinear characteristic models.
2. according to the method described in claim 1, it is characterized in that, the test platform is believed by vector signal generator, vector
The equipment such as number analyzer, attenuator and PC machine composition.
3. according to the method described in claim 1, it is characterized in that, the test platform is measured and recorded inputs hair by signal source
The input digital baseband signal of letter machine obtains after the received machine demodulation of carrier frequency and sender output signal of sender
Digital baseband signal, wherein signal source can be vector signal generator, receiver can be Vector Signal Analyzer.
4. according to the method described in claim 1, it is characterized in that, the kinematic nonlinearity sample fingerprint data set is by input number
The memory depth of word baseband signal is the amplitude fingerprint and frequency fingerprint of m, phase fingerprint, and demodulation corresponding with input fingerprint
The amplitude fingerprint of digital baseband signal, phase fingerprint afterwards.
5. according to the method described in claim 1, it is characterized in that, the dynamic multinuclear bandwidth generalized regression nerve networks are topological
It is made of input layer, mode layer, summation layer and output layer.
6. according to the method described in claim 1, it is characterized in that, the frequency segmentation technology is non-thread according to unit frequency-is transmitted
Property working band classification is handled, determine numerical value and the stepping of frequency fingerprint point, control dynamic multinuclear bandwidth general regression neural
The quantity and bandwidth of network mode layer neuron.
7. according to the method described in claim 1, it is characterized in that, the amplitude non-uniform quantizing technology increases according to signal amplitude
Beneficial compression ratio carries out segment processing to the amplitude of input signal, determines numerical value and the stepping of amplitude fingerprint point, controls dynamic multinuclear
The quantity and bandwidth of bandwidth generalized regression nerve networks mode layer neuron.
8. according to the method described in claim 1, it is characterized in that, the dynamic memory fingerprint technique is non-by setting sender
Linear memory depth, the length of time-ofday signals amplitude and carrier frequency before changing in sample fingerprint data, control dynamic are more
The dimension of the quantity and mode layer kernel function of the wide generalized regression nerve networks input layer of nucleus band.
9. according to the method described in claim 1, it is characterized in that, the optimization algorithm is gradient descent method, Newton method, grain
Swarm optimization, genetic algorithm or annealing algorithm.
10. according to the method described in claim 1, it is characterized in that, the dynamic multinuclear bandwidth generalized regression nerve networks mould
Type uses incremental learning pattern, further increases model accuracy by inputting new training sample, and can track in time
The nonlinear variation of power amplifier, corrected parameter improve the robustness of model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810648237.9A CN108768550B (en) | 2018-06-21 | 2018-06-21 | Wide-band transmitter nonlinear modeling method based on dynamic multi-core bandwidth generalized regression neural network algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810648237.9A CN108768550B (en) | 2018-06-21 | 2018-06-21 | Wide-band transmitter nonlinear modeling method based on dynamic multi-core bandwidth generalized regression neural network algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108768550A true CN108768550A (en) | 2018-11-06 |
CN108768550B CN108768550B (en) | 2021-07-06 |
Family
ID=63976468
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810648237.9A Active CN108768550B (en) | 2018-06-21 | 2018-06-21 | Wide-band transmitter nonlinear modeling method based on dynamic multi-core bandwidth generalized regression neural network algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108768550B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109657390A (en) * | 2018-12-28 | 2019-04-19 | 中国电子科技集团公司第二十九研究所 | A kind of technique IP statistical modeling method in radio frequency Integrated manufacture |
CN110035425A (en) * | 2019-04-04 | 2019-07-19 | 中国科学技术大学 | Based on wireless network card to the physical fingerprint extracting method of wireless device |
CN112231986A (en) * | 2020-11-04 | 2021-01-15 | 中国电子科技集团公司第二十九研究所 | Numerical control attenuator modeling method |
CN113553771A (en) * | 2021-07-30 | 2021-10-26 | 海宁利伊电子科技有限公司 | Dynamic X parameter kernel calculation method based on RNN (radio network) |
CN115327373A (en) * | 2022-04-20 | 2022-11-11 | 岱特智能科技(上海)有限公司 | Hemodialysis equipment fault diagnosis method based on BP neural network and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2005032189A1 (en) * | 2003-09-26 | 2005-04-07 | Universite Du Quebec En Abitibi-Temiscamingue (Uqat) | Method and system for indoor geolocation using an impulse response fingerprinting technique |
CN102437822A (en) * | 2011-11-30 | 2012-05-02 | 上海瑞和安琦通信科技有限公司 | Self-adaptive digital pre-distortion linear system of radio frequency power amplifier |
CN102983819A (en) * | 2012-11-08 | 2013-03-20 | 南京航空航天大学 | Imitating method of power amplifier and imitating device of power amplifier |
CN106130661A (en) * | 2016-06-13 | 2016-11-16 | 杭州电子科技大学 | Broadband wireless transmitter recognition methods based on Hammerstein Wiener model |
CN107104746A (en) * | 2017-04-26 | 2017-08-29 | 中央军委装备发展部第六十三研究所 | Frequency hopping radio set sender nonlinear characteristic modeling method |
-
2018
- 2018-06-21 CN CN201810648237.9A patent/CN108768550B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2005032189A1 (en) * | 2003-09-26 | 2005-04-07 | Universite Du Quebec En Abitibi-Temiscamingue (Uqat) | Method and system for indoor geolocation using an impulse response fingerprinting technique |
CN102437822A (en) * | 2011-11-30 | 2012-05-02 | 上海瑞和安琦通信科技有限公司 | Self-adaptive digital pre-distortion linear system of radio frequency power amplifier |
CN102983819A (en) * | 2012-11-08 | 2013-03-20 | 南京航空航天大学 | Imitating method of power amplifier and imitating device of power amplifier |
CN106130661A (en) * | 2016-06-13 | 2016-11-16 | 杭州电子科技大学 | Broadband wireless transmitter recognition methods based on Hammerstein Wiener model |
CN107104746A (en) * | 2017-04-26 | 2017-08-29 | 中央军委装备发展部第六十三研究所 | Frequency hopping radio set sender nonlinear characteristic modeling method |
Non-Patent Citations (1)
Title |
---|
尹思源等: "广义记忆型神经网络射频功放数字预失真器", 《微波学报》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109657390A (en) * | 2018-12-28 | 2019-04-19 | 中国电子科技集团公司第二十九研究所 | A kind of technique IP statistical modeling method in radio frequency Integrated manufacture |
CN110035425A (en) * | 2019-04-04 | 2019-07-19 | 中国科学技术大学 | Based on wireless network card to the physical fingerprint extracting method of wireless device |
CN110035425B (en) * | 2019-04-04 | 2021-10-01 | 中国科学技术大学 | Physical fingerprint extraction method for wireless equipment based on wireless network card |
CN112231986A (en) * | 2020-11-04 | 2021-01-15 | 中国电子科技集团公司第二十九研究所 | Numerical control attenuator modeling method |
CN113553771A (en) * | 2021-07-30 | 2021-10-26 | 海宁利伊电子科技有限公司 | Dynamic X parameter kernel calculation method based on RNN (radio network) |
CN113553771B (en) * | 2021-07-30 | 2024-02-20 | 海宁利伊电子科技有限公司 | Dynamic X parameter accounting method based on RNN network |
CN115327373A (en) * | 2022-04-20 | 2022-11-11 | 岱特智能科技(上海)有限公司 | Hemodialysis equipment fault diagnosis method based on BP neural network and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN108768550B (en) | 2021-07-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108768550A (en) | Wide-band transmitter nonlinear modeling method based on dynamic multi-core bandwidth generalized regression neural network algorithm | |
Liu et al. | Dynamic behavioral modeling of 3G power amplifiers using real-valued time-delay neural networks | |
Oyedare et al. | Estimating the required training dataset size for transmitter classification using deep learning | |
CN107064913A (en) | A kind of wireless location method and system based on deep learning | |
CN112640299A (en) | Digital predistortion under varying operating conditions | |
CN103941254A (en) | Soil physical property classification recognition method and device based on geological radar | |
Li et al. | Wireless transmitter identification based on device imperfections | |
CN105426921B (en) | A kind of RFID label antenna optimization method | |
CN111245512A (en) | Neural network-based nonlinear channel modeling method for visible light communication system | |
CN109088645A (en) | Rfid transmissions Poewr control method and device | |
CN112784690A (en) | Broadband signal parameter estimation method based on deep learning | |
CN107104746B (en) | Nonlinear characteristic modeling method for frequency hopping radio transmitter | |
CN113267718B (en) | Power amplifier testing method and device and computer readable storage medium | |
US20030076896A1 (en) | Method and apparatus for the linearization of a radio frequency high-power amplifier | |
Luongvinh et al. | Behavioral modeling of power amplifiers using fully recurrent neural networks | |
CN113591390A (en) | Model screening method and platform for receiver radio frequency link nonlinear effect | |
Simbelie et al. | Envelope time-domain characterizations to assess in-band linearity performances of pre-matched MASMOS® power amplifier | |
CN110161471A (en) | It is a kind of for the sample rate of cloud MIMO radar and the calculation method of quantization bit | |
RV et al. | Optimization of digital predistortion models for RF power amplifiers using a modified differential evolution algorithm | |
Yin et al. | Pattern recognition of RF power amplifier behaviors with multilayer perceptron | |
CN108268700A (en) | A kind of RF power amplification temperature characterisitic modeling method based on BPNN | |
Zhao et al. | Multi-band behavioral modeling of power amplifier using carrier frequency-dependent time delay neural network model | |
Thielecke et al. | On Evaluating V2X Channel Models with COTS 802.11 p Devices in Urban Field-Trials | |
Li et al. | Complex radial basis function networks trained by QR‐decomposition recursive least square algorithms applied in behavioral modeling of nonlinear power amplifiers | |
Sukkhawatchani et al. | Performance evaluation of anomaly detection in cellular core networks using self-organizing map |
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 |