CN106209072A - A kind of crystal oscillator based on artificial neural network - Google Patents
A kind of crystal oscillator based on artificial neural network Download PDFInfo
- Publication number
- CN106209072A CN106209072A CN201610493237.7A CN201610493237A CN106209072A CN 106209072 A CN106209072 A CN 106209072A CN 201610493237 A CN201610493237 A CN 201610493237A CN 106209072 A CN106209072 A CN 106209072A
- Authority
- CN
- China
- Prior art keywords
- data
- input
- neural network
- artificial neural
- processing unit
- 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
- H03—ELECTRONIC CIRCUITRY
- H03L—AUTOMATIC CONTROL, STARTING, SYNCHRONISATION, OR STABILISATION OF GENERATORS OF ELECTRONIC OSCILLATIONS OR PULSES
- H03L1/00—Stabilisation of generator output against variations of physical values, e.g. power supply
- H03L1/02—Stabilisation of generator output against variations of physical values, e.g. power supply against variations of temperature only
- H03L1/022—Stabilisation of generator output against variations of physical values, e.g. power supply against variations of temperature only by indirect stabilisation, i.e. by generating an electrical correction signal which is a function of the temperature
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention belongs to integrated circuit fields, a kind of crystal oscillator based on artificial neural network.This crystal oscillator includes controlling voltage generating module and crystal oscillation module.Control voltage generating module to be made up of temperature sensor, cymometer, data selector, input data processing unit, artificial neural network and output data processing unit.The present invention produces the control voltage needed for varactor two ends by artificial neural network, utilize artificial neural network self unintentional nonlinearity mapping ability, make this network can be with highly precise approach nonlinear function, thus generation is the control voltage of nonlinear function to temperature, this voltage accurately compensates, by the capacitance controlling varactor, the skew that the frequency of TCXO and VCTCXO produces because of variations in temperature, significantly improves its frequency stability.
Description
Technical field
The invention belongs to integrated circuit fields, a kind of crystal oscillator based on artificial neural network.
Background technology
Crystal oscillator is widely used in portable electric appts and wireless communication technology etc. as stable frequency source
Field, its frequency is to weigh one of good and bad important indicator of crystal oscillator performance to the degree of stability of variations in temperature.Temperature-compensating
Crystal oscillator (TCXO) is to reduce, by temperature compensation network, the skew that oscillator output frequencies produces because of variations in temperature, thus
Improve frequency stability.
Temperature compensating crystal oscillator (TCXO) is typically by temperature compensation network, varactor, crystal resonator and vibration electricity
Road module composition, as shown in Figure 1.Its operation principle is: temperature compensation network produces the control voltage varied with temperature, this control voltage
Act on varactor, thus it is possible to vary the capacitance of varactor, compensate the resonant frequency of crystal resonator because of variations in temperature
The skew produced, thus improve the frequency stability of crystal oscillator.
It is temperature that the voltage at varactor two ends and crystal oscillator resonant frequency non-linear relation result in control voltage V
The higher-order function of degree, voltage and frequency relation non-linear the biggest, then the higher order term in voltage and temperature relation is the biggest
(K.R.Ward, A novel approach to improving the stability of TCVCXO temperature
performance,Frequency Control Symposium and PDA Exhibition Jointly with the
17th European Frequency and Time Forum,2003.Proceedings of the 2003IEEE
International,pp.473–477,2003)。
Existing compensation schemes mainly has two kinds, and one is analog temperature compensation scheme, and another kind is digital temperature
Compensation scheme.In analog temperature compensation scheme, because producing is that the voltage of higher-order function is relatively difficult to temperature, general with linear
The approximation to function voltage V control action to frequency of oscillation, thus producing temperature with temperature compensation network is polynomial function (
As be cube function) control voltage V, compensation precision is limited;In design scheme, by voltage temperature letter
The matching of number and the mode of interpolation obtain the control voltage needed for each temperature spot, although ratio of precision analog temperature compensation scheme
Precision is high, but its precision is limited to matching and the form of Interpolation-Radix-Function and exponent number, and exponent number is the highest, and the precision of compensation is the highest,
System resource and the complexity that however it is necessary that are the biggest, are not suitable for the application scenario having further requirement to compensation precision.
The non-linear relation controlling voltage and resonant frequency of a crystal not only have impact on temperature compensating crystal oscillator (TCXO)
Compensation precision, had a strong impact on the frequency stable of the adjustable temperature compensating crystal oscillator of output frequency (VCTCXO) especially
Property.At present, (VCTCXO) realizes again by varactor, but the control voltage V of varactor is temperature compensation network to be produced
Raw compensation voltage V1Voltage V with outside regulation mid frequency2Cum rights is added generation.
The non-linear relation controlling voltage V and resonant frequency of a crystal causes VCTCXO at different regulation voltage V2Under, temperature
Degree characteristic is different, i.e. after the mid frequency of crystal oscillator changes, owing to oscillator frequency temperature characteristics revolves
Turn (J.R.Vig, Quartz crystal resonators and oscillators for frequency control
and timing applications.A tutorial.[online]Available:http://www.ieee-
Uffc.org), same set temperature compensation data can cause compensation precision inadequate.For solving this problem, patent
A kind of method proposing linearisation frequency-voltage relationship in US6549055B2, by controlling voltage and varactor
Between insert a circuit module with non-linear transfer function so that between input voltage to output frequency, be approximately line
Sexual relationship, thus improve the temperature stability of VCTCXO, but the program has the circuit module knot of non-linear transfer function
Structure is complicated, increases the difficulty of crystal oscillator design.
Summary of the invention
For existing TCXO and VCTCXO because the temperature that the non-linear relation compensated between voltage and frequency causes is mended
Repay the shortcoming that precision is inadequate, the invention provides a kind of crystal oscillator based on artificial neural network.
Technical scheme is as follows:
A kind of crystal oscillator based on artificial neural network, including controlling voltage generating module and crystal oscillation module.
Control voltage generating module by temperature sensor, cymometer, data selector, input data processing unit, artificial
Neutral net and output data processing unit composition, its effect is that to produce be the control voltage of non-linear relation to temperature.
Crystal oscillation module is made up of varactor, crystal resonator and oscillating circuit, and its effect is to produce vibration letter
Number.
In described control voltage generating module, temperature sensor is connected with input data processing unit, is used for detecting and passing
The temperature data T of defeated crystal oscillation module;
The input of cymometer is connected with the output of crystal oscillation module, outfan and input 1 port of data selector
Being connected, its effect is the frequency monitoring and transmitting the oscillator signal that crystal oscillation module produces in data acquisition phase;
Input 2 port of data selector is connected with the input of outside hub frequency, output termination input data processing unit,
Its effect is to select the data path by data selector under signal Contr2 controls;
The input of input data processing unit is also connected with the control voltage V of varactor in crystal oscillation module,
Output be connected with the input of artificial neural network, its effect be receive input data, and it is processed into artificial neural network can
With the data directly processed;
Artificial neural network is feedforward network, and its output termination output data processing unit, effect is at signal Contr2
To input data X it is anticipated that V=g (X) makes nonlinear response under control, produce non-linear control voltage;
The output of output data processing unit connects the control voltage of varactor, and its effect is to artificial neural network
Output data process and are followed by varactor two end, by controlling the vibration of the capacitance compensated crystal oscillator of varactor
The skew that frequency produces because of variations in temperature.
Further, Contr1 is mode switching signal, and this signal is input to input data processing unit by control
Data dimension controls crystal oscillator and switches between TCXO pattern and VCTCXO pattern.
Further, Contr2 is artificial neural network state switching signal, and this signal controls artificial neural network and learning
Switch between habit state and duty, control the data by data selector simultaneously.
Further, Contr1 signal controls crystal oscillator in different pattern, different together with Contr2 signal
Work under state.Specifically, crystal oscillator switching between different patterns, different states is real in the following manner
Existing:
Under TCXO pattern, when Contr2 signal control artificial neural network is in learning state, signal Contr1 controls
The data being input to input data processing unit are 2 dimensions, including temperature data T and control voltage data V, i.e. X=(V, T);When
When Contr2 signal control artificial neural network is in running order, signal Contr1 controls to be input to input data processing unit
Data be 1 dimension, only temperature data T, it may be assumed that X=(T);
Under VCTCXO pattern, when Contr2 signal control artificial neural network is in learning state, signal Contr1 is controlled
It is 3-dimensional that system is input to input the data of data processing unit, including temperature data T, controls voltage data V and frequency data f,
That is: X=(V, f, T);When Contr2 signal control artificial neural network is in running order, signal Contr1 controls to be input to
The data of input data processing unit are 2 dimensions, including temperature data T and frequency data f, it may be assumed that X=(f, T).
Further, the work process being somebody's turn to do crystal oscillator based on artificial neural network is divided into three phases, particularly as follows:
First stage: collecting training data
During TCXO pattern, in temperature T that each is differentiUnder, i is temperature label, applies control at varactor two ends
Voltage V processediSo that at different temperature spot Ti, the frequency shift (FS) of crystal oscillator output is zero, records data (Vi,Ti),
To training sample Y=(V, T), now training sample Y is two-dimensional array, and wherein, T is as the input of artificial neural network, V conduct
The target output of artificial neural network;
During VCTCXO pattern, first keep temperature T constant, change the control voltage V at varactor two ends, use cymometer
Record different ViThe output frequency f of lower agitator;Then keep the control voltage V at varactor two ends constant, in difference
Temperature spot TiThe output frequency f of lower cymometer record crystal oscillator, obtains training sample Y=(f, V, T), now trains
Sample Y is three-dimensional array, and wherein, f and T exports as the target of artificial neural network as the input of artificial neural network, V.
Second stage: the training (study) of artificial neural network
In this stage, Contr2 signal controls artificial neural network and is in learning state, controls to pass through data selector simultaneously
The data that data are 1 port, the frequency data f i.e. obtained in the collecting training data stage.The training sample that first stage gathers
Notebook data is input to artificial neural network after input data processing unit processes, and artificial neural network is to input variable (TCXO
During pattern, the input of artificial neural network is X=T;During VCTCXO pattern, the input of artificial neural network is X=(f, T)) do
Go out response, produce network output, then network output and target output V are compared, when both errors are unsatisfactory for presetting
During required precision, artificial neural network adjusts network weight W, until error is less than the precision preset, training terminates, and this process is such as
Shown in Fig. 3.
Phase III: work
In this stage, it is in running order that Contr2 signal controls artificial neural network.Under TCXO pattern, Contr1 controls defeated
Enter to input data processing unit data be X=(T), artificial neural network is according to the network weight trained in second stage
Input being made nonlinear response, produces non-linear control voltage V=g (T), this control voltage is after output data processing unit
The capacitance of control varactor, thus the skew that the frequency of oscillation compensating TCXO produces because of variations in temperature;
Under VCTCXO pattern, Contr2 controls by the data of data selector from port 2, the center of i.e. outside input
Frequency, meanwhile, Contr1 control be input to input data processing unit data be X=(f, T), artificial neural network according to
Nonlinear response is made in input by the network weight that second stage trains, and produces non-linear control voltage V=g (f, T), manually
The output of neutral net controls the capacitance of varactor after output data processing unit, thus accurately compensated crystal oscillator
The skew that produces because of variations in temperature of frequency of oscillation.Now, the binary function that voltage V is f and T is controlled, i.e. in different
Under frequency of heart f, produce the non-linear control voltage that temperature characterisitic is different.
Present invention artificial neural network produces the control voltage needed for varactor two ends.Artificial neural network self
Unintentional nonlinearity mapping ability makes this network can be with highly precise approach nonlinear function, thus producing is non-thread to temperature
Property function control voltage, this voltage by control varactor capacitance accurately compensate the frequency of TCXO and VCTCXO because of temperature
The skew that degree change produces, significantly improves its frequency stability.
Artificial neural network produces and meets the non-linear control voltage presetting precision, with existing polynomial function form
Compensation voltage is compared, and compensation precision is high, it is possible to be applicable to the occasion having high request to temperature stability;For VCTCXO, due to
Artificial neural network can approach binary nonlinear function, i.e. can produce different temperatures characteristic under different mid frequencyes
Non-linear control voltage, it is to avoid oscillator frequency temperature characterisitic rotation after mid frequency changes in existing compensation schemes
The shortcoming that the compensation effect quoted is the best;The present invention is applicable to the quartz crystal of any cut type, lithium tantalate, lithium niobate crystal
The crystal such as body, gallium lanthanum and MEMS.
In sum, compensation precision of the present invention is high, and overcomes VCTCXO oscillator frequency after mid frequency changes
Temperature characterisitic rotates the shortcoming that the compensation effect caused is the best;It is applicable to the quartz crystal of any cut type, lithium tantalate, niobic acid
Crystalline lithium, gallium lanthanum and MEMS crystal.
Accompanying drawing explanation
Fig. 1 is existing temperature compensating crystal oscillator structure schematic diagram;
Fig. 2 is the structural representation of the present invention;
Fig. 3 is the training process schematic of artificial neural network;
Fig. 4 is VCTCXO schematic diagram based on BP neutral net;
Fig. 5 (a) is structural representation during VCTCXO based on artificial neural network training, and (b) is based on artificial neuron
The structural representation when VCTCXO of network works.
Detailed description of the invention
Below in conjunction with Fig. 4 and Fig. 5, with BP neutral net as model, as a example by VCTCXO pattern, provide the one of the present invention
The specific embodiment of crystal oscillator based on artificial neural network.Artificial neural network in the present invention can be to be other forms
Model, and the present invention is applicable not only to VCTCXO, is also applied for TCXO.
Fig. 4 is as a example by VCTCXO pattern, it is provided that a kind of based on artificial neural network the crystal oscillator of the present invention
Embodiment, this embodiment includes controlling voltage generating module and crystal oscillation module.
Control voltage generating module by temperature sensor, cymometer, data selector, input data processing unit, artificial
Neutral net and output data processing unit composition, its effect is to produce the nonlinear function V about temperature T and frequency f;
Crystal oscillation module includes varactor, crystal resonator and oscillating circuit, is used for producing oscillator signal.
In described control voltage generating module, temperature sensor is connected with input data processing unit, is used for detecting and passing
The temperature data T of defeated crystal oscillation module;
The input of cymometer is connected with the output of crystal oscillation module, outfan and input 1 port of data selector
Being connected, its effect is the frequency monitoring and transmitting the oscillator signal that crystal oscillation module produces in data acquisition phase;
Input 2 port of data selector is connected with the input of outside hub frequency, output termination input data processing unit,
Its effect is to select the data path by data selector under signal Contr2 controls;
The input of input data processing unit is also connected with the control voltage V of varactor in crystal oscillation module,
Output be connected with the input of artificial neural network, its effect be receive input data, and it is processed into artificial neural network can
With the data directly processed;
The output termination output data processing unit of artificial neural network, its effect is to defeated under signal Contr2 controls
Enter data X=(f, T) it is anticipated that V=g (f, T) makes nonlinear response, produce non-linear control voltage;
The output of output data processing unit connects the control voltage of varactor, and its effect is to artificial neural network
Output data process and are followed by varactor two ends, by controlling the vibration of the capacitance compensated crystal oscillator of varactor
The skew that frequency produces because of variations in temperature.
Further, Contr1 is mode switching signal, and this signal is input to input data processing unit by control
The dimension of data X controls crystal oscillator and switches between TCXO pattern and VCTCXO pattern, and now in example, Contr1 controls
Crystal oscillator is in VCTCXO pattern;Contr2 is artificial neural network state switching signal, and this signal controls artificial neuron
Network switches between learning state and duty, controls the data by data selector simultaneously.
Contr1 signal control together with Contr2 signal crystal oscillator when different patterns, different work
Make.Specifically, crystal oscillator switching between different patterns, different states is accomplished by:
In this embodiment, Contr1 signal controls crystal oscillator and is under VCTCXO pattern, when Contr2 signal controls
When artificial neural network is in learning state, the data that signal Contr1 controls to be input to input data processing unit are 3-dimensional, bag
Include temperature data T, control voltage data V and frequency data f, it may be assumed that X=(V, f, T);When artificial neural network is in running order
Time, the data that signal Contr1 controls to be input to input data processing unit are 2 dimensions, including temperature data T and frequency data f,
That is: X=(f, T).
Further, use three layers of BP neural network model artificial neural network include by two neurons form defeated
Enter a layer Ii(i=1,2), hidden layer H being made up of 4 neuronsj(j=1,2,3,4) output layer and by 1 neuron formed
Ok(k=1), i, j, k are input layer, hidden layer, output layer neuron label respectively;Hidden nodes purpose selects by being actually subjected to
The problem solved determines, is chosen as 4 the most for convenience of description;Any one hidden neuron receives all neurons of input layer
The signal of transmission, and it is transferred to output layer neuron after it is carried out Nonlinear Processing, do not have signal to pass with between layer neuron
Pass.Nonlinear transformation is by the activation primitive of each neuronDetermine, activation primitive hereinSelect sigmoid function;Input layer
Information to hidden layer is transmitted by weight wijDetermining, the information between hidden layer to output layer is transmitted by weightsDetermine.
Further, the work process of the VCTCXO based on BP neutral net of the present embodiment is divided into three phases, specifically
For:
First stage: collecting training data
For the VCTCXO being made up of the crystal of different qualities and the varactor of different qualities, first keep temperature not
Become, change the control voltage V that varactor two ends are additional, with the output frequency f of cymometer detection VCTCXO, record data;
Then keep the control voltage V at varactor two ends constant, at different temperature spot TiThe defeated of VCTCXO is detected with cymometer
Going out frequency f, record data, obtain training sample Y=(f, T, V), this training sample Y is three-dimensional array, wherein frequency data f and
Temperature data T is the input of BP neutral net, and V is the target output of BP neutral net.
Second stage: the training (study) of artificial neural network
Shown in structural representation such as Fig. 5 (a) of this stage VCTCXO, Contr2 signal controls BP neutral net and is in study
State, and control by the data of data selector from the output of port 1, i.e. cymometer;Contr1 signal controls first
The training sample Y=(f, T, V) of phase acquisition is input to input data processing unit, and wherein frequency data f and temperature data T is
The input of BP neutral net, V is the target output of BP neutral net.Each sample of input is made sound by BP neutral net
Should, produce reality output, then compare the error between the actual output of BP neutral net and target output, when error is unsatisfactory for
During the precision set, BP neutral net adjusts network weight wij、Until error is less than the precision set.This stage includes two
Individual process: signal forward-propagating and error back propagation, specific as follows:
Signal forward-propagating process: signal is the most successively transmitted by input neuron, non-through hidden layer and output layer
Linear process, is finally exported by output neuron.During Gai, network weight be constant.
Output for some sample S, BP neutral net can be expressed as:
Wherein,It is hidden layer and the activation primitive of output layer neuron, selection sigmoid function here:
Error back propagation process: the actual output of BP neutral net is compared with target output, when differing bigger,
Then the error signal of the two is propagated the most forward as input signal from the outfan of network.Back propagation makes BP nerve net
The network weight w of networkij、The direction reduced towards error function is constantly revised, until error is reduced to the precision preset.If
Target corresponding for sample S is output as Ts, the error of the most all samples is:
Wherein, n is sample size.When this error ratio preset precision big time, neutral net adjust network weight wijWith
Until the error shown in above formula is less than presetting precision, then the training stage of BP neutral net completes, now artificial neural network energy
Enough with permissible accuracy approximating function V=g (f, T).
Phase III: work
In this stage, it is in running order that Contr2 signal controls artificial neural network, and controls by data selector
Data are from port 2, the mid frequency of i.e. outside input.The data that Contr1 controls to be input to input data processing unit only have
Temperature data T and mid frequency data f.BP neutral net according to the network weight trained in second stage to input X=(f,
T) making nonlinear response, produce non-linear control voltage V=g (f, T), this control voltage is controlled after output data processing unit
The capacitance of varactor processed, thus the skew that the frequency of oscillation accurately compensating VCTCXO produces because of variations in temperature.Due to BP
Neutral net can produce the control voltage that temperature characterisitic is different under different mid frequencyes, it is to avoid in traditional VCTCXO
Frequency-temperature characteristic is caused to rotate the problem that the compensation precision caused is inadequate because mid frequency changes.
Claims (5)
1. a crystal oscillator based on artificial neural network, including controlling voltage generating module and crystal oscillation module, its
It is characterised by:
Control voltage generating module by temperature sensor, cymometer, data selector, input data processing unit, artificial neuron
Network and output data processing unit composition, its effect is that to produce be the control voltage of non-linear relation to temperature;
Crystal oscillation module is made up of varactor, crystal resonator and oscillating circuit, and its effect is to produce oscillator signal;
In described control voltage generating module, temperature sensor is connected with input data processing unit, is used for detecting and transmitting crystalline substance
The temperature data T of oscillation body module;
The input of cymometer is connected with the output of crystal oscillation module, and outfan is connected with input 1 port of data selector,
Its effect is the frequency monitoring and transmitting the oscillator signal that crystal oscillation module produces in data acquisition phase;
Input 2 port of data selector is connected with the input of outside hub frequency, and output termination input data processing unit, it is made
Be signal Contr2 control under select the data path by data selector;
The input of input data processing unit is also connected with the control voltage V of varactor in crystal oscillation module, output
Be connected with the input of artificial neural network, its effect be receive input data, and it is processed into artificial neural network can be straight
Connect the data of process;
Artificial neural network is feedforward network, and its output termination output data processing unit, effect is to control at signal Contr2
Under to input data X it is anticipated that V=g (X) makes nonlinear response, produce non-linear control voltage;
The output of output data processing unit connects the control voltage of varactor, and its effect is the output to artificial neural network
Data process and are followed by varactor two end, by controlling the frequency of oscillation of the capacitance compensated crystal oscillator of varactor
The skew produced because of variations in temperature;
Described Contr1 is mode switching signal, and this signal is input to input the data dimension numerical control of data processing unit by control
Crystal oscillator processed switches between TCXO pattern and VCTCXO pattern;
Described Contr2 is artificial neural network state switching signal, and this signal controls artificial neural network in learning state and work
Make to switch between state, control the data by data selector simultaneously.
2. crystal oscillator based on artificial neural network as claimed in claim 1, it is characterised in that: described different pattern,
The different switching modes between state is specific as follows:
Under TCXO pattern, when Contr2 signal control artificial neural network is in learning state, signal Contr1 controls input
Data to input data processing unit are 2 dimensions, including temperature data T and control voltage data V, i.e. X=(V, T);When
When Contr2 signal control artificial neural network is in running order, signal Contr1 controls to be input to input data processing unit
Data be 1 dimension, only temperature data T, it may be assumed that X=(T);
Under VCTCXO pattern, when Contr2 signal control artificial neural network is in learning state, signal Contr1 controls input
Data to input data processing unit are 3-dimensional, including temperature data T, control voltage data V and frequency data f, it may be assumed that X=
(V,f,T);When artificial neural network is in running order, signal Contr1 controls the number being input to input data processing unit
According to being 2 dimensions, including temperature data T and frequency data f, it may be assumed that X=(f, T).
3. crystal oscillator based on artificial neural network as claimed in claim 1, it is characterised in that: brilliant in crystal oscillation module
Body resonator is the quartz crystal of any cut type, lithium tantalate, lithium columbate crystal, gallium lanthanum or MEMS crystal.
4. crystal oscillator based on artificial neural network as claimed in claim 1, its work process is divided into three phases, specifically
For:
First stage, collecting training data:
During TCXO pattern, in temperature T that each is differentiUnder, i is temperature label, the control electricity applied at varactor two ends
Pressure ViSo that at different temperature spot Ti, the frequency shift (FS) of crystal oscillator output is substantially zeroed, records data (Vi,Ti),
To training sample Y=(V, T), now training sample Y is two-dimensional array, and wherein, T is as the input of artificial neural network, V conduct
The target output of artificial neural network;
During VCTCXO pattern, first keep temperature T constant, change the control voltage V at varactor two ends, use cymometer record
Different ViThe output frequency f of lower agitator;Then keep the control voltage V at varactor two ends constant, in different temperature
Degree point TiThe output frequency f of lower cymometer record crystal oscillator, obtains training sample Y=(f, V, T), now training sample
Y is three-dimensional array, and wherein, f and T exports as the target of artificial neural network as the input of artificial neural network, V;
Second stage, the training of artificial neural network i.e. learn:
Contr2 signal controls artificial neural network and is in learning state, and control by the data of data selector is 1 end simultaneously
The data of mouth, the frequency data f i.e. obtained in the collecting training data stage;The training sample data that first stage gathers are through input
Data processing unit is input to artificial neural network after processing, and input variable is responded by artificial neural network, produces network
Output, then compares, when both errors are unsatisfactory for the required precision preset, manually network output and target output V
Neutral net adjusts network weight W, until error is less than the precision preset, training terminates;
Phase III, work:
It is in running order that Contr2 signal controls artificial neural network, and under TCXO pattern, Contr1 controls to be input to input number
Being X=(T) according to the data of processing unit, input is made by artificial neural network according to the network weight trained in second stage
Nonlinear response, produces non-linear control voltage V=g (T), and this control voltage controls transfiguration two after output data processing unit
The capacitance of pole pipe, thus the skew that the frequency of oscillation compensating TCXO produces because of variations in temperature;
Under VCTCXO pattern, Contr2 controls by the data of data selector from port 2, the center frequency of i.e. outside input
Rate, meanwhile, the data that Contr1 controls to be input to input data processing unit are X=(f, T), and artificial neural network is according to the
Nonlinear response is made in input by the network weight that the two-stage trains, and produces non-linear control voltage V=g (f, T), manually god
After output data processing unit, the capacitance of varactor is controlled through the output of network, thus accurately compensated crystal oscillator
The skew that frequency of oscillation produces because of variations in temperature.
5. the work process of crystal oscillator based on artificial neural network as claimed in claim 4, it is characterised in that:
Described input variable is: during TCXO pattern, and the input of artificial neural network is X=T;During VCTCXO pattern, artificial neuron
The input of network is X=(f, T).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610493237.7A CN106209072B (en) | 2016-06-27 | 2016-06-27 | A kind of crystal oscillator based on artificial neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610493237.7A CN106209072B (en) | 2016-06-27 | 2016-06-27 | A kind of crystal oscillator based on artificial neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106209072A true CN106209072A (en) | 2016-12-07 |
CN106209072B CN106209072B (en) | 2018-10-16 |
Family
ID=57462331
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610493237.7A Active CN106209072B (en) | 2016-06-27 | 2016-06-27 | A kind of crystal oscillator based on artificial neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106209072B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107508576A (en) * | 2017-08-15 | 2017-12-22 | 电子科技大学 | A kind of active complex filter based on neutral net |
CN107562113A (en) * | 2017-08-22 | 2018-01-09 | 电子科技大学 | A kind of low line regulation reference circuit and production method based on neutral net |
CN107578096A (en) * | 2017-09-21 | 2018-01-12 | 胡明建 | A kind of voltage-frequency formula selects the design method of end artificial neuron |
CN107612546A (en) * | 2017-08-29 | 2018-01-19 | 电子科技大学 | A kind of phase-locked loop circuit based on neutral net |
CN112953460A (en) * | 2021-01-28 | 2021-06-11 | 武汉市博畅软件开发有限公司 | Response surface method-based frequency calibration method and system for electrically tunable filter |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1588030A (en) * | 2004-07-09 | 2005-03-02 | 大连理工大学 | Portable gas detection analyzer with temperature and humidity compensation function |
US20150142715A1 (en) * | 2011-08-11 | 2015-05-21 | Greenray Industries, Inc. | Neural network frequency control and compensation of control voltage linearity |
CN105336081A (en) * | 2015-11-26 | 2016-02-17 | 电子科技大学中山学院 | Fuzzy neural network-based electrical fire monitoring terminal and processing steps thereof |
-
2016
- 2016-06-27 CN CN201610493237.7A patent/CN106209072B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1588030A (en) * | 2004-07-09 | 2005-03-02 | 大连理工大学 | Portable gas detection analyzer with temperature and humidity compensation function |
US20150142715A1 (en) * | 2011-08-11 | 2015-05-21 | Greenray Industries, Inc. | Neural network frequency control and compensation of control voltage linearity |
CN105336081A (en) * | 2015-11-26 | 2016-02-17 | 电子科技大学中山学院 | Fuzzy neural network-based electrical fire monitoring terminal and processing steps thereof |
Non-Patent Citations (1)
Title |
---|
康波 等: "一种基于神经网络的混沌控制方法", 《系统工程与电子技术》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107508576A (en) * | 2017-08-15 | 2017-12-22 | 电子科技大学 | A kind of active complex filter based on neutral net |
CN107508576B (en) * | 2017-08-15 | 2020-12-29 | 电子科技大学 | Active complex filter based on neural network |
CN107562113A (en) * | 2017-08-22 | 2018-01-09 | 电子科技大学 | A kind of low line regulation reference circuit and production method based on neutral net |
CN107562113B (en) * | 2017-08-22 | 2019-04-05 | 电子科技大学 | One kind low line regulation reference circuit neural network based and production method |
CN107612546A (en) * | 2017-08-29 | 2018-01-19 | 电子科技大学 | A kind of phase-locked loop circuit based on neutral net |
CN107612546B (en) * | 2017-08-29 | 2020-10-23 | 电子科技大学 | Phase-locked loop circuit based on neural network |
CN107578096A (en) * | 2017-09-21 | 2018-01-12 | 胡明建 | A kind of voltage-frequency formula selects the design method of end artificial neuron |
CN112953460A (en) * | 2021-01-28 | 2021-06-11 | 武汉市博畅软件开发有限公司 | Response surface method-based frequency calibration method and system for electrically tunable filter |
Also Published As
Publication number | Publication date |
---|---|
CN106209072B (en) | 2018-10-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106253893A (en) | A kind of High Accuracy Microcomputer compensated crystal oscillator | |
CN106209072A (en) | A kind of crystal oscillator based on artificial neural network | |
US7932789B2 (en) | Frequency adjustable surface acoustic wave oscillator | |
CN105222765B (en) | The temperature-compensation method and system of MEMS gyro | |
CN104020453A (en) | Electronic circuit, radar apparatus, and method of performing self-diagnosis on radar apparatus | |
CN106597149A (en) | Oscillator residual life estimation method based on acceleration sensitivity | |
CN106292839A (en) | Real-time clock compensates circuit and calibration steps, device | |
CN109709802A (en) | The control method of active electronic ladder circuit based on iterative learning control | |
CN106603011B (en) | Voltage-controlled temperature compensation crystal oscillator | |
CN103279038A (en) | Self-adaptive control method of sliding formwork of micro gyroscope based on T-S fuzzy model | |
CN104467674B (en) | A kind of voltage controlled temperature compensated crystal oscillator | |
CN101626656B (en) | Radio frequency matching method and plasma processor | |
CN106208968A (en) | Signal frequency mixing method and frequency mixer | |
CN103699006A (en) | Control method based on fuzzy variable sliding model surface tracking differentiator | |
RU2487444C2 (en) | Method of generating high-frequency signals and apparatus for realising said method | |
CN102751920B (en) | Method for identifying and controlling two-induction-motor synchronous system | |
CN107106140B (en) | Ultrasonic probe and the diagnostic ultrasound equipment for having ultrasonic probe | |
CN107508576A (en) | A kind of active complex filter based on neutral net | |
Singh et al. | Design and VLSI implementation of fuzzy logic controller | |
CN110531625A (en) | The limited frequency range iterative learning fault tolerant control method of active electronic ladder circuit | |
RU2490780C2 (en) | Method for amplitude, phase and frequency modulation of high-frequency signals and multifunctional apparatus for realising said method | |
CN116430713B (en) | Method for improving control loop bandwidth of full-angle hemispherical resonator gyroscope | |
CN117761925A (en) | Liquid crystal phase shifter response time test system and test method | |
CN114360989B (en) | Magnetron automatic phase modulation device, system and method based on injection frequency locking | |
RU2488944C2 (en) | Method for amplitude, phase and frequency modulation of high-frequency signals and multifunctional apparatus for realising said method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |