CN107169476A - A kind of frequency identification system based on neutral net - Google Patents
A kind of frequency identification system based on neutral net Download PDFInfo
- Publication number
- CN107169476A CN107169476A CN201710468638.1A CN201710468638A CN107169476A CN 107169476 A CN107169476 A CN 107169476A CN 201710468638 A CN201710468638 A CN 201710468638A CN 107169476 A CN107169476 A CN 107169476A
- Authority
- CN
- China
- Prior art keywords
- frequency
- neural network
- training
- module
- neutral net
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Stabilization Of Oscillater, Synchronisation, Frequency Synthesizers (AREA)
- Synchronisation In Digital Transmission Systems (AREA)
Abstract
The invention belongs to integrated circuit fields, and in particular to a kind of frequency identification system based on neutral net.The present invention utilizes neural network module, sample comprising frequency information is handled, after sample training terminates, neural network module preserves current state, when subsequently being recognized to incoming frequency, search procedure need not be repeated, it is directly exported and just omits scope target code Cbit, the quick identification of complete paired frequency.The present invention can be used for phase-locked loop circuit in, it is calibrated after can obtain accurate output frequency.
Description
Technical field
The invention belongs to integrated circuit fields, and in particular to a kind of frequency identification system based on neutral net.
Background technology
Frequency identification, i.e., divided roughly to the single frequency in a frequency range, and identifies that its omits scope at the beginning of affiliated,
To allow circuit quickly to recognize it.During frequency identification, the frequency divided roughly in the frequency range is compiled first
Code, whole process is exactly that corresponding target code is searched out from the sequence of an order, the lookup mode of natural full blast
For dichotomy, the other modes such as certain sequential search can equally complete this task.
When using binary search, we provide an initial value, are compared with incoming frequency, then carry out multiple
Binary chop, obtains frequency range where it.
Searched for generic sequence, it is then slower than binary search in all frequency ranges, speed due to travel through
Some, are equal to the time for adding and tentatively being recognized to frequency.
Carrying out binary search to generic sequence is the mode of full blast, but be original frequency difference than it is larger when
Wait, it is necessary to which the multiple binary search of execution of repeatability, limits the possibility that recognition time further shortens.
Using traditional lookup method, when we repeatedly recognize, identification every time is required for being repeated once search procedure, this
Sample adds the total time that frequency is identified.
The content of the invention
There is problem or deficiency for above-mentioned, in order to complete that the frequency in a frequency range is tentatively recognized, quick point
Class, it is relevant with state before using Processing with Neural Network the invention provides a kind of frequency identification system based on neutral net
Data, the sample comprising frequency information is trained, classification to frequency is realized, it is final to obtain where frequency just slightly scope
Target code.
The frequency identification system based on neutral net, as shown in figure 1, including neural network module, sampling module, multichannel
Multiplexer and supervised learning information acquisition module.
The sampling module has two input signals:Sampling clock fclk will be gathered with signal fin, sampling module is sampled
Fin be converted into the sequence Serial data being made up of 0 and 1 for a string, i.e., the signal that neural network module can be handled is simultaneously defeated
Go out;It exports the input for connecing neural network module.
The supervised learning information acquisition module has two and sampling module identical input signal:Sampling clock fclk and
It is sampled signal fin;Its output signal is first slightly scope target code Cbit, the input phase with neural network module where fin
Even;By the way of binary chop, the corresponding Cbit of fin are determined, training objective is provided for neural network module, passes through controller
The training and output to neural network module are contrasted and supervised in real time.
The neural network module, there is two input signals:The Serial data and supervised learning exported by sampling module
The Cbit of information acquisition module output;There are two Enable Pins EN_TRAIN and EN_OUT, the work for switching neural network module
Make state;Its output signal is the target output Train after the completion of neural network module is trained;And pass through controller and nerve net
Network enters row data communication, display training process and accuracy rate;The output of neural network module meets the input S2 of multiplexer.
The multiplexer has two input signals:By outside, offer is designated as S1 and S2, wherein S1 input signal
DataIn, S2 input signal are Train;There are an Enable Pin EN and a signal behavior end Con, selected as Con=1
S1, realizes and the outside of target code is write, S2 is selected as Con=0, is selected by neural network module write-in;Its is defeated
Go out signal for final output Cbit.
Further, the neural network module is when EN-TRAIN is enabled, the serial data that sampling module is obtained
It is input to simultaneously among neural network module with Cbit, wherein serial data are mainly comprising information:Leading edge position, under
Drop is along the position where position and current frequency, and whether Cbit data train effective result for final inspection network;When full
After sufficient supervised learning during the requirement of accuracy, EN-OUT is enabled, and exports final target output.
Further, the neutral net is Recognition with Recurrent Neural Network, and the model of use is long memory models LSTM in short-term.
Its workflow is:
First stage:Training sample is produced with obtaining training objective
Input one is sampled frequency finWith clock frequency fclk, sampling module is to incoming frequency finSampling, comprising
The serial binary code serial data of frequency and phase information, wherein, incoming frequency f each timeinCorrespondence one is desired
Represent that its just slightly target code Cbit of scope, the Cbit are provided by supervised learning information acquisition module;
For i-th of incoming frequency fi, correspondence target code Ci, incoming frequency is by frequency range (fmin, fmax) traveled through
Into obtaining training sample X=[fi, Ci];
Second stage:The training of neural network module
EN_TRAIN signals are enabled, and neutral net enters physical training condition;
A, neutral net and training sample are initialized;
Neural network model and network initial weight are initialized, initialization input neuron number is n, n >=1, and just
Beginningization training sample is X=[fi, Ci], wherein fiInputted as the sample of neutral net, fiIt is a line serial binary code, by
0th, 1 composition, 0 and 1 sequencing includes frequency and phase information;Cbit as neutral net training objective, by binary system
Represent;
B, training
Using neural network BP training algorithm, network output is produced according to sample first, training result is monitored by controller
The degree of accuracy, when being unsatisfactory for default accuracy requirement, adjust the number and network weight of hidden layer, will until meeting the degree of accuracy
After asking, neural network parameter, including neural network model, input neuron number and network weight are preserved, training terminates;
Phase III:Work
A, reading incoming frequency fin
Read input fin, it is serial binary code serial data that sampling processing is carried out to it;
B, neutral net export the corresponding just slightly scope target code Cbit of the incoming frequency
The neural network parameter that neural network module is preserved according to step is responded to input serial data, is produced
The corresponding Cbit of the incoming frequency, when EN_OUT is enabled, closes neural metwork training state, Cbit is exported real by neutral net
The identification of existing frequency.
Further, the above-mentioned frequency identification system based on neutral net, in phase-locked loop circuit, it is calibrated after
Obtain accurate output frequency.
The data relevant with state before are handled in the present invention by using neural network module, so as to realize to comprising frequency
The training of the sample of rate information:Training sample is input to neural network module first, it is defeated that neural network module can produce network
Go out, by controller to monitor the degree of accuracy of training result, when being unsatisfactory for default accuracy requirement, adjust the number of hidden layer
With network weight, until meeting accuracy requirement, training terminates.
In summary, the present invention utilizes neural network module, the sample comprising frequency information is handled, sample training
After end, neural network module preserves current state, when subsequently being recognized to incoming frequency, it is not necessary to repeat search procedure, directly
Connecing output, it just omits scope target code Cbit, the quick identification of complete paired frequency.
Brief description of the drawings
Fig. 1 is the structural representation of the present invention;
Fig. 2 is the sample schematic diagram that embodiment uses frequency sampling mode;
Fig. 3 a are the advanced schematic diagrames of sample phase that embodiment uses phase sample mode, and Fig. 3 b are that embodiment uses phase
The delayed schematic diagram of sample phase of sample mode;
Fig. 4 is the schematic block diagram for applying the present invention to phase-locked loop circuit;
Fig. 5 is the phase-locked loop circuit schematic diagram that embodiment uses frequency sampling;
Fig. 6 is the phase-locked loop circuit schematic diagram that embodiment uses phase sample;
Fig. 7 is the topological structure of RNN neutral nets;
Fig. 8 is the sequential expanded view of RNN neutral nets;
Fig. 9 is the RNN neutral net flow charts based on LSTM.
Embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.
With reference to Fig. 5 and Fig. 6, using Recognition with Recurrent Neural Network structure RNN (Recurrent Neural Networks), LSTM
The pattern of (Long Short Term Memory) algorithm, frequency sampling and phase sample, provides the present invention and is applied to phaselocked loop
The specific implementation example of loop quick lock in.
Recognition with Recurrent Neural Network (RNN) topological structure is as shown in fig. 7, comprises input layer, hidden layer and output layer, wherein inputting
Collection be labeled as { x0, x1 ..., xt, xt+1 ... }, output stage be labeled as { y0, y1 ..., yt, yt+1 ... }, hidden unit it is defeated
Go out collection and be labeled as { s0, s1 ..., st, st+1 ... }, these hidden units complete topmost work.
Fig. 7 is the incomplete structure chart of Recognition with Recurrent Neural Network.Because the ring of Recognition with Recurrent Neural Network hidden layer is included from ring, friendship
Ring is pitched, often RNN is typically drawn as sequential expanded view as shown in Figure 8.
It can easily be seen that the input of hidden layer also includes the state of a upper hidden layer, with sequential from sequential expanded view
Increase, network depth is increasing.
ot=g (Vst) (1)
st=f (Uxt+Wst-1) (2)
Formula 1 is the calculation formula of output layer, and output layer is a full articulamentum, that is, its each node and is hidden
Each node of layer is connected.V is the weight matrix of output layer, and g is activation primitive.
Formula 2 is the calculation formula of hidden layer, and it is circulation layer.U is the weight matrix for inputting x, and W is last value conduct
The weight matrix of input this time, f is activation primitive.It can be seen that current time input layer to the information of hidden layer by formula 2
Determined with the information transmission of last moment input layer to hidden layer by weight matrix U and W, hidden layer to the information between output layer
Transmission is determined by weight matrix V.
This example is implemented by neutral net frequency identification module and cycle of phase-locked loop module, and the system is used for into phaselocked loop
The quick lock in of circuit, its topological structure is as shown in Figure 5.
Wherein, neutral net frequency identification module includes neural network module, sampling module, multiplexer, supervision
Information acquisition module is practised, the effect of supervised learning information acquisition module is that neutral net is trained, produce correct tuning bent
Scope target code Cbit is just omited where line coding, i.e. frequency;
Controller is connected with neural network module, for realizing the data communication with Recognition with Recurrent Neural Network, and shows training
Process, accuracy rate etc.;
Cycle of phase-locked loop module includes phase frequency detector (PFD), charge pump (CP), loop filter (LPF), VCO
Device (PFD), frequency divider (Divider), its effect is to realize PLL loop-locking functions;
The output of supervised learning information acquisition module is connected with the input of neural network module in neural network module, supervision
The input of learning information acquisition module is connected with reference frequency Fref and crossover frequency Fdiv, and its effect is believed by supervised learning
Cease acquisition module and obtain correct tuning curve control word under current frequency dividing ratio, the control word is used for following RNN supervised learnings
Training objective is provided.By the way of binary chop, then fixed frequency dividing ratio carries out multiple binary chop, obtains optimal tune
Harmonic curve control word.
The input of sampling module is connected with reference frequency Fref and crossover frequency Fdiv, and output is inputted with neural network module
It is connected, its effect is that crossover frequency Fdiv is sampled with reference frequency Fref, and it is serial that frequency information is then converted into 0,1
Sequence, then send sequence as sample into neural network module;
Wherein, when when by the way of to frequency sampling, its nucleus module is A/D converter, with Fclk pairs of sampling clock
It is sampled signal Fin to be sampled, Fin is sampled in each Fclk rising edge, obtains a series of 0,1 serial sequence
Row (as shown in Fig. 2 its square wave frequency is higher, illustrate that be sampled frequency differs bigger with sampling clock, square wave frequency is lower, says
The bright frequency that is sampled differs smaller with sampling clock).When processing is by the data of frequency sampling, one determination of individual data correspondence
Frequency, wherein sampling clock Fclk correspondence clock signal Fref are sampled signal Fin i.e. crossover frequency Fdiv, sampling comes out
Data fi represent currently to be sampled the frequency of signal, we carry out multiple repairing weld to each frequency dividing ratio, construct neutral net
The training sample and test sample of module provide input for follow-up neural network module.
Further, the input of phase frequency detector (PFD) is reference frequency Fref and crossover frequency Fdiv, and it exports connection
The input of charge pump (CP), discharge and recharge is carried out to subsequent capacitance, and CP outputs are adjusted by loop filter (LPF), loop
Wave filter (LPF) output is the control voltage Vcont, VCO of voltage controlled oscillator (VCO) output vo1, vo2 and point below
Frequency device is connected, signal (Fdiv) after frequency divider is divided, then is connected with PFD, a composition PFD input, with reference frequency
Fref is compared, and is linked to be cycle of phase-locked loop.
But the loop does not have the function of locking automatically, it is necessary to manually adjust frequency dividing ratio to find lock-out state.Will ginseng
Frequency Fref and the obtained frequency Fdiv of frequency dividing are examined as the input of the frequency identification module based on neural network module, with the frequency
Rate identification module helps phase locked loop fast lock.
Further, when when by the way of to phase sample, its nucleus module is TDC modules, and effect is to use TDC modules
Read out sampling clock Fclk and be sampled signal Fin phase difference, wherein, Fin enters a time delay chain, Fclk rising edge
The level of the time delay chain is sampled, it (is that phase is advanced as shown in Fig. 3 .a that phase information, which is converted into 0,1 serial sequence,
When, schematic diagram of the frequency by time delay chain is sampled, shown in Fig. 3 .b, when being delayed phase, frequency is sampled and passes through time delay chain
Schematic diagram).When processing is by the data of phase sample, wherein sampling clock Fclk correspondence clock signal Fref are, it is necessary into prolonging
When chain signal be crossover frequency Fdiv, sampling out data fiIn, individual data does not correspond to the frequency of a determination, still
Difference between consecutive number string, correspond to the phase difference between the frequency Fdiv and clock signal Fref that are sampled, wherein ordered series of numbers A
(A0、A1、A2...) be by Fdiv carry out multiple Fref rising edges sampling extraction Lai TDC sequences, be designated as fi。
Wherein, the input all the way of A bit multiplexeds device is connected with the output of RNN modules, can be controlled all the way by outside write-in
System, while having Enable Pin and signal behavior end, output is connected to VCO, and it, which is acted on, can be achieved on to the outer of tuning curve
Portion writes or selected by RNN write-ins;
Wherein, neural network module can handle the data relevant with state before, so as to realize to comprising frequency
The training of the sample of information:Training sample is input to neural network module first, neural network module can produce network output,
By controller, we can monitor the degree of accuracy of training result, when being unsatisfactory for default accuracy requirement, adjustment hidden layer
Number and network weight, until meeting accuracy requirement, training terminates.
The identifying system of frequency based on neutral net, the quick lock in function for phase-locked loop circuit is main by following several
Individual step is realized:
1st, training sample is produced.The training sample is obtained by sampling module, sample mode frequency sampling pattern, this data
Obtained by A/D samplings.Related to time order and function due to sampling, 0 and 1 context then includes frequency information, remembers serial data
Serial-data is fi。
2nd, training objective is produced.The training objective is obtained by supervised learning information acquisition module, by binary chop, suitable
Sequence lookup method etc. looks for the training objective of Recognition with Recurrent Neural Network and is designated as Ci, thus obtain training sample X=[fi, Ci], wherein
fiIt is used as the input of Recognition with Recurrent Neural Network, CiAs the training objective of Recognition with Recurrent Neural Network, by binary representation.
3rd, said process is repeated to whole frequency dividing ratios, obtains the frequency range (f of samplemin, fmax), construct data volume
Sufficiently large training sample and test sample.
4th, when RNN enters training mode, by different tuning curve control word CbitSequentially input, be designated as Ci, it is different
Control word correspond to different tuned frequencies, obtain binary serial code.
5th, Recognition with Recurrent Neural Network is trained.Recognition with Recurrent Neural Network and training sample are initialized first, initially
Change Recognition with Recurrent Neural Network model and network initial weight, initialization input neuron number is 1, and initializes training sample
For X=[fi, Ci];
6th, the weight matrix and bias matrix in neural network module are trained.EN-TRAIN is enabled, such as Fig. 7 institutes
Show, the data f that sampling is obtainediWith corresponding CiSend into input input modules, wherein CiSend into prediction and expect module
To detect whether to obtain correct answer, sampled data fiFeeding pre_processing pretreatment modules are pre-processed, and are located
Manage as [x0, x1..., xn] (wherein n is neuron number in the dimension of data, correspondence first_layer, can be according to training
As a result self-defining), the data after processing are then sent into the layer networks of first_layler first, with reference to weights matrixes and
Biases matrixes are calculated, and export hW, b(x)=f (wx, b).Together sent by LSTM networks and by first_layer output
Enter LSTM_layer calculating, take once result every 5000 times (checking the number of times of result oneself can define), send into output_
Layer output modules, by the data in output_layer result and initial input Input, calculate its difference coefficient, using gradient
Descent method is trained (train).An iteration is often carried out, a weight matrix weights and bias matrix biases is updated.
Test sample is put into the network to be tested, final correct probability is exported by prediction modules.Enabled in EN-TRAIN
When, when we use frequency sampling Frequency Patterns, the f that A/D modules are obtainediData and CiData are input to input moulds simultaneously
Among block, wherein fiData are mainly comprising information:Position where leading edge position, trailing edge position and current frequency, CiNumber
Effective result whether is trained according to for final inspection network;When by the way of phase sample, the f that TDC modules are obtainedi
Data and CiData are input among input modules simultaneously, wherein fiData are mainly comprising information:Leading edge position, trailing edge
Position where position and current frequency, CiWhether data train effective result for final inspection network;
7th, neural network module final output.By controller, we can monitor the degree of accuracy of training result, when discontented
When foot presets accuracy requirement, the number and network weight of hidden layer are adjusted, until meeting accuracy requirement, circulation nerve is preserved
Network parameter, including Recognition with Recurrent Neural Network model, input neuron number and network weight, training terminate, when meeting accuracy
It is required that when, EN-OUT is enabled, the sequence of the final corresponding tuning curve of output.
8th, output control word completes the quick lock in of PLL loops into PLL loops.The control word enable output when,
Control word is input to VCO capacitor array, phase lock loop circuit loop work, to realize quick lock in.
In summary, it is seen that the present invention can be used for the identification of frequency, simultaneously because neutral net can be handled with frequency
The sequence of rate information, can be obtained with different sample modes for its sample mode for providing sample, be not limited to frequency sampling
With phase sample pattern.
Claims (5)
1. a kind of frequency identification system based on neutral net, it is characterised in that:Including neural network module, sampling module, many
Path multiplexer and supervised learning information acquisition module;
The sampling module has two input signals:Sampling clock fclk and it is sampled signal fin, sampling module is by collection
Fin is converted into a string of sequence Serial data being made up of 0 and 1, i.e., the signal that neural network module can be handled is simultaneously defeated
Go out;It exports the input for connecing neural network module;
The supervised learning information acquisition module has two and sampling module identical input signal:Sampling clock fclk and adopted
Sample signal fin;First slightly scope target code Cbit where its output signal is fin, is connected with the input of neural network module;Adopt
With the mode of binary chop, the corresponding Cbit of fin are determined, training objective is provided for neural network module, it is real-time by controller
Training and output to neural network module are contrasted and supervised;
The neural network module, there is two input signals:The Serial data and supervised learning information exported by sampling module
The Cbit of acquisition module output;There are two Enable Pins EN_TRAIN and EN_OUT, the work shape for switching neural network module
State;Its output signal is the target output Train after the completion of neural network module is trained;And entered by controller with neutral net
Row data communication, display training process and accuracy rate;The output of neural network module meets the input S2 of multiplexer;
The multiplexer has two input signals:By outside, offer is designated as DataIn to S1 and S2, wherein S1 input signal,
S2 input signal is Train;There are an Enable Pin EN and a signal behavior end Con, S1, realization pair are selected as Con=1
The outside write-in of target code, S2 is selected as Con=0, is selected by neural network module write-in;Its output signal is most
Whole output Cbit.
2. the frequency identification system as claimed in claim 1 based on neutral net, it is characterised in that:
The neural network module is when EN-TRAIN is enabled, and the serial data and Cbit that sampling module is obtained is simultaneously defeated
Enter among neural network module, wherein serial data are mainly comprising information:Leading edge position, trailing edge position and work as
Whether the position where preceding frequency, Cbit data train effective result for final inspection network;After supervised learning is met
During the requirement of accuracy, EN-OUT is enabled, and exports final target output.
3. the frequency identification system as claimed in claim 1 based on neutral net, it is characterised in that:The neutral net is circulation
Neutral net, the model of use is long memory models LSTM in short-term.
4. the frequency identification system based on neutral net as claimed in claim 1, its workflow is:
First stage:Training sample is produced with obtaining training objective
Input one is sampled frequency finWith clock frequency fclk, sampling module is to incoming frequency finSampling, obtain comprising frequency and
The serial binary code serial data of phase information, wherein, incoming frequency f each timeinCorrespondence one is desired to represent it
The target code Cbit of scope is just omited, the Cbit is provided by supervised learning information acquisition module;
For i-th of incoming frequency fi, correspondence target code Ci, incoming frequency is by frequency range (fmin, fmax) traversal completion, obtain
To training sample X=[fi, Ci];
Second stage:The training of neural network module
EN_TRAIN signals are enabled, and neutral net enters physical training condition;
A, neutral net and training sample are initialized;
Neural network model and network initial weight are initialized, initialization input neuron number is n, n >=1, and is initialized
Training sample is X=[fi, Ci], wherein fiInputted as the sample of neutral net, fiIt is a line serial binary code, by 0,1 group
Into 0 and 1 sequencing includes frequency and phase information;Cbit as neutral net training objective, by binary representation;
B, training
Using neural network BP training algorithm, network output is produced according to sample first, the standard of training result is monitored by controller
Exactness, when being unsatisfactory for default accuracy requirement, adjusts the number and network weight of hidden layer, until meeting accuracy requirement
Afterwards, neural network parameter, including neural network model, input neuron number and network weight are preserved, training terminates;
Phase III:Work
A, reading incoming frequency fin
Read input fin, it is serial binary code serial data that sampling processing is carried out to it;
B, neutral net export the corresponding just slightly scope target code Cbit of the incoming frequency
The neural network parameter that neural network module is preserved according to step is responded to input serial data, produces this defeated
Enter the corresponding Cbit of frequency, when EN_OUT is enabled, close neural metwork training state, Cbit outputs are realized frequency by neutral net
The identification of rate.
5. the frequency identification system based on neutral net as claimed in claim 1, in phase-locked loop circuit, it is calibrated after
Obtain accurate output frequency.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710468638.1A CN107169476B (en) | 2017-06-20 | 2017-06-20 | Frequency identification system based on neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710468638.1A CN107169476B (en) | 2017-06-20 | 2017-06-20 | Frequency identification system based on neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107169476A true CN107169476A (en) | 2017-09-15 |
CN107169476B CN107169476B (en) | 2020-06-16 |
Family
ID=59819820
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710468638.1A Active CN107169476B (en) | 2017-06-20 | 2017-06-20 | Frequency identification system based on neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107169476B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109002891A (en) * | 2018-03-15 | 2018-12-14 | 小蚁科技(香港)有限公司 | The selectivity control based on feature of neural network |
CN110009186A (en) * | 2019-02-19 | 2019-07-12 | 国网江西省电力有限公司电力科学研究院 | Power distribution station top-oil temperature prediction technique and device |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070132525A1 (en) * | 2004-02-13 | 2007-06-14 | Koninklijke Philips Electronics N.C. | Method and device for testing a phase locked loop |
CN105929419A (en) * | 2016-04-13 | 2016-09-07 | 南京理工大学 | BP artificial neural network-based GPS carrier tracking method |
-
2017
- 2017-06-20 CN CN201710468638.1A patent/CN107169476B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070132525A1 (en) * | 2004-02-13 | 2007-06-14 | Koninklijke Philips Electronics N.C. | Method and device for testing a phase locked loop |
CN105929419A (en) * | 2016-04-13 | 2016-09-07 | 南京理工大学 | BP artificial neural network-based GPS carrier tracking method |
Non-Patent Citations (2)
Title |
---|
徐明哲: "BP神经网络在信号发生器中的应用", 《科技视界》 * |
杨晓明 等: "基准结构自振频率的人工神经网络识别", 《辽宁工程技术大学学报(自然科学版)》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109002891A (en) * | 2018-03-15 | 2018-12-14 | 小蚁科技(香港)有限公司 | The selectivity control based on feature of neural network |
CN110009186A (en) * | 2019-02-19 | 2019-07-12 | 国网江西省电力有限公司电力科学研究院 | Power distribution station top-oil temperature prediction technique and device |
Also Published As
Publication number | Publication date |
---|---|
CN107169476B (en) | 2020-06-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107463966B (en) | Radar range profile's target identification method based on dual-depth neural network | |
Vanzella et al. | Photometric redshifts with the Multilayer Perceptron Neural Network: Application to the HDF-S and SDSS | |
CN110020667A (en) | Searching method, system, storage medium and the equipment of neural network structure | |
EP0823090B1 (en) | Adaptive filtering neural network classifier | |
CN105510970B (en) | Obtain seismic facies optimal classes purpose method | |
US20230196202A1 (en) | System and method for automatic building of learning machines using learning machines | |
CN108478216A (en) | A kind of epileptic seizure intelligent Forecasting early period based on convolutional neural networks | |
CN107169476A (en) | A kind of frequency identification system based on neutral net | |
Birgmeier | A fully Kalman-trained radial basis function network for nonlinear speech modeling | |
CN108732620A (en) | A kind of non-supervisory multi-wave seismic oil and gas reservoir prediction technique under supervised learning | |
CN110414723A (en) | The method, apparatus and system of fractured hydrocarbon reservoir history matching based on microseismic event | |
CN109948452A (en) | A kind of clock signal prediction technique and device | |
CN109919312A (en) | Operation method, device and the DPU of convolutional neural networks | |
CN111553513A (en) | Medium-and-long-term runoff prediction method based on quadratic decomposition and echo state network | |
CN114897144A (en) | Complex value time sequence signal prediction method based on complex value neural network | |
US6650162B2 (en) | Digital clock generator circuit with built-in frequency and duty cycle control | |
CN110378035A (en) | It is a kind of that soft-measuring modeling method is hydrocracked based on deep learning | |
KR102428833B1 (en) | Method for manufacturing frequency controlled source loop and system therefor | |
CN104966074B (en) | Coal-rock identification method based on transform domain broad sense Gamma distribution | |
CN103873056B (en) | Agitator self-checking device and calibration steps thereof | |
CN109709045A (en) | Landwaste kind identification method and system based on optoacoustic spectroscopy integrated signal | |
CN106896404B (en) | The recognition methods of thin reservoir and device | |
CN104852877B (en) | Signal receiving device and its two benches adaptive equalization method | |
CN107463750A (en) | Capacitance-resistance parameter determination method in phase-locked loop circuit | |
CN103746691B (en) | Clock signal control system and method |
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 |