CN105787473B - A method of extracting the Random telegraph noise signal with trap coupling effect - Google Patents
A method of extracting the Random telegraph noise signal with trap coupling effect Download PDFInfo
- Publication number
- CN105787473B CN105787473B CN201610183059.8A CN201610183059A CN105787473B CN 105787473 B CN105787473 B CN 105787473B CN 201610183059 A CN201610183059 A CN 201610183059A CN 105787473 B CN105787473 B CN 105787473B
- Authority
- CN
- China
- Prior art keywords
- formula
- hidden state
- trap
- moment
- random telegraph
- 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.)
- Active
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
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
- G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Monitoring And Testing Of Transmission In General (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a kind of methods for extracting the Random telegraph noise signal with trap coupling effect, are based on hidden Markov model, and from actual measurement in noisy leakage current signal, extraction obtains Random telegraph noise that is clean and having coupling information;It include: that the trap number N in device is set according to obtained leakage current signaltrapRange;For each NtrapValue carries out primary complete modeling and extracts, obtains corresponding Random telegraph noise signal RTN (t);Respectively to each NtrapCorresponding Bayesian information criterion BIC is calculated in value;By N corresponding to Bayesian information criterion BIC minimum valuetrapUnder extraction result RTN (t) be used as true Random telegraph noise signal, complete dependent on hidden Markov model Random telegraph noise signal extraction.The RTN that the present invention extracts can really reflect influencing each other between trap, while will not be influenced by state missing.
Description
Technical field
The invention belongs to field of microelectronic devices, are related to Random telegraph noise method for extracting signal more particularly to a kind of base
In the method that Hidden Markov Model extracts the Random telegraph noise signal with trap coupling effect.
Background technique
With the continuous diminution of semiconductor devices scale, the trap number in gate medium is gradually decreased, trap behavior with
Machine feature more highlights, and causes more and more attention.The presence of trap in device, can cause device property and circuit
Serious influence, for example, the delay of logic circuit is caused to increase, Static RAM (SRAM) reads failure etc..Therefore, it grinds
Study carefully the trap behavior in small size device, grasps degradation characteristics and circuit of the fundamental characteristics for prediction circuit of trap comprehensively
Reliability design be of great significance.
Trap has the characteristic of random capture and transmitting channel carrier, so that corresponding become occurs for the state of leakage current
Change, generates random fluctuation, i.e. Random telegraph noise (RTN).Therefore, RTN phenomenon is depended on for the research of trap behavior
Research.The measurement of RTN mainly directly detects the state of leakage current under a gate voltage.When due to actual test, it can not keep away
Exempt from there are noises, therefore, when practical operation, hardly result in the RTN signal of " clean ".
Due to the generation mechanism of RTN, meeting hidden Markov model (HMM) therefore can use HMM, from noise
Measured signal in extract and obtain the RTN signal of " clean ".Currently used HMM extracts method, it is assumed that the trap phase in device
It is mutually independent, i.e., coupling effect is not present between trap, and then extract RTN signal.However, in practice, between trap there is
Coupling of different strengths and weaknesses, i.e. whether occupying of trap will affect characteristic (including the amplitude, capture time of another trap
Constant, launch time constant).Coupling effect can cause very big influence to device property and circuit characteristic.Moreover, due to
Trap capture and transmitting channel carrier have very big randomness, and average capture and launch time are also different because of trap, practical
Measurement there is measurement window, might not it is observed that trap whole behaviors, the case where thus missing there is state so that
The extraction of RTN is more difficult.Therefore, existing method is difficult to realize extract from the signal of actual measurement clean and with coupling
RTN signal.
Summary of the invention
In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a kind of random telegraph of the extraction with trap coupling effect
The method of noise signal, this method are based on hidden Markov model, realize from actual measurement with mentioning in noisy leakage current signal
Take out Random telegraph noise that is clean and having coupling information.
In this specification, parameter definition involved in hidden Markov model is as follows:
Ntrap: the trap number in device;
Nstate: the hidden state number of RTN,
The hidden state of subscript i:RTN, and i ∈ [1,2 ... Nstate];
Ai: the corresponding current value of hidden state i, i.e., there is no the current values under noise situations;
Oi: there is the current value under noise situations, there is randomness, obey in the corresponding current observation of hidden state i
Gaussian Profile, Oi~N (Ai,σ2), wherein N (Ai,σ2) indicate, mean value Ai, standard deviation is the Gaussian Profile of σ;
K:K={ KijIt is a Nstate×NstateThe matrix of size indicates the transition probability between hidden state, matrix K
The element of i-th row j column is denoted as Kij, indicate the probability for being transferred to hidden state i in unit testing time interval from hidden state j;
T: the testing time of device current, t ∈ [1,2 ... T], T is total testing time;
I(t): current value measured by moment t institute's reality;
S(t): hidden state, S corresponding to moment t(t)∈[1,2,...Nstate]。
Present invention provide the technical scheme that
A method of the Random telegraph noise signal with trap coupling effect being extracted, the extracting method is based on hidden Ma Er
Can husband's model, obtain random telegraph that is clean and having coupling information from actual measurement with extracting in noisy leakage current signal
Noise;Include the following steps:
One) the leakage current signal of setting grid voltage and leakage pressure is obtained by actual measurement;
Two) the leakage current signal obtained according to sets the trap number N in devicetrapRange be 1~Nmax;
NmaxFor maximum trap number;For each NtrapValue carries out primary complete modeling and extracts, obtains the NtrapBe worth it is corresponding with
Electromechanics report noise signal RTN (t);The modeling extraction process specifically:
1) Random telegraph noise model is established based on hidden Markov model, obtains the state transition probability between hidden state
Matrix K;Each of matrix K element KijIt is the probability for being transferred to hidden state i in the unit time from hidden state j;Hidden state
Number is
2) according to the histogram of leakage current signal, the corresponding current amplitude of each hidden state is determined, according at the beginning of current amplitude
Step extracts Random telegraph noise signal, obtains corresponding hidden state S under each moment t(t), then state transition probability is calculated
Intensity matrix K and Gaussian noise amplitude σ, the initial value as Random telegraph noise model;
3) Random telegraph noise model is trained using feedforward-rear feed algorithm, sets convergence threshold, is asked by iteration
Solve the Random telegraph noise model, obtain model parameter, including state transition probability intensity matrix K, Gaussian noise amplitude σ and
There is no the current value A under noise situations;Random telegraph noise signal RTN (t) is obtained by the extraction of formula 6 after iteration convergence;
In formula 6, S(t)Indicate hidden state locating for t moment;Indicate S(t)Current value corresponding to state, S(t)It is counted by formula 5
It obtains:
In formula 5, ai(t) it indicates t moment, be in hidden state i and it has been observed that I(1),I(2),…I(t)Probability, by formula 1
It is calculated:
In formula 1, each of matrix K element KijIt is the probability for being transferred to hidden state i in the unit time from hidden state j;σ
For noise amplitude;T indicates the testing time;I(t)Indicate electric current measured by t moment;AiIndicate electric current corresponding to hidden state i
Value;
Three) respectively to each NtrapValue, is calculated the N using formula 9trapIt is worth corresponding Bayesian information criterion BIC:
Wherein, NtrapFor trap number;T is testing time overall length;S(T)It indicates hidden state locating for the T moment, is counted by formula 5
It obtains;For the T moment, it is in hidden stateS (T)), and observe sequence I(1),I(2),…I(T)Probability, calculated by formula 1
It obtains;
Four) by N corresponding to Bayesian information criterion BIC minimum valuetrapUnder extraction result RTN (t) as true
Thus Random telegraph noise signal completes the extraction process of the Random telegraph noise signal dependent on hidden Markov model.
The method of Random telegraph noise signal for said extracted with trap coupling effect, further, step 1) institute
It states setting grid voltage and leakage pressure and presses 0.9V for fixed grid voltage 0.9V and fixed leakage;Step 2) the maximum trap number NmaxIt is set as
3。
The method of Random telegraph noise signal for said extracted with trap coupling effect, further, step 3) institute
It states and Random telegraph noise model is trained using feedforward-rear feed algorithm, by iteratively solving the Random telegraph noise mould
Type obtains model parameter, specifically comprises the following steps:
3a) by feedforward arithmetic, model is calculated using formula 1 and is located at hidden state i in t moment, and has produced
Survey sequence I(1)~I(t)Probability ai(t);
3b) by rear feed algorithm, model is calculated using formula 2 and is located at hidden state i in t moment, and has produced
Survey sequence I(t)~I(T)Probability bi(t):
In formula 2, KjiIndicate the probability for being transferred to hidden state j in unit testing time interval from hidden state i;σ indicates noise
Amplitude;T indicates the testing time;T indicates total testing time;I(t)Indicate electric current measured by t moment;AiIndicate hidden state i
Corresponding current value;Last current state is the T moment;
Model parameter 3c) is obtained by iteratively solving the Random telegraph noise model according to generalized expectation-maximization;
Obtained Random telegraph noise signal RTN (t) 3d) is extracted by formula 6 according to after iteration convergence.
The method of Random telegraph noise signal of the said extracted with trap coupling effect, wherein step 3c) it is described by repeatedly
In generation, solves the Random telegraph noise model, obtains model parameter, specifically: constantly repeating step 3c1)~3c4) iterative solution
Model parameter terminates iteration when the iterative parameter knots modification between two step iteration steps is less than the convergence threshold:
Formula 1 and formula 2 3c1) are substituted into formula 3, acquire intermediate variable γij(t):
In formula 3, γij(t) it indicates t moment, the probability of hidden state j is transferred to by hidden state i;ai(t-1) when indicating t-1
It carves, is in hidden state i, and it has been observed that I(1),I(2),…I(t-1)Probability, can be calculated by formula 1;KjiIndicate unit testing
The probability of hidden state j is transferred in time interval from hidden state i;bj(t) in t moment, it is located at hidden state j, and has observed
To actual measurement sequence I(t)~I(T)Probability, can be calculated by formula 2;
3c2) formula 3 is substituted into formula 4, solution obtains parameter K:
In formula 4, γij(t) it indicates t moment, the probability of hidden state j is transferred to by hidden state i, can be calculated by formula 3;Formula 4
Put in marks ^ above the letter of equal sign left end, and expression is the value of current iteration step;
3c3) according to formula 5 and formula 6, it is calculated moment t=1, when 2 ... T, hidden state S locating for Random telegraph noise(t)
With corresponding Random telegraph noise signal RTN (t):
Wherein, S(t)Indicate hidden state locating for t moment;ai(t) t moment is indicated, in hidden state i, and it has been observed that
I(1),I(2),…I(t)Probability, can be calculated by formula 1;Indicate S(t)Current value corresponding to state, S(t)It is calculated by formula 5
It arrives;
σ and A 3c4) are obtained according to the solution of formula 7:
Wherein, T is total length of testing speech;T indicates the testing time;I(t)For current value measured by t moment;S(t)Indicate t
Hidden state locating for moment;Indicate S(t)Current value corresponding to state is calculated by formula 8:
Wherein, T is total length of testing speech;T indicates the testing time;I(t)For current value measured by t moment;S(t)Indicate t
Hidden state locating for moment, is calculated by formula 5;||·||BooleanFor Boolean formula, the value 1 when condition is set up works as item
Value 0 when part is invalid.
The method of Random telegraph noise signal for said extracted with trap coupling effect, further, the convergence
Threshold value is set as 10-3~10-6;When (hidden state is corresponding by iterative parameter K (state transition probability matrix), σ (amplitude of noise), A
Current amplitude) knots modification when being respectively less than set convergence threshold, terminate iteration.
Compared with prior art, the beneficial effects of the present invention are:
The present invention provides a kind of Random telegraph noise letter based on the extraction of hidden Markov model with trap coupling effect
Number method, for from actual measurement in noisy leakage current signal, extract it is clean and with coupling information with electromechanics
Report noise.This extracting method at the extraction, does not require an assumption that mutually indepedent between trap, and therefore, the RTN of extraction can be true
Influencing each other between real reflection trap, i.e. coupling effect between trap.In addition, in the RTN of actual measurement, since test window has
Limit, often there is the phenomenon that state missing, such case can not be extracted by current technology, and this method due to it is not assumed that state it is complete
Whole property so will not be lacked by state is influenced, and can be realized and extract RTN clean and with coupling from the signal of actual measurement
Signal.
Detailed description of the invention
Fig. 1 is the schematic diagram of different RTN signals.
Fig. 2 is the Random telegraph noise provided by the invention for being extracted based on Hidden Markov Model and having trap coupling effect
The flow diagram of the extracting method of signal.
Fig. 3 is the leakage current signal graph of the nMOSFET surveyed in the embodiment of the present invention.
Fig. 4 is to work as N in the embodiment of the present inventiontrapWhen=2, the schematic diagram of RTN model calculation process and major parameter relationship.
Fig. 5 is to work as N in the embodiment of the present inventiontrapIt when=2, is restrained after iteration 23 times, obtained extraction result schematic diagram.
Fig. 6 be in the embodiment of the present invention Bayesian information criterion BIC with trap number NtrapThe tendency chart of variation.
Specific embodiment
With reference to the accompanying drawing, the present invention, the model of but do not limit the invention in any way are further described by embodiment
It encloses.
One ideal clean RTN signal and as shown in Figure 1 with noisy RTN signal.The present invention provides a kind of band
The extracting method of the Random telegraph noise signal of trap coupling effect, this method is based on hidden Markov model, from the band of actual measurement
In noisy leakage current signal, Random telegraph noise that is clean and having coupling information is extracted.
Flow chart such as Fig. 2 institute of the extracting method of Random telegraph noise signal provided by the invention with trap coupling effect
Show, includes the following steps:
One) by directly testing, the leakage current signal of certain grid voltage and leakage pressure is obtained;
Two) according to signal the case where sets the range N of trap number that may be presenttrap=1~Nmax.Small size device
In trap number it is fewer, can generally take Nmax=3.To each possible NtrapValue all carries out primary complete modeling
It extracts;It comprises the following processes:
1) it is based on HMM, establishes the model of RTN;
According to trap number Ntrap, since each trap can produce two hidden states, NtrapA trap, at most may be used
To generate Nstate=2NtrapA hidden state, although this 2NtrapA hidden state can all might not occur in test window.Into one
Step ground, the hidden state space of available RTN, it is possible to occur in measured signal hidden state i ∈ [1,2 ... Nstate],
The corresponding current value of each hidden state is denoted as A=[A1,A2,...ANstate].The hidden state of each of hidden state space, it is corresponding
NtrapOne group of a trap may occupancy.Due to the presence of Gaussian noise in test process, reality is observed on hidden state i
The current value O measurediWith randomness, meet Gaussian Profile, i.e. Oi~N (Ai,σ2), σ is the amplitude of Gaussian noise.
According to trap occupancy corresponding to hidden state space and each hidden state, the state between state and state is obtained
Transition probability matrix K;Each of matrix K element KijIt is meant that in the unit time and is transferred to hidden state i's from hidden state j
Probability.Since the generating state conversion simultaneously of two traps is the event of minimum probability, each hidden state can only be transferred to
In addition NtrapA hidden state, i.e. K have certain sparsity.
2) initial value of RTN model is determined;
According to leakage current signal I(1),I(2),...I(T)Histogram determine that each hidden state is corresponding using clustering method
Current amplitudeAnd accordingly, it tentatively extracts RTN signal and obtains the corresponding hidden state S under each moment t(t),
And thus calculate transition probability intensity matrix K and noise amplitude σ.
3) RTN model is trained using feedforward-rear feed algorithm;
3a) feedforward arithmetic
Utilize recurrence formula:
Computation model is located at hidden state i in t moment, and has produced actual measurement sequence I(1)~I(t)Probability ai
(t).Wherein, KijIndicate the probability for being transferred to hidden state i in unit testing time interval from hidden state j;The width of σ expression noise
Degree;T indicates the testing time;I(t)Indicate electric current measured by t moment;AiIndicate current value corresponding to hidden state i.
3b) rear feed algorithm
Utilize recurrence formula:
Computation model is located at hidden state i in t moment, and has produced actual measurement sequence I(t)~I(T)Probability bi
(t).Wherein, KjiIndicate the probability for being transferred to hidden state j in unit testing time interval from hidden state i;The width of σ expression noise
Degree;T indicates the testing time;T indicates total testing time;I(t)Indicate electric current measured by t moment;AiIndicate that hidden state i institute is right
The current value answered;Last current state is the T moment, and locating hidden state, general, last current state can be denoted as 1, extracts result to final
Influence it is little.
3c) solve RTN model parameter
According to generalized expectation-maximization, following 3c1 is constantly repeated) to 3c4) step, model parameter is iteratively solved, until iteration
As a result meet iteration convergence condition, i.e. between two step iteration steps, iterative parameter K (state transition probability matrix), the σ (width of noise
Degree), the knots modification of A (hidden state corresponding current range value) be respectively less than set convergence threshold;Convergence threshold mentioning depending on needed for
Depending on taking precision, it generally may be configured as 10-3, when needing high-precision, can be set as 10-6:
Formula 1 and formula 2 3c1) are substituted into formula 3, acquire intermediate variable γij(t):
γij(t) it is meant that, t moment, the probability of hidden state j is transferred to by hidden state i.Wherein, ai(t-1) t-1 is indicated
Moment is in hidden state i, and it has been observed that I(1),I(2),…I(t-1)Probability, can be calculated by formula 1;KjiIndicate that unit is surveyed
The probability of hidden state j is transferred in examination time interval from hidden state i;bj(t) in t moment, it is located at hidden state j, and has seen
Measure actual measurement sequence I(t)~I(T)Probability, can be calculated by formula 2.
3c2) formula 3 is substituted into following iterative formula, solution matrix K, wherein the letter of equal sign "=" left end adds above
Symbol ^, expression are the values of current iteration step, to distinguish preceding iteration step Chinese style 1, formula 2, the K in formula 3:
Wherein, γij(t) it indicates t moment, the probability of hidden state j is transferred to by hidden state i, can be calculated by formula 3.
3c3) determine moment t=1, when 2 ... T, hidden state S locating for RTN(t)And corresponding RTN signal RTN (t):
Wherein, S(t)Indicate hidden state locating for t moment;ai(t) t moment is indicated, in hidden state i, and it has been observed that
I(1),I(2),…I(t)Probability, can be calculated by formula 1;Indicate S(t)Current value corresponding to state, S(t)It is calculated by formula 5
It arrives.
3c4) solve σ and A, wherein the letter of equal sign "=" left end has added symbol ^ above, and expression is current iteration step
Value, to distinguish in preceding iteration step, i.e. formula 1, σ and A in formula 2:
Wherein, T is total length of testing speech;T indicates the testing time;I(t)For current value measured by t moment;S(t)Indicate t
Hidden state, S locating for moment(t)Gained is calculated by formula 5;Indicate S(t)Current value corresponding to state.
Wherein, T is total length of testing speech;T indicates the testing time;I(t)For current value measured by t moment;S(t)Indicate t
Hidden state locating for moment, can be calculated by formula 5;||·||BooleanFor Boolean formula, condition sets up value 1, condition not at
Vertical value 0.
3d) according to the RTN signal RTN (t) that can be extracted after iteration convergence by formula 6.
Three) to each NtrapExtraction under value is as a result, calculate separately Bayesian information criterion BIC using following formula:
Wherein, NtrapFor trap number;T is testing time overall length;S(T)It indicates hidden state locating for the T moment, is counted by formula 5
It obtains;For the T moment, it is in hidden state S(T), and observe sequence I(1),I(2),…I(T)Probability, calculated by formula 1
It obtains.
Four) N corresponding to Bayesian information criterion BIC minimum valuetrapUnder that group extraction result RTN (t), as very
Real RTN signal.
The entire extraction process of the RTN signal dependent on hidden Markov model is completed as a result,.
Following embodiment actual measurement at fixed grid voltage 0.9V and fixed leakage pressure 0.9V obtains nMOSFET leakage current signal, base
RTN signal is extracted in hidden Markov model, specific implementation step is as follows:
One) actual measurement obtains the leakage current signal of fixed grid voltage 0.9V and the nMOSFET under fixed leakage pressure 0.9V, and signal is lasting
Time 1000s, as shown in Figure 3;
Two) leakage current signal according to Fig.3, it can be seen that the step of variation is few, illustrates that trap number is less, because
This present embodiment sets maximum trap number NmaxIt is 3, i.e. consideration trap number NtrapIt is 1,2,3 three kinds of situations, and carries out respectively
It extracts, step below is with NtrapIn case where=2:
1) by Ntrap=2, therefore the hidden state i of RTN has 4 kinds of possible cases, is denoted as [1,2,3,4], respectively corresponds two and falls into
Trap all prevents take up, only trap 1 occupies, only trap 2 occupies, two traps all occupy these four situations.Four hidden states point
Not Dui Ying a current value, be denoted as amplitude A (A1~A4);OiFor the corresponding current observation of hidden state i;Fig. 4 is implementation of the present invention
In example as Ntrap=2, major parameter and its schematic diagram of relationship in RTN model.Since trap 1 and trap 2 will not be sent out simultaneously
Raw overturning, therefore the state transition probability matrix K between hidden state are as follows:
Wherein, in the state transition probability matrix K between hidden state, K12It indicates to be jumped in unit interval by hidden state 2
To the probability of hidden state 1;Remaining element is similarly in K.
2) clustering method is first used, primary just extraction is carried out to signal, RTN model parameter A, K are determined according to first extract, σ's
Initial value is respectively as follows:
σ=1.31E-6;
3) solution is iterated to RTN model parameter using feedforward-rear feed algorithm, iteration repeatedly restrains afterwards, the present embodiment
The number of iterations is 23;Obtained extraction result RTN (t) is as shown in figure 5, RTN model parameter are as follows:
σ=0.9E-6;
Three) to Ntrap=1 and Ntrap=3 both situations, execute step 2 respectively) operation carry out model extraction, calculate
Each NtrapBayesian information criterion BIC value under value, as a result as shown in fig. 6, working as NtrapWhen=2, BIC is minimum.Therefore,
Ntrap=2 be most reasonable trap number assumption value;
Four) final RTN is extracted as a result, as NtrapExtraction result RTN (t) when=2, i.e., shown in Fig. 5.
It should be noted that the purpose for publicizing and implementing example is to help to further understand the present invention, but the skill of this field
Art personnel, which are understood that, not to be departed from the present invention and spirit and scope of the appended claims, and various substitutions and modifications are all
It is possible.Therefore, the present invention should not be limited to embodiment disclosure of that, and the scope of protection of present invention is with claim
Subject to the range that book defines.
Claims (5)
1. a kind of method for extracting the Random telegraph noise signal with trap coupling effect, the extracting method are based on hidden Ma Erke
Husband's model, from actual measurement in noisy leakage current signal, extraction obtains clean and with coupling information random telegraph and makes an uproar
Sound;Include the following steps:
One) the leakage current signal of setting grid voltage and leakage pressure is obtained by actual measurement;
Two) the leakage current signal obtained according to sets the trap number N in devicetrapRange be 1~Nmax; NmaxFor
Maximum trap number;For each NtrapValue carries out primary complete modeling and extracts, obtains the NtrapIt is worth corresponding with electromechanics
It reports noise signal RTN (t);The modeling extraction process specifically:
1) Random telegraph noise model is established based on hidden Markov model, obtains the state transition probability matrix between hidden state
K;Each of matrix K element KijIt is the probability for being transferred to hidden state i in the unit time from hidden state j;Hidden state number is
NState,
2) it according to the histogram of leakage current signal, determines the corresponding current amplitude of each hidden state, is tentatively mentioned according to current amplitude
Random telegraph noise signal is taken, corresponding hidden state S under each moment t is obtained(t), then state transition probability intensity is calculated
Matrix K and Gaussian noise amplitude σ, the initial value as Random telegraph noise model;
3) Random telegraph noise model is trained using feedforward-rear feed algorithm, sets convergence threshold, by iteratively solving institute
Random telegraph noise model is stated, obtains model parameter, including state transition probability intensity matrix K, Gaussian noise amplitude σ and is not deposited
Current value A under noise situations;Random telegraph noise signal RTN (t) is obtained by the extraction of formula 6 after iteration convergence;
In formula 6, S(t)Indicate hidden state locating for t moment;Indicate S(t)Current value corresponding to state, S(t)It is calculated by formula 5
It arrives:
In formula 5, ai(t) it indicates t moment, be in hidden state i and it has been observed that I(1),I(2),…I(t)Probability, calculated by formula 1
It obtains:
In formula 1, each of matrix K element KijIt is the probability for being transferred to hidden state i in the unit time from hidden state j;σ is to make an uproar
Sound amplitude;T indicates the testing time;I(t)Indicate electric current measured by t moment;AiIndicate current value corresponding to hidden state i;
Three) respectively to each NtrapValue, is calculated the N using formula 9trapIt is worth corresponding Bayesian information criterion BIC:
Wherein, NtrapFor trap number;T is testing time overall length;S(T)It indicates hidden state locating for the T moment, is calculated by formula 5
It arrives;For the T moment, it is in hidden stateS (T), and observe sequence I(1),I(2),…I(T)Probability, be calculated by formula 1;
Four) by N corresponding to Bayesian information criterion BIC minimum valuetrapUnder extraction result RTN (t) as really with electromechanics
Noise signal is reported, the extraction process of the Random telegraph noise signal dependent on hidden Markov model is thus completed.
2. extracting the method for the Random telegraph noise signal with trap coupling effect as described in claim 1, characterized in that step
One) the setting grid voltage and leakage pressure are fixed grid voltage 0.9V and fixed leakage pressure 0.9V;Step 2) the maximum trap number Nmax
It is set as 3.
3. extracting the method for the Random telegraph noise signal with trap coupling effect as described in claim 1, characterized in that step
3) described that Random telegraph noise model is trained using feedforward-rear feed algorithm, it is made an uproar by iteratively solving the random telegraph
Acoustic model obtains model parameter, specifically comprises the following steps:
3a) by feedforward arithmetic, model is calculated using formula 1 and is located at hidden state i in t moment, and has produced actual measurement
Sequence I(1)~I(t)Probability ai(t);
3b) by rear feed algorithm, model is calculated using formula 2 and is located at hidden state i in t moment, and has produced actual measurement
Sequence I(t)~I(T)Probability bi(t):
In formula 2, KjiIndicate the probability for being transferred to hidden state j in unit testing time interval from hidden state i;The width of σ expression noise
Degree;T indicates the testing time;T indicates total testing time;I(t)Indicate electric current measured by t moment;AiIndicate that hidden state i institute is right
The current value answered;Last current state is the T moment;
Model parameter 3c) is obtained by iteratively solving the Random telegraph noise model according to generalized expectation-maximization;
Obtained Random telegraph noise signal RTN (t) 3d) is extracted by formula 6 according to after iteration convergence.
4. extracting the method for the Random telegraph noise signal with trap coupling effect as claimed in claim 3, characterized in that step
It is 3c) described by iteratively solving the Random telegraph noise model, model parameter is obtained, specifically: constantly repeating step 3c1)
~3c4) iterative solution model parameter, the knot when the iterative parameter knots modification between two step iteration steps is respectively less than the convergence threshold
Beam iteration:
Formula 1 and formula 2 3c1) are substituted into formula 3, acquire intermediate variable γij(t):
In formula 3, γij(t) it indicates t moment, the probability of hidden state j is transferred to by hidden state i;ai(t-1) t-1 moment, place are indicated
In hidden state i, and it has been observed that I(1),I(2),…I(t-1)Probability, can be calculated by formula 1;KjiIndicate the unit testing time
The probability of hidden state j is transferred in interval from hidden state i;bj(t) in t moment, it is located at hidden state j, and it has been observed that real
Sequencing column I(t)~I(T)Probability, can be calculated by formula 2;
3c2) formula 3 is substituted into formula 4, solution obtains parameter K:
In formula 4, γij(t) it indicates t moment, the probability of hidden state j is transferred to by hidden state i, can be calculated by formula 3;4 equal sign of formula
Put in marks ^ above the letter of left end, and expression is the value of current iteration step;
3c3) according to formula 5 and formula 6, it is calculated moment t=1, when 2 ... T, hidden state S locating for Random telegraph noise(t)With it is right
The Random telegraph noise signal RTN (t) answered:
Wherein, S(t)Indicate hidden state locating for t moment;ai(t) it indicates t moment, is in hidden state i, and it has been observed that I(1),I(2),…I(t)Probability, can be calculated by formula 1;Indicate S(t)Current value corresponding to state, S(t)It is calculated by formula 5;
σ and A 3c4) are obtained according to the solution of formula 7:
Wherein, T is total length of testing speech;T indicates the testing time;I(t)For current value measured by t moment;S(t)Indicate t moment
Locating hidden state;Indicate S(t)Current value corresponding to state is calculated by formula 8:
Wherein, T is total length of testing speech;T indicates the testing time;I(t)For current value measured by t moment;S(t)Indicate t moment
Locating hidden state, is calculated by formula 5;||·||BooleanFor Boolean formula, the value 1 when condition is set up, when condition not
Value 0 when establishment.
5. extracting the method for the Random telegraph noise signal with trap coupling effect as claimed in claim 4, characterized in that described
Iterative parameter knots modification includes the knots modification of the knots modification of K, the knots modification of σ and A;The convergence threshold is set as 10-3~10-6。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610183059.8A CN105787473B (en) | 2016-03-28 | 2016-03-28 | A method of extracting the Random telegraph noise signal with trap coupling effect |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610183059.8A CN105787473B (en) | 2016-03-28 | 2016-03-28 | A method of extracting the Random telegraph noise signal with trap coupling effect |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105787473A CN105787473A (en) | 2016-07-20 |
CN105787473B true CN105787473B (en) | 2019-01-29 |
Family
ID=56392216
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610183059.8A Active CN105787473B (en) | 2016-03-28 | 2016-03-28 | A method of extracting the Random telegraph noise signal with trap coupling effect |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105787473B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107292061A (en) * | 2017-07-28 | 2017-10-24 | 西安交通大学 | A kind of process industry complex electromechanical systems information modelling approach of data-driven |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7203635B2 (en) * | 2002-06-27 | 2007-04-10 | Microsoft Corporation | Layered models for context awareness |
RU2598601C2 (en) * | 2012-03-30 | 2016-09-27 | Сони Корпорейшн | Data processing device, data processing method and software |
CN103607219B (en) * | 2013-11-07 | 2016-02-03 | 电子科技大学 | A kind of noise prediction method of electric line communication system |
CN104501832A (en) * | 2014-12-02 | 2015-04-08 | 中国航天科工集团第三研究院第八三五七研究所 | Improved practical inertial sensor noise reduction device |
-
2016
- 2016-03-28 CN CN201610183059.8A patent/CN105787473B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN105787473A (en) | 2016-07-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10290066B2 (en) | Method and device for modeling a long-time-scale photovoltaic output time sequence | |
Zhang et al. | Sparse target counting and localization in sensor networks based on compressive sensing | |
CN104766175A (en) | Power system abnormal data identifying and correcting method based on time series analysis | |
CN106896306B (en) | A kind of GIS oscillatory surge pressure test signal antinoise method | |
CN101614786B (en) | Online intelligent fault diagnosis method of power electronic circuit based on FRFT and IFSVC | |
Xue et al. | Empirical likelihood for single-index varying-coefficient models | |
CN104376231B (en) | Based on the damnification recognition method for improving approximate Bayes's calculating | |
CN109102032A (en) | A kind of pumping plant unit diagnostic method based on depth forest and oneself coding | |
Chen et al. | Multiple fault condition recognition of gearbox with sequential hypothesis test | |
CN110969194B (en) | Cable early fault positioning method based on improved convolutional neural network | |
CN103064008A (en) | Nolinear analog circuit soft fault diagnostic method based on Hilbert-huang transform | |
CN103439413A (en) | Acoustic emission signal analytical method for automatically identifying damage mode of thermal barrier coating | |
CN104050604A (en) | Electric power system static safety assessment method based on probabilistic tide | |
CN105138770B (en) | Space product Reliablility simulation appraisal procedure based on indirect characteristic quantities | |
CN112465124B (en) | Twin depth space-time neural network model acquisition/fault diagnosis method and device | |
Hu et al. | Fault feature recovery with Wasserstein generative adversarial imputation network with gradient penalty for rotating machine health monitoring under signal loss condition | |
CN103543026B (en) | Structural damage identification method based on vibration transfer rate function and support vector machine | |
Tupikina et al. | Characterizing the evolution of climate networks | |
CN105228185B (en) | Method used for identifying identity of fuzzy redundant node in communication network | |
Lin et al. | Engine components fault diagnosis using an improved method of deep belief networks | |
CN112231775B (en) | Hardware Trojan horse detection method based on Adaboost algorithm | |
CN105787473B (en) | A method of extracting the Random telegraph noise signal with trap coupling effect | |
CN105823801B (en) | A kind of electronic nose drift compensation method based on deepness belief network feature extraction | |
CN103219998A (en) | Hybrid parameter estimation method for use under multi-channel compressed sensing framework | |
CN103093094A (en) | Software failure time forecasting method based on kernel partial least squares regression algorithm |
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 | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20220620 Address after: 201210 room 122, building 1, No. 200, Jichuang road and No. 491, Yindong Road, China (Shanghai) pilot Free Trade Zone, Pudong New Area, Shanghai Patentee after: Shanghai Lunke Electronic Technology Co.,Ltd. Address before: 100871 No. 5, the Summer Palace Road, Beijing, Haidian District Patentee before: Peking University |