CN110443223A - A kind of signal automatic Modulation classification method and system based on K-means - Google Patents
A kind of signal automatic Modulation classification method and system based on K-means Download PDFInfo
- Publication number
- CN110443223A CN110443223A CN201910748250.6A CN201910748250A CN110443223A CN 110443223 A CN110443223 A CN 110443223A CN 201910748250 A CN201910748250 A CN 201910748250A CN 110443223 A CN110443223 A CN 110443223A
- Authority
- CN
- China
- Prior art keywords
- signal
- standard deviation
- lfm
- bpsk
- 16qam
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention discloses a kind of signal automatic Modulation classification method and system based on K-means, its modulation signature is extracted using signal processing algorithms such as time frequency analysis, instantaneous auto-correlation and Fourier transformations, machine learning algorithm based on naive Bayesian and K-means realizes the layer-by-layer classification of the radar and communications signal of different modulating type.Simulation result shows that the multi-signal sorting algorithm network can effectively realize the Modulation recognition of 6 kinds of different modulating types.
Description
Technical field
The invention belongs to radars and electronic technology field, and in particular to a kind of signal automatic Modulation based on K-means point
Class method and system.
Background technique
Radar is recognized to need to realize perception to external electromagnetic environment, including realize for detect the other radars received and
The Classification and Identification of signal of communication.To the modulation type Classification and Identification field of emitter Signals, there are the following problems at present:
(1) when parameter generation signal is arranged, influence of some parameters to algorithm classification performance is not accounted for, sample is caused
This amount is smaller.
(2) only consider a type of modulated signal, and practical radar emitter signal has more modulation type.
(3) a variety of differences between signal are not accounted for, a variety of different characteristics are extracted, realize more letters according only to a kind of feature
Number classification, increase classification difficulty.
(4) feature extracting method is more single, does not focus on the expansion in terms of feature extracting method.
Summary of the invention
It is an object of the invention to overcome deficiency in the prior art, it is automatic to propose a kind of signal based on K-means
Modulation classification method realizes the layer-by-layer classification of different modulating type signal.
In order to solve the above technical problems, the present invention provides a kind of signal automatic Modulation classification method based on K-means,
It is characterized in that including following procedure:
Determine different modulating type signal, including SF, LFM, BPSK, QPSK, 16QAM and 2FSK signal；
The standard deviation that the time-frequency plane crest frequency first derivative of 6 signals is obtained using Short Time Fourier Transform, is based on
6 signals are divided into two major class { SF, LFM } and { BPSK, QPSK, 16QAM, 2FSK } by standard deviation；
Classified using Timed automata to SF signal and LFM signal；
Using instantaneous auto-correlation real part standard deviation, spectrum peak number and envelope standard deviation characteristic to BPSK, QPSK,
16QAM and 2FSK classify.
Further, based on standard deviation by 6 signals be divided into two major class { SF, LFM } with BPSK, QPSK, 16QAM,
2FSK } detailed process are as follows:
Standard deviation β based on each signal peak frequency first derivative1, set thresholding ε1If β1< ε1It is { SF, LFM } letter
Number, β1≥ε1It is { BPSK, QPSK, 16QAM, 2FSK } signal.
Further, carrying out classification to SF signal and LFM signal using Timed automata includes:
Timed automata β based on SF and LFM signal2, set thresholding ε2If β2< ε2, then it is judged as SF signal, if β2≥
ε2, then it is judged as LFM signal.
Further, using instantaneous auto-correlation real part standard deviation, spectrum peak number and envelope standard deviation characteristic to BPSK,
The detailed process that QPSK, 16QAM and 2FSK classify are as follows:
The instantaneous auto-correlation of 4 kinds of signals, the information for extracting instantaneous auto-correlation real part standard deviation are joined as one group of feature respectively
Number, is denoted as β3；
Envelope standard deviation is extracted as one group of characteristic parameter, is denoted as β4；
Discrete Fourier transform is done to 4 kinds of signals, the information of spectrum peak number is extracted as one group of characteristic parameter, is denoted as
β5；
Three-dimensional feature vector plane space { β is constructed using this three groups of characteristic parameters3,β4,β5, in conjunction with K-means cluster point
Similar object is formed one kind by the algorithm of analysis, identifies BPSK, QPSK, 16QAM and 2FSK signal.
Correspondingly, the present invention also provides a kind of signal automatic Modulation categorizing system based on K-means, characterized in that
Categorization module on the right side of signal type determining module, first layer categorization module, second layer left side categorization module and the second layer；
Signal type determining module, for determining different modulating type signal, including SF, LFM, BPSK, QPSK, 16QAM
With 2FSK signal；
First layer categorization module, for obtaining the time-frequency plane crest frequency one of 6 signals using Short Time Fourier Transform
The standard deviation of order derivative, based on standard deviation by 6 signals be divided into two major class { SF, LFM } with BPSK, QPSK, 16QAM,
2FSK}；
Categorization module on the left of the second layer, for being classified using Timed automata to SF signal and LFM signal；
Categorization module on the right side of the second layer, for utilizing instantaneous auto-correlation real part standard deviation, spectrum peak number and envelope mark
Quasi- difference feature classifies to BPSK, QPSK, 16QAM and 2FSK.
Further, in first layer categorization module, based on standard deviation by 6 signals be divided into two major class { SF, LFM } with
The detailed process of { BPSK, QPSK, 16QAM, 2FSK } are as follows:
Standard deviation β based on each signal peak frequency first derivative1, set thresholding ε1If β1< ε1It is { SF, LFM } letter
Number, β1≥ε1It is { BPSK, QPSK, 16QAM, 2FSK } signal.
Further, classified using Timed automata to SF signal and LFM signal in categorization module on the left of the second layer
Include:
Timed automata β based on SF and LFM signal2, set thresholding ε2If β2< ε2, then it is judged as SF signal, if β2≥
ε2, then it is judged as LFM signal.
Further, on the right side of the second layer in categorization module, using instantaneous auto-correlation real part standard deviation, spectrum peak number and
The detailed process that envelope standard deviation characteristic classifies to BPSK, QPSK, 16QAM and 2FSK are as follows:
The instantaneous auto-correlation of 4 kinds of signals, the information for extracting instantaneous auto-correlation real part standard deviation are joined as one group of feature respectively
Number, is denoted as β3；
Envelope standard deviation is extracted as one group of characteristic parameter, is denoted as β4；
Discrete Fourier transform is done to 4 kinds of signals, the information of spectrum peak number is extracted as one group of characteristic parameter, is denoted as
β5；
Three-dimensional feature vector plane space { β is constructed using this three groups of characteristic parameters3,β4,β5, in conjunction with K-means cluster point
Similar object is formed one kind by the algorithm of analysis, identifies BPSK, QPSK, 16QAM and 2FSK signal.
Compared with prior art, the beneficial effects obtained by the present invention are as follows being: the present invention utilizes time frequency analysis, instantaneous auto-correlation
Its modulation signature is extracted with signal processing algorithms such as Fourier transformations, the machine learning based on naive Bayesian and K-means is calculated
Method realizes the layer-by-layer classification of the radar and communications signal of different modulating type.Simulation result shows the multi-signal sorting algorithm
Network can effectively realize the Modulation recognition of 6 kinds of different modulating types.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention；
Fig. 2 is signal time-frequency contour map: (a) SF signal time-frequency contour map；(b) LFM signal time-frequency contour map；
(c) bpsk signal time-frequency contour map；(d) QPSK signal time-frequency contour map；(e) 16QAM signal time-frequency contour map；(f)
2FSK signal time-frequency contour map；
Fig. 3 is instantaneous auto-correlation real axis perspective view: (a) the instantaneous auto-correlation real axis perspective view of BPSK；(b) QPSK is instantaneously from phase
Close real axis perspective view；
Fig. 4 is signal spectrum figure: (a) bpsk signal spectrogram；(b) 2FSK signal spectrum figure；
Fig. 5 is the characteristic profile of the standard deviation of crest frequency first derivative；
Fig. 6 is envelope standard deviation characteristic figure；
Fig. 7 is instantaneous auto-correlation real part standard deviation characteristic figure；
Fig. 8 is spectrum peak number distribution map；
Fig. 9 is the classification results under 20dB signal-to-noise ratio.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.Following embodiment is only used for clearly illustrating the present invention
Technical solution, and not intended to limit the protection scope of the present invention.
The present invention utilizes time frequency analysis, wink for the received other radar and communications signals of cognition reconnaissance system for radar institute
When the signal processing algorithms such as auto-correlation and Fourier transformation extract its modulation signature, the machine based on naive Bayesian and K-means
Device learning algorithm realizes the layer-by-layer classification of the radar and communications signal of different modulating type.Simulation result shows the multi signal
Sorting algorithm network can effectively realize the Modulation recognition of 6 kinds of different modulating types.
A kind of signal automatic Modulation classification method based on K-means of the invention, realizes the classification of signal modulation style,
It is shown in Figure 1, including following procedure:
S1 determines different modulating type signal, including SF, LFM, BPSK, QPSK, 16QAM and 2FSK signal.
Cognition radar is being modulated classification of type Study of recognition first to external electromagnetic environment signal (emitter Signals)
It needs to be determined that certain signal form set.Different signal types has different time-frequencies, envelope and phase property, needs to make
Feature extraction algorithm and signal sorting algorithm are all not quite similar.For the ease of feature extraction algorithm and Modulation recognition machine
The determination of algorithm is practised, the present invention targeted communication and radar signal include: SF, LFM, BPSK, QPSK, 16QAM, 2FSK signal
Totally 6 kinds of signal forms.
Signal automatic Modulation classification method, flow diagram are as shown in Figure 1.Since signal kinds are more, multiple features are used
The method of multistratum classification.
First layer classification: 6 kinds of signal sets are divided by S2 first using Short Time Fourier Transform and NB Algorithm
Two major class { SF, LFM } and { BPSK, QPSK, 16QAM, 2FSK }.
There is larger difference in the first derivative standard deviation that above-mentioned two classes signal time-frequency characteristics are utilized.In short-term in Fu
Leaf transformation basic thought is signal to be added to time slip-window, and do Fourier transformation to signal in window, obtains the when frequency conversion of signal
Spectrum.The Short Time Fourier Transform formula of signal x (n) are as follows:
N is sampling number in formula, and l is time delay, and g (n) is window function, g*(n-l) conjugate operation is indicated.
6 kinds of signals do the frequency information of window section when obtaining each after Short Time Fourier Transform, time-frequency contour map such as Fig. 2 institute
Show.By Fig. 2 (a) it is found that the frequency of SF signal remains unchanged, the crest frequency difference of window section is 0 when adjacent, standard deviation 0；By
Fig. 2 (b) is it is found that the frequency of LFM signal changes linearly, and the crest frequency difference of window section remains unchanged when adjacent, and standard deviation is
0；By Fig. 2 (c), (d) with (e) it is found that since there are the changes of instantaneous frequency in phase hit for BPSK, QPSK and 16QAM signal
Change, the crest frequency difference of window section has size fluctuating when adjacent, and standard deviation is larger；By Fig. 2 (f) it is found that there are two types of 2FSK signals
The variation of frequency, the crest frequency difference of window section is influenced by two kinds of frequencies when adjacent, and standard deviation is larger.Therefore, it can extract phase
The standard deviation of window section crest frequency poor (crest frequency first derivative) classifies to 6 kinds of signals when adjacent.
The crest frequency of window section, expression formula when finding out each after signal Short Time Fourier Transform first are as follows:
F=max | ξ (l, n) | } (2)
In formula, F for window sometimes crest frequency set, adjacent peak frequency first derivative expression formula are as follows:
Δ F (i)=F (i+1)-F (i), i=1,2 ..., M-1 (3)
Standard deviation β is asked to crest frequency first derivative1, expression formula are as follows:
Window number when M is in formula, the crest frequency of window section when F (i) is i-th,For crest frequency
The mean value of first derivative.
In the standard deviation β for finding out each signal peak frequency first derivative1Afterwards, divided based on Nae Bayesianmethod
Class.Classifying step is as follows:
The classification set C={ y of 6 kinds of radar and communications signals is defined first1,y2, y1It indicates { SF, LFM }, y2It indicates
{ BPSK, QPSK, 16QAM, 2FSK }.
(1) characteristic attribute and division: characteristic attribute β are determined1Indicate the standard deviation of crest frequency first derivative.It provides and draws
Point: β1:{β1< ε1,β1≥ε1, classification thresholding ε is determined according to training set1, classify set y1And y2。
(2) obtain training sample: 6 kinds of Signal-to-Noise variations are 10dB, 15dB and 20dB, the parameter setting of various signals
As shown in table 1.
Table is arranged in 1 signal parameter of table
According to the parameter setting of table 1,1200 training samples are constructed altogether.
(3) sample characteristics attribute value: the characteristic profile of the standard deviation of the crest frequency first derivative of training sample is calculated
As shown in Figure 5.As seen from the figure, the standard deviation of the crest frequency first derivative of SF signal and LFM signal is close to 0, and remaining 4 kinds
Signal is not 0.The variation range of BPSK, QPSK and 16QAM signal is larger, and 2FSK signal intensity is smaller.
(4) characteristic attribute demarcation interval is determined: according to distribution map as shown in Figure 5, thresholding ε1The range that can be selected 0~
Between 0.6258, thresholding ε is determined1For the median 0.3129 of demarcation interval.
That is, in the standard deviation β for finding out each signal peak frequency first derivative1Afterwards, thresholding ε is set1If β1< ε1
It is { SF, LFM } signal, β1≥ε1It is { BPSK, QPSK, 16QAM, 2FSK } signal.
S3, second layer classification left-hand branch: using Timed automata and NB Algorithm to SF signal and LFM signal
Classify.
By { SF, LFM } and { BPSK, QPSK, 16QAM, 2FSK } after first layer differentiation, in Fig. 1 second layer left-hand branch
Realize the classification of SF and LFM signal.Since the Timed automata of SF signal is definite value 1, LFM signal is by carrying out carrier frequency
(because frequency of carrier signal is constant, LFM signal converts carrier frequency, changes linearly its frequency, i.e. carrier frequency for modulation
Rate modulation), the transmitted bandwidth of carrier signal is increased, Timed automata is greater than 1, and the area of 2 kinds of signals can be realized with this feature
Point.According to this difference, Timed automata is found out, and SF signal and LFM signal are distinguished in conjunction with NB Algorithm.
Timed automata β2Expression formula are as follows:
β2=Bw·Tw (5)
In formula, BwFor the bandwidth of signal, TwFor the time width of signal.
In the Timed automata β for finding out SF and LFM signal2Afterwards, classify based on Nae Bayesianmethod.Classification step
It is rapid as follows:
The classification set C={ y of SF signal and LFM signal is defined first3,y4, y3Indicate SF signal, y4Indicate LFM letter
Number.
(1) characteristic attribute and division: characteristic attribute β are determined2It indicates Timed automata, provides and divide β2:{β2< ε2,β2≥
ε2, thresholding ε is determined according to training set2, set y3And y4。
(2) obtain training sample: the variation of the signal-to-noise ratio of SF signal and LFM signal is 10dB, 15dB and 20dB, parameter setting
As shown in table 1,400 training samples are constructed altogether.
(3) calculate sample characteristics attribute value: the Timed automata of training sample is sought by formula (5).SF signal is wide
Bandwidth product is definite value 1, and LFM signal Timed automata is not 1 and variation range is larger.According to calculated result, LFM signal when
Wide bandwidth accumulates value from 500~2000.
(4) characteristic attribute demarcation interval is determined: according to the Timed automata acquired, thresholding ε2It can be selected between 1~500,
Since Timed automata variation range is larger, thresholding ε is determined2For the median 250 of demarcation interval.
That is, in the Timed automata β for finding out SF and LFM signal2Afterwards, thresholding ε is set2If β2< ε2, then judge
For SF signal, if β2≥ε2, then it is judged as LFM signal.
Second layer classification right-hand branch: S4 utilizes instantaneous auto-correlation real part standard deviation, spectrum peak number, envelope standard
3 kinds of features such as difference distinguish BPSK, QPSK, 16QAM and 2FSK.
Specific practice is: calculating the instantaneous auto-correlation of 4 kinds of signals, extracts the information conduct of instantaneous auto-correlation real part standard deviation
One group of characteristic parameter, is denoted as β3.Envelope standard deviation is extracted as one group of characteristic parameter, is denoted as β4.Direct computation of DFT is done to 4 kinds of signals
Leaf transformation extracts the information of spectrum peak number as one group of characteristic parameter, is denoted as β5.Finally, utilizing this three groups of characteristic parameters
Construct three-dimensional feature vector plane space { β3,β4,β5, similar object is formed one in conjunction with the algorithm of K-means clustering
Class identifies BPSK, QPSK, 16QAM and 2FSK signal.The initial mean value needed for the training stage, progress K-means algorithm
It seeks, classifies to training sample, obtain optimal mean value center.In test phase, using the optimal mean value trained as survey
The initial mean value of sample sheet utilizes K-means algorithm again, final to realize signal modulation style classification.
(1) instantaneous autocorrelation characteristic extracts
Receiving signal is x (n), instantaneous auto-correlation B (n, l) expression formula are as follows:
B (n, l)=x (n) x*(n-l) (6)
In formula, l is time delay, and B (n, l) does not have time integral, remains the prompting message of relevant treatment.
Psk signal expression formula are as follows:
A is amplitude, f in formula0For carrier frequency, ΦiIndicate phase.The instantaneous auto-correlation expression formula of psk signal are as follows:
Corresponding real part expression formula are as follows:
In formula m be symbol width in sampling number, time delay l < m, by formula (9) it is found that bpsk signal has 0, π, two kinds of phases,
Cos (2 π f in same symbol0It l) is certain value, instantaneous auto-correlation real part output is direct current.Phase hit between adjacent symbol
Φi+1-ΦiWhen being 0, i.e. cos (2 π f0L+0)=cos (0)=1, instantaneous auto-correlation real part show positive value jump；Work as Φi+1-
ΦiWhen for π, i.e. cos (2 π f0L+ π)=cos (π)=- 1, instantaneous auto-correlation real part shows negative value and jumps, therefore bpsk signal
Instantaneous auto-correlation real part output shows as the jump of 2 values.And QPSK hasFour kinds of phases, cos (2 π f in same symbol0l)
For certain value, instantaneous auto-correlation real part output is direct current.As the phase hit Φ between adjacent symboli+1-ΦiWhen being 0, i.e. cos (2
πf0L+0)=cos (0)=1, instantaneous auto-correlation real part show positive value jump；Work as Φi+1-ΦiWhen for π, i.e. cos (2 π f0l+
π)=cos (π)=- 1, instantaneous auto-correlation real part show negative value jump；Work as Φi+1-ΦiForWhen, i.e.,Its instantaneous auto-correlation real part shows zero and jumps, therefore QPSK signal transient auto-correlation real part
Output shows as the jump of 3 values.
QAM signal expression are as follows:
A in formulaiFor amplitude, QAM signal transient auto-correlation expression formula are as follows:
Corresponding real part expression formula are as follows:
16QAM has the characteristics that several degree variations of leggy.By formula (12) it is found that A in same symbol2cos(2πf0L) it is
Definite value, instantaneous auto-correlation real part output is direct current.Real part output between adjacent symbol generates Φi+1-ΦiA variety of phase hits
And Ai+1·AiThe jump of a variety of amplitudes, therefore 16QAM instantaneous auto-correlation real part output shows as multivalue jump.
2FSK signal expression are as follows:
F in formulaiFor carrier frequency, there are two types of frequencies to change for 2FSK signal, Φ0For initial phase.Instantaneous auto-correlation expression formula
Are as follows:
Corresponding real part expression formula are as follows:
By formula (15) it is found that cos (2 π f in same symbol0It l) is definite value, instantaneous auto-correlation real part output is direct current.It is adjacent
Between subcode, instantaneous auto-correlation real part output is the jump signal modulated by adjacent frequency difference.
According to above-mentioned analysis, the extractable differentiation that signal is realized based on the feature of instantaneous auto-correlation real part.The wink of signal
When auto-correlation real axis perspective view as shown in figure 3, by Fig. 3 (a) it is found that bpsk signal real part jump be 2 values, can by Fig. 3 (b)
Know, the real part jump of QPSK signal is 3 values, and the real part of 16QAM is multivalue jump, and the real part of 2FSK is to be modulated by adjacent frequency difference
Jump signal, also show as multivalue jump, therefore 4 kinds of signal real part standard deviation sizes have differences, and can choose instantaneous auto-correlation
Real part standard deviation is as one group of characteristic parameter, expression formula are as follows:
In formula, R (n) indicates instantaneous auto-correlation real part,For its mean value.
(2) envelope standard deviation characteristic is extracted
Since 16QAM has the characteristics that several degree jumps, and BPSK, QPSK and 2FSK do not have the characteristics of amplitude jumps,
Therefore envelope standard deviation can be chosen as characteristic parameter, to distinguish 16QAM signal and { BPSK, QPSK, 2FSK } signal.Envelope
Standard deviation expression formula are as follows:
In formulaFor the envelope of signal,For the mean value of signal envelope.
(3) spectrum peak feature extraction
Due to BPSK, QPSK and 16QAM spectrum peak number have it is multiple, and by amount of bandwidth influenced variation range compared with
Greatly, and 2FSK spectrum peak number maintains essentially in 2.The spectrogram of BPSK and 2FSK is as shown in figure 4, setting thresholding ε3, Fig. 4
(a) bpsk signal can detect that multiple spectrum peaks, the 2FSK signal of Fig. 4 (b) can detect that peak value number be 2, QPSK and
16QAM is similar with BPSK.Peak value number can be chosen as one group of characteristic parameter using this feature.Signal x (n) is done first
Discrete Fourier transform, expression formula are
It chooses at 0.707 times of maximum value as detection threshold ε3, spectrum peak number expression formula is
(4) first with instantaneous auto-correlation real part standard deviation β3, envelope standard deviation β4With spectrum peak number β5Three spies
Sign constructs three-dimensional feature space, distinguishes 2FSK signal, BPSK, QPSK and 16QAM in space utilization K-means clustering algorithm
Signal.
K-means algorithm is using error sum of squares criterion function as the objective function of cluster, error sum of squares criterion function
It is defined as
In formula, k is cluster number of clusters, LiFor the set of sample in i-th of cluster, ρ indicates the sample point in the cluster, μiExpression pair
The central point of cluster is answered, dist indicates that the calculation of sample point and central point distance is Euclidean distance.In sample pointρIt is constant
In the case where, the size of error sum of squares criterion function and the central point μ of clusteriIt is related, make in the smallest cluster of J to acquire
The heart, can J to derivation, enabling derivative is 0, and expression formula is
By formula (22) it is found that making the mean value of each point in the smallest cluster centre point cluster of error squared criterion function.K-
Cluster number of clusters and initial mean value need to be arranged in means algorithm in advance.
Classifying step based on K-means is as follows:
First with instantaneous auto-correlation real part standard deviation β3, envelope standard deviation β4With spectrum peak number β5Three features, structure
Build three-dimensional feature space, space utilization K-means clustering algorithm distinguish 2FSK signal, BPSK, QPSK and 16QAM signal,
Realize the classification of Fig. 3 second layer right-hand branch.Sample set D={ β3,β4,β5, cluster number of clusters k=4, corresponding 4 initial mean values
Vector is { μ1,μ2,μ3,μ4}。L1,L2,L3,L4The corresponding cluster of BPSK, QPSK, 16QAM and 2FSK signal is respectively indicated, when initial
It is empty set.The following steps are recycled, until reaching stop condition:
1) it calculates distance: calculating separately the Euclidean distance of each sample point ρ to 4 mean vector
2) sorted out according to apart from nearest criterion: if
min{dist(ρ,μ1),dist(ρ,μ2),dist(ρ,μ3),dist(ρ,μ4)=dist (ρ, μ1) (24)
Then ρ ∈ L1, i.e. the sample point belongs to bpsk signal, and excess-three kind situation is similar.
3) it generates new mean vector: recalculating each cluster LiMean vector, generate new mean vector
4) it is unchanged to judge that new and old mean vector has, mean vector is updated if changing, new mean vector replaces
Old mean vector.If mean vector no longer changes, illustrates that clustering criteria function J has restrained, export the classification knot of 4 kinds of signals
Fruit L1,L2,L3,L4。
Embodiment
Using the recognition performance of Computer Simulation verifying multi-signal sorting algorithm.To generate based on parameter ergodic theorem
Data sample, signal under the conditions of trained and test phase generates different parameters, carries out feature extraction, realizes classification.It is anti-
Only excessively caused sample size is excessive and the very few caused sample covering of Parameters variation is not comprehensive for Parameters variation, main herein
The variation for considering the parameters such as the signal-to-noise ratio, bandwidth and the chip rate that are affected to classification influences little sampling frequency on classification
The parameters such as rate and carrier frequency remain unchanged.
SF, LFM, BPSK, QPSK, 16QAM and 2FSK signal are generated according to parameter setting shown in table 1, by first layer
Classifying step realizes the first layer classification of the standard deviation based on NB Algorithm and crest frequency first derivative, distinguishes
{ SF, LFM } and { BPSK, QPSK, 16QAM, 2FSK } then presses the classifying step on the left of the second layer and realizes based on naive Bayesian
The classification of the second layer left-hand branch of algorithm and Timed automata is distinguished SF signal and LFM signal, is divided on the right side of the second last layer
Zhi Liyong K-means algorithm realizes classification, distinguishes BPSK, QPSK, 16QAM and 2FSK signal.Main introduce is distinguished below
The simulation process of BPSK, QPSK, 16QAM and 2FSK signal.
In the training stage, the signal-to-noise ratio variation of 4 kinds of signals is 10dB, 15dB and 20dB, is constructed altogether by the parameter setting of table 1
800 training samples.It is jumped according to 16QAM signal with amplitude, envelope standard deviation is larger, and other 3 kinds of signals do not have width
Degree jump, envelope standard deviation is smaller, takes envelope to 4 kinds of signals, extracts the information of envelope standard deviation as one group of characteristic parameter,
Characteristic profile as shown in fig. 6, as seen from the figure, the envelope standard deviation size of 4 kinds of signals is influenced have certain variation by signal-to-noise ratio,
In the case where different signal-to-noise ratio, the envelope standard deviation of 16QAM has differences with another three kinds of signals；Further according to 4 kinds of signal transients
4 kinds of signals are done instantaneous auto-correlation, extract the information conduct of instantaneous auto-correlation real part standard deviation by the difference of auto-correlation real part output
One group of characteristic parameter, characteristic profile is as shown in fig. 7, as seen from the figure, the size of 4 kinds of signal transient auto-correlation real part standard deviations
It is influenced by signal-to-noise ratio variation, under different state of signal-to-noise, instantaneous auto-correlation real part standard deviation and the QPSK signal of BPSK is deposited
In difference；Then according to the difference of 4 kinds of signal spectrum peak values, Fourier transformation is done to signal, extracts the letter of spectrum peak number
Breath is used as one group of characteristic parameter, and characteristic profile is as shown in figure 8, as seen from the figure, the spectrum peak of BPSK, QPSK and 16QAM
Number variation range is larger, and the spectrum peak number of 2FSK signal maintains 2；Finally, utilizing this three groups of characteristic parameter structures
Three-dimensional feature vector plane space is built, realizes that Modulation recognition, classification results are as shown in Figure 9 in conjunction with K-means algorithm.It can from figure
Find out, constructs three-dimensional union feature using instantaneous auto-correlation real part standard deviation, envelope standard deviation and spectrum peak number, can incite somebody to action
BPSK, QPSK, 16QAM and 2FSK are distinguished.
In test phase, test sample, performance of the testing algorithm under different signal-to-noise ratio are generated.
Under different state of signal-to-noise, the classification accuracy rate of SF, LFM, BPSK, QPSK, 16QAM and 2FSK signal such as table 2
It is shown.As shown in Table 2, each Modulation recognition accuracy increases with the increase of signal-to-noise ratio, when signal-to-noise ratio reaches 15dB, the classification
Structural behaviour is preferable.
The classification accuracy rate of SF, LFM, BPSK, QPSK, 16QAM and 2FSK under the different signal-to-noise ratio of table 2
The method of the present invention extracts its modulation using signal processing algorithms such as time frequency analysis, instantaneous auto-correlation and Fourier transformations
Feature, the machine learning algorithm based on naive Bayesian and K-means realize the radar and communications signal of different modulating type
Layer-by-layer classification.Simulation result shows that the multi-signal sorting algorithm network can effectively realize the signal point of 6 kinds of different modulating types
Class.
Correspondingly, the present invention also provides a kind of signal automatic Modulation categorizing system based on K-means, characterized in that
Categorization module on the right side of signal type determining module, first layer categorization module, second layer left side categorization module and the second layer；
Signal type determining module, for determining different modulating type signal, including SF, LFM, BPSK, QPSK, 16QAM
With 2FSK signal；
First layer categorization module, for obtaining the time-frequency plane crest frequency one of 6 signals using Short Time Fourier Transform
The standard deviation of order derivative, based on standard deviation by 6 signals be divided into two major class { SF, LFM } with BPSK, QPSK, 16QAM,
2FSK}；
Categorization module on the left of the second layer, for being classified using Timed automata to SF signal and LFM signal；
Categorization module on the right side of the second layer, for utilizing instantaneous auto-correlation real part standard deviation, spectrum peak number and envelope mark
Quasi- difference feature classifies to BPSK, QPSK, 16QAM and 2FSK.
Further, in first layer categorization module, based on standard deviation by 6 signals be divided into two major class { SF, LFM } with
The detailed process of { BPSK, QPSK, 16QAM, 2FSK } are as follows:
Standard deviation β based on each signal peak frequency first derivative1, set thresholding ε1If β1< ε1It is { SF, LFM } letter
Number, β1≥ε1It is { BPSK, QPSK, 16QAM, 2FSK } signal.
Further, classified using Timed automata to SF signal and LFM signal in categorization module on the left of the second layer
Include:
Timed automata β based on SF and LFM signal2, set thresholding ε2If β2< ε2, then it is judged as SF signal, if β2≥
ε2, then it is judged as LFM signal.
Further, on the right side of the second layer in categorization module, using instantaneous auto-correlation real part standard deviation, spectrum peak number and
The detailed process that envelope standard deviation characteristic classifies to BPSK, QPSK, 16QAM and 2FSK are as follows:
The instantaneous auto-correlation of 4 kinds of signals, the information for extracting instantaneous auto-correlation real part standard deviation are joined as one group of feature respectively
Number, is denoted as β3；
Envelope standard deviation is extracted as one group of characteristic parameter, is denoted as β4；
Discrete Fourier transform is done to 4 kinds of signals, the information of spectrum peak number is extracted as one group of characteristic parameter, is denoted as
β5；
Three-dimensional feature vector plane space { β is constructed using this three groups of characteristic parameters3,β4,β5, in conjunction with K-means cluster point
Similar object is formed one kind by the algorithm of analysis, identifies BPSK, QPSK, 16QAM and 2FSK signal.
It should be understood by those skilled in the art that, embodiments herein can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the application
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the application, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
The application is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present application
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the technical principles of the invention, several improvements and modifications, these improvements and modifications can also be made
Also it should be regarded as protection scope of the present invention.
Claims (8)
1. a kind of signal automatic Modulation classification method based on K-means, characterized in that including following procedure:
Determine different modulating type signal, including SF, LFM, BPSK, QPSK, 16QAM and 2FSK signal；
The standard deviation of the time-frequency plane crest frequency first derivative of 6 signals is obtained using Short Time Fourier Transform, is based on standard
6 signals are divided into two major class { SF, LFM } and { BPSK, QPSK, 16QAM, 2FSK } by difference；
Classified using Timed automata to SF signal and LFM signal；
Using instantaneous auto-correlation real part standard deviation, spectrum peak number and envelope standard deviation characteristic to BPSK, QPSK, 16QAM and
2FSK classifies.
2. a kind of signal automatic Modulation classification method based on K-means according to claim 1, characterized in that be based on
6 signals are divided into the detailed process of two major class { SF, LFM } and { BPSK, QPSK, 16QAM, 2FSK } by standard deviation are as follows:
Standard deviation β based on each signal peak frequency first derivative1, set thresholding ε1If β1< ε1It is { SF, LFM } signal, β1
≥ε1It is { BPSK, QPSK, 16QAM, 2FSK } signal.
3. a kind of signal automatic Modulation classification method based on K-means according to claim 1, characterized in that utilize
Timed automata carries out classification to SF signal and LFM signal
Timed automata β based on SF and LFM signal2, set thresholding ε2If β2< ε2, then it is judged as SF signal, if β2≥ε2,
Then it is judged as LFM signal.
4. a kind of signal automatic Modulation classification method based on K-means according to claim 1, characterized in that utilize
Instantaneous auto-correlation real part standard deviation, spectrum peak number and envelope standard deviation characteristic carry out BPSK, QPSK, 16QAM and 2FSK
The detailed process of classification are as follows:
The instantaneous auto-correlation of 4 kinds of signals respectively extracts the information of instantaneous auto-correlation real part standard deviation as one group of characteristic parameter, note
For β3；
Envelope standard deviation is extracted as one group of characteristic parameter, is denoted as β4；
Discrete Fourier transform is done to 4 kinds of signals, the information of spectrum peak number is extracted as one group of characteristic parameter, is denoted as β5；
Three-dimensional feature vector plane space { β is constructed using this three groups of characteristic parameters3,β4,β5, in conjunction with K-means clustering
Similar object is formed one kind by algorithm, identifies BPSK, QPSK, 16QAM and 2FSK signal.
5. a kind of signal automatic Modulation categorizing system based on K-means, characterized in that signal type determining module, first layer
Categorization module and second layer right side categorization module on the left of categorization module, the second layer；
Signal type determining module, for determining different modulating type signal, including SF, LFM, BPSK, QPSK, 16QAM and
2FSK signal；
First layer categorization module, the time-frequency plane crest frequency single order for obtaining 6 signals using Short Time Fourier Transform are led
6 signals are divided into two major class { SF, LFM } and { BPSK, QPSK, 16QAM, 2FSK } based on standard deviation by several standard deviations；
Categorization module on the left of the second layer, for being classified using Timed automata to SF signal and LFM signal；
Categorization module on the right side of the second layer, for utilizing instantaneous auto-correlation real part standard deviation, spectrum peak number and envelope standard deviation
Feature classifies to BPSK, QPSK, 16QAM and 2FSK.
6. a kind of signal automatic Modulation categorizing system based on K-means according to claim 5, characterized in that first
In layer categorization module, 6 signals are divided by two major class { SF, LFM } and { BPSK, QPSK, 16QAM, 2FSK } based on standard deviation
Detailed process are as follows:
Standard deviation β based on each signal peak frequency first derivative1, set thresholding ε1If β1< ε1It is { SF, LFM } signal, β1
≥ε1It is { BPSK, QPSK, 16QAM, 2FSK } signal.
7. a kind of signal automatic Modulation categorizing system based on K-means according to claim 5, characterized in that second
In the categorization module of layer left side, carrying out classification to SF signal and LFM signal using Timed automata includes:
Timed automata β based on SF and LFM signal2, set thresholding ε2If β2< ε2, then it is judged as SF signal, if β2≥ε2,
Then it is judged as LFM signal.
8. a kind of signal automatic Modulation categorizing system based on K-means according to claim 5, characterized in that second
Layer right side categorization module in, using instantaneous auto-correlation real part standard deviation, spectrum peak number and envelope standard deviation characteristic to BPSK,
The detailed process that QPSK, 16QAM and 2FSK classify are as follows:
The instantaneous auto-correlation of 4 kinds of signals respectively extracts the information of instantaneous auto-correlation real part standard deviation as one group of characteristic parameter, note
For β3；
Envelope standard deviation is extracted as one group of characteristic parameter, is denoted as β4；
Discrete Fourier transform is done to 4 kinds of signals, the information of spectrum peak number is extracted as one group of characteristic parameter, is denoted as β5；
Three-dimensional feature vector plane space { β is constructed using this three groups of characteristic parameters3,β4,β5, in conjunction with K-means clustering
Similar object is formed one kind by algorithm, identifies BPSK, QPSK, 16QAM and 2FSK signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910748250.6A CN110443223A (en) | 2019-08-14 | 2019-08-14 | A kind of signal automatic Modulation classification method and system based on K-means |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910748250.6A CN110443223A (en) | 2019-08-14 | 2019-08-14 | A kind of signal automatic Modulation classification method and system based on K-means |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110443223A true CN110443223A (en) | 2019-11-12 |
Family
ID=68435291
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910748250.6A Withdrawn CN110443223A (en) | 2019-08-14 | 2019-08-14 | A kind of signal automatic Modulation classification method and system based on K-means |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110443223A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112769722A (en) * | 2021-01-05 | 2021-05-07 | 河海大学 | Automatic identification method and system for communication signal modulation type |
| CN112859025A (en) * | 2021-01-05 | 2021-05-28 | 河海大学 | Radar signal modulation type classification method based on hybrid network |
-
2019
- 2019-08-14 CN CN201910748250.6A patent/CN110443223A/en not_active Withdrawn
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112769722A (en) * | 2021-01-05 | 2021-05-07 | 河海大学 | Automatic identification method and system for communication signal modulation type |
| CN112859025A (en) * | 2021-01-05 | 2021-05-28 | 河海大学 | Radar signal modulation type classification method based on hybrid network |
| CN112859025B (en) * | 2021-01-05 | 2023-12-01 | 河海大学 | Radar signal modulation type classification method based on hybrid network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109802905B (en) | CNN convolutional neural network-based digital signal automatic modulation identification method | |
| CN108600135A (en) | A kind of recognition methods of signal modulation mode | |
| CN105119862B (en) | A kind of identification of signal modulation method and system | |
| CN102760444B (en) | Support vector machine based classification method of base-band time-domain voice-frequency signal | |
| CN107682109B (en) | A kind of interference signal classifying identification method suitable for UAV Communication system | |
| CN111832462B (en) | Frequency hopping signal detection and parameter estimation method based on deep neural network | |
| CN110443223A (en) | A kind of signal automatic Modulation classification method and system based on K-means | |
| Davies et al. | Deep neural networks for appliance transient classification | |
| EP3826203A1 (en) | Signal detection device and signal detection method | |
| CN110232371A (en) | High-precision HRRP Radar Multi Target recognition methods based on small sample | |
| CN108470155A (en) | A kind of extensive stream data processing method of Radar emitter individual identification | |
| CN109656366A (en) | Emotional state identification method and device, computer equipment and storage medium | |
| CN107067022B (en) | Method, device and equipment for establishing image classification model | |
| CN112749633B (en) | Separate and reconstructed individual radiation source identification method | |
| CN109088837A (en) | A kind of many kinds of radar and automatic recognition of communication signals based on clustering | |
| Norouzi et al. | Adaptive modulation recognition based on the evolutionary algorithms | |
| CN108809874B (en) | Radar and communication multi-signal classification method based on circulation support vector machine | |
| CN104091178A (en) | Method for training human body sensing classifier based on HOG features | |
| CN109034087B (en) | PCA (principal component analysis) dimension reduction-based hybrid machine learning signal classification method | |
| CN113452637B (en) | Underwater acoustic communication signal modulation identification method based on feature selection and support vector machine | |
| CN113449682B (en) | Method for identifying radio frequency fingerprints in civil aviation field based on dynamic fusion model | |
| CN109472216A (en) | Radiation source feature extraction and individual discrimination method based on signal non-Gaussian feature | |
| Huynh-The et al. | Chain-Net: Learning deep model for modulation classification under synthetic channel impairment | |
| CN102006252A (en) | Single-tone signal identification method | |
| CN114615007B (en) | Tunnel mixed flow classification method and system based on random forest |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20191112 |