CN111814777A

CN111814777A - Modulation pattern recognition method based on characteristic quantity grading

Info

Publication number: CN111814777A
Application number: CN202010963725.6A
Authority: CN
Inventors: 朱江; 姜南; 杨虎; 胡登鹏; 高凯; 杨军; 李二保; 朱立; 王新建
Original assignee: Hunan Guoke Ruicheng Electronic Technology Co ltd
Current assignee: Hunan Guoke Ruicheng Electronic Technology Co ltd
Priority date: 2020-09-15
Filing date: 2020-09-15
Publication date: 2020-10-23
Anticipated expiration: 2040-09-15
Also published as: CN111814777B

Abstract

The invention discloses a modulation pattern recognition method based on characteristic quantity grading, which comprises the following steps: selecting a characteristic quantity to be used according to an identification signal set, and ensuring that the characteristic quantity covers all signals to be identified in the identification signal set; grading the sensitivity and complexity of signal-to-noise ratio and frequency offset according to the characteristic quantity; determining the judgment sequence of the decision tree according to the distinguished characteristic quantity series and the distinguished signal identification type; and step four, sequentially judging from top to bottom according to the judging sequence, outputting the signal type if the signals are identified, and adopting the next-stage characteristic quantity to identify if the signals are not identified until all the signals in the identification signal set are identified. The method can improve the signal identification probability of the signal under the conditions of low signal-to-noise ratio and multiple signal types by carrying out multiple identification on one hand, and reduces the complexity of the realization of the whole identification system by grading the characteristic quantity according to the complexity on the other hand.

Description

Modulation pattern recognition method based on characteristic quantity grading

Technical Field

The invention relates to the technical field of wireless communication reconnaissance, in particular to a modulation pattern recognition method based on characteristic quantity grading.

Background

The method for identifying the modulation pattern based on the characteristic quantity mainly comprises two steps: firstly, the calculation of the characteristic quantity and secondly the classification and identification of the signals. The characteristic quantity is compared and selected according to the signal type in the signal set, and the signal type is identified by adopting a certain classifier based on the value of the characteristic quantity. The basic flow of signal identification is as follows: firstly, selecting characteristic quantity according to a signal set, determining a decision threshold, then determining a classifier according to the characteristic quantity and the signal set, finally calculating the characteristic quantity of a signal to be detected, and carrying out signal identification according to the classifier. The feature quantities and the classifiers will be described separately below.

The most commonly used feature quantities mainly include instantaneous features, transform domain features, statistical features, constellation features and the like.

1) Temporal characteristics

The instantaneous characteristic is mainly related to the instantaneous amplitude, phase and frequency of the signal, and the three parameters are used for further calculation to derive more characteristic values. Non-patent document 1 realizes identification of a low-order digitally modulated signal by extracting a characteristic parameter based on instantaneous information. Non-patent document 2 realizes classification of 9 common modulation signals, i.e., 2ASK, 4ASK, 8ASK, 2FSK, 4FSK, 8FSK, 2PSK, 4PSK, and 8PSK, based on temporal characteristics formed by parameters such as temporal amplitude, frequency, and phase. Transient characteristics are easy to extract, the implementation complexity is low, but the transient characteristics are easily influenced by noise and are more robust only for part of the modulation pattern.

2) Transform domain characterization

The transform domain mainly comprises a Fourier transform domain and a wavelet transform domain. By converting the signal into the transform domain and subjecting it to processing such as smoothing, normalization and median filtering, the eigenvalues of the transform domain can be obtained. Through Fourier transform, the signal can be converted from a time domain to a frequency domain, and the frequency domain can reflect the change of the modulation signal parameters more intuitively, so that some characteristics of the frequency domain can be used for signal modulation classification. Wavelet transform is a commonly used feature extraction technique that has the advantage of reducing the effects of noise. The features extracted from the wavelet transform domain contain both signal time domain information and frequency domain information, and thus are more robust to modulation classification than fourier transforms. Non-patent document 3 creatively applies wavelet transform to modulation recognition, and performs classification recognition on MPSK and MFSK signals. Non-patent document 4 proposes to use two characteristic parameters, namely, a curve kurtosis and a ratio of a variance of a wavelet transform coefficient to an absolute value of a mean value, to realize classification and identification of three signals, namely FM, MSK and QPSK, when a signal-to-noise ratio is greater than 4 dB.

3) Statistical features

The statistical characteristics include High Order Moments (HOMs), High Order Cumulants (HOC), high Order Cyclic Cumulants (HOC), Cyclic stationarity, autocorrelation characteristics, and the like. Higher Order Statistics (HOS) can be classified into HOM and HOC, which are widely used in the classification of ASK, PSK and QAM signals. The high-order statistic has the advantages of reflecting the high-order statistic characteristics of the signal, eliminating noise influence, having good anti-phase rotation performance and the like. Non-patent document 5 proposes a feature parameter based on signal statistical characteristics and transient characteristics, and combines with a back propagation neural network, to solve the problem of classification and identification of classical digital modulation and binary offset carrier and derivative modulation thereof under a low signal-to-noise ratio, thereby realizing classification and identification with higher accuracy and speed when the signal-to-noise ratio is 3 dB. Non-patent document 6 proposes a joint feature parameter based on a high-order cumulant and a fractal box dimension, which realizes classification and identification of 7 modulation signals, and the algorithm has a low computational complexity and can have a high identification rate even when the signal-to-noise ratio is low and the number of samples is small. Non-patent document 7 proposes, for classification and identification of three signals, namely BPSK, QPSK and 16QAM, in an underwater channel environment including fading, doppler shift, phase noise and additive white gaussian noise, to first classify the signals into BPSK and non-BPSK signals by using a second-order cyclostationary feature, and then further identify the QPSK signal from the remaining signals by using a maximum likelihood detection method. Non-patent document 8 discusses a new modulation classification method under a multipath fading channel, and the algorithm processes channel impulse response by establishing a relationship between the cumulant of the received signal and the multipath fading effect by using the normalized fourth-order cumulant, thereby realizing the classification and identification of MPSK and MQAM signals.

4) Constellation features

The shape characteristics of the constellation can be obtained by, for example, calculating the number of constellation points. The constellation diagrams of PSK and QAM signals can reflect the modulation schemes of the PSK and QAM signals, and the modulation parameters can be obtained through the amplitude and phase information of each constellation point. Non-patent document 9 classifies and identifies BPSK, QPSK, 8PSK, v.29f, v.29, and v.32f based on a constellation fuzzy C-means clustering algorithm. Non-patent document 10 proposes a constellation-based method for achieving PSK and QAM signal identification, which is not very complex but has limited applicability.

The above four types of identification features are characterized by different application occasions and signal types. The signal characteristic quantity based on the instantaneous characteristic is easy to obtain, and the realization complexity is low. However, for different signals which are affected by noise differently, the method classifies the characteristic quantities from low implementation complexity, and completes the identification of the signals through multiple judgments.

The classifiers that are widely used mainly include 3 types: decision trees, neural networks and support vector machines.

1) Decision tree

The decision tree classifier is a relatively simple classifier, and the idea is to divide data to be classified into different subclasses by setting a threshold value, and then continue to set the threshold value for the different subclasses for classification. Non-patent document 11 studies that 10 digital modulation signals, such as 2ASK, 4ASK, 8ASK, 2FSK, 4FSK, 8FSK, 4PSK, 8PSK, 16QAM, and 32QAM, are classified and identified by using a combined feature parameter based on a decision tree classifier, and the identification rate can reach 100% when the signal-to-noise ratio is greater than 5 dB. The decision tree is easy to realize in practice and flexible to expand, when the types of data to be classified are increased, only the branches of the existing decision tree need to be expanded without redesigning the whole decision tree, and the realization complexity is relatively low, so that the decision tree is used more.

2) Artificial neural network

An artificial neural network is one of the most commonly used techniques in a signal classifier, and is convenient to implement due to its flexible structure. And, the artificial neural network can adapt to and process complex signals through self-learning. Non-patent document 12 implements classification identification of FSK, PSK, ASK, and QAM signals using hierarchical neural networks, in which a first network identifies a modulation class and a second group of networks identifies a constellation order of the modulation class. Non-patent document 13 realizes classification of AM, DSB, FM, 2FSK, 4FSK, BPSK, and QPSK signals by using a BP neural network in combination with three feature vectors extracted by a spectral correlation function, and the recognition rate is 94% or more when the signal-to-noise ratio is greater than 0 dB.

The learning and training modes of the artificial neural network can be divided into two types, one is supervised learning, and the other is unsupervised learning. In the supervised mode, the input data can comprise additional training data, while in the unsupervised mode, only a learning mode and relevant rules are specified, and the specific learning content is different according to the input signals, so that the system can automatically search for rules and characteristics. Clearly, the results obtained by supervised learning are more accurate. But has the disadvantage that it requires a large amount of training data. Most artificial neural networks for signal classification employ supervised learning techniques such as multi-layer perceptrons and radial basis functions. However, unsupervised artificial neural network techniques, such as self-organizing map networks, have also been used for automatic classification of signals.

3) Support vector machine

The principle of the support vector machine is to find an optimal plane or hyperplane, so that the distance between the plane and the data point is maximized, and the points closest to the classification plane are called support vectors. In the case where data points are linearly inseparable, the data points of different classes can be effectively separated by mapping the low-dimensional data to the high-dimensional data by using a kernel function. Non-patent document 14 proposes an algorithm for realizing automatic modulation recognition of a signal by using a support vector machine based on statistical characteristic parameters, and the algorithm has high robustness in a wide signal-to-noise ratio range.

Because the artificial neural network and the support vector machine both need to be trained greatly and the implementation complexity is high, the method is mainly a decision tree method in engineering application at present, and therefore the method is mainly designed aiming at the decision tree method.

After the adopted characteristic quantity and the classifier are determined, the signal to be identified can be identified. In the current signal identification method, the adopted characteristic quantity is used as the identification basis of a certain signal or a certain type of signal, and the signal is judged to be the signal if the characteristic quantity is in accordance with the signal or is judged to be the other signal if the characteristic quantity is not in accordance with the signal by setting a threshold value. When the identification signal set is small, the signal can be accurately identified, but when the signal set is large, the identification effect is poor, and the threshold selection is difficult.

Prior art documents

Non-patent document

Non-patent document 1: nandi A K, Azzouz E. Algorithms for automatic registration of communications signals [ J ]. Biulleten Eksperimentno ĭ biologici I meditsinny, 1998, 37(7):23-35.

Non-patent document 2: moser E, Moran M K, Hillen E, et al, Automatic modulation via in-stance deficiencies [ C ]// Aerospace and electronics reference IEEE, 2016: 218-.

Non-patent document 3: ho K C, Prokopiw W, Chan Y T Modulation identification of digital signals by the wave transmission [ J ]. IEE procedures-Radar, Sonarand Navigation, 2002, 147(4): 169-.

Non-patent document 4: xijihui, Zhang Yiwen, Zhaoyaxin, etc. ultrashort wave signal modulation identification based on kurtosis and wavelet transform [ J ] modern electronic technology, 2016, 39(23):9-12.

Non-patent document 5: zhou Q, Lu H, Jia L, et al, Automatic modulation with genetic feedback neural network [ C ]// evolution calculation. IEEE, 2016:4626-4633.

Non-patent document 6: the application of high-order cumulant and fractal theory in signal modulation and identification is researched in J signal processing, 2013, 29(6):761-765.

Non-patent document 7: sanderson J, Li X, Liu Z, et al, Hierarchical Blanking modulation Classification for the purpose of the Underwater acoustical Communication Signal visual circulation and maximum Likeliod Acoustic Analysis [ C ]// Military Communication conference, Milcom 2013-IEEE 2014:29-34.

Non-patent document 8: chang D C, Shih P K, Cumulants-based modulation detection technique in multiple channels [ J ]. Communications Iet, 2015, 9(6): 828-.

Non-patent document 9: schreaygg C, Reichert J. Modulation classification of QAMSCHEME using the DFT of phase history combined with module information [ C ]// MILCOM 97 procedures IEEE, 2002: 1372-.

Non-patent document 10: wu Z, Like E, ChakraVarthy V. replaceable modulation at low SNR Using spectral correction [ C ]. Consumercommunications and Networking correction. IEEE, 2007: 1134.1138.

non-patent document 11: age stamp, digital communication signal automatic modulation recognition algorithm research [ D ]. university of Chongqing, 2012.

Non-patent document 12: naraghipour M. Robust modulation classification detection techniques cumulants and structural neural networks [ J ]. Proceedings of SPIE-The International Society for Optical Engineering, 2007:65671J-65671J-11.

Non-patent document 13: qian L, Zhu C. Modulation classification based on cyclic features and neural network [ C ]// International consistency on Image and Signal processing. IEEE, 2010: 3601-.

Non-patent document 14: zhang W. Automatic modulation classification based on statistical defects and Support Vector Machine [ C ]// general Assembly and scientific symposium. IEEE, 2014: 1-4.

Disclosure of Invention

In order to solve the problem that in the prior art, for some signals, part of characteristic quantity may appear under a certain condition, but the signals can be judged when the characteristic appears, and some characteristic quantity is not obvious enough for several signals and can only be judged by mutual comparison; on the other hand, when the signal set is large, the threshold of each characteristic quantity is very difficult to determine, and the implementation complexity of different characteristic quantities is not the same in the implementation process.

The technical scheme of the invention is as follows: a modulation pattern recognition method based on characteristic quantity grading comprises the following steps:

selecting a characteristic quantity to be used according to an identification signal set, and ensuring that the characteristic quantity covers all signals to be identified in the identification signal set;

grading the sensitivity and complexity of signal-to-noise ratio and frequency offset according to the characteristic quantity;

determining the judgment sequence of the decision tree according to the distinguished characteristic quantity series and the distinguished signal identification type;

and step four, sequentially judging from top to bottom according to the judging sequence, outputting the signal type if the signals are identified, and adopting the next-stage characteristic quantity to identify if the signals are not identified until all the signals in the identification signal set are identified.

Preferably, the main factors of the classification level for classifying the feature quantity are the signal set to be identified and the selected feature quantity, and the classification level includes a first-level feature quantity, a second-level feature quantity and a third-level feature quantity.

Preferably, the determination of ranking the feature amount is a step of:

firstly, judging a first-level characteristic quantity;

secondly, judging the characteristic quantity which is not identified by the signal of the first-level characteristic quantity by adopting the second-level characteristic quantity,

and finally, if the signal identification is not obtained, identifying by using a third-level characteristic quantity, and repeating the steps until all the signals to be identified in the identification signal set are completely identified.

Preferably, the signal identification set comprises 3 analog signals: AM, SSB, CW, and 9 digital signals: 2ASK, MSK, 2FSK, 4FSK, BPSK, QPSK, 8PSK, pi/4 DQPSK, 16 QAM.

Preferably, the feature quantity to be used selected according to the signal identification set includes the following hierarchical feature quantities:

a first stage: a bandwidth for distinguishing a single carrier from a modulated signal;

and a second stage: the single-peak spectrum line number of the salient pulses in the frequency spectrum comprises the single-peak spectrum line number of the salient pulses in various frequency spectrums of an amplitude square spectrum, a frequency spectrum, a square spectrum and a quartic spectrum and is used for identifying 2ASK, BPSK, pi/4 DQPSK and 8PSK signals and partial 2FSK and 4FSK signals;

and a third stage: the peak value number after the frequency spectrum smoothing filtering comprises the peak value number after the frequency spectrum, the square spectrum and the fourth power spectrum of the signal are smoothed and filtered; combining the number of single-peak spectral lines of the burst pulses in the frequency spectrum to complete the identification of the MSK signal and the identification of unidentified 2FSK and 4FSK signals;

fourth stage: and (3) signal amplitude distribution parameters are combined with the number of single-peak spectral lines of the outstanding pulses in each frequency spectrum to finish the identification of QPSK, 16QAM, SSB and AM signals.

Preferably, the signal identification process includes the following steps:

the first step is as follows: carrying out first-stage characteristic quantity single carrier and modulation signal identification on the signal identification set, judging the signal to be single carrier CW when the bandwidth of the identification signal after preprocessing is smaller than a threshold, and adopting second-stage characteristic quantity for identification when the bandwidth of the identification signal after preprocessing is larger than the threshold;

the second step is that: identifying 2ASK, BPSK, pi/4 QPSK and 8PSK signals according to the number of single peak spectral lines of the salient pulses in each frequency spectrum of the identification signals after the identification in the first step, and judging the signals to be 2ASK signals if the number of peak values of the peak spectral lines of the amplitude square spectrum salient peaks is 1 and the number of spectral lines of the peak spectral protrusions peaks is 1; if the number of peaks of the amplitude square spectrum peak value spectral line is 1, the number of lines of the spectrum peak value spectral line is 0, and the number of lines of the square spectrum peak value spectral line is 1, determining the signal as a BPSK signal; if the peak number of the amplitude square spectrum peak value spectral line is 1, the number of the spectrum peak value spectral lines is 0, the number of the square spectrum peak value spectral lines is 0, and the number of the quartic spectrum peak value spectral lines is 2, judging the signal to be a pi/4 DQPSK signal; if the peak number of the amplitude square spectrum peak value spectral line is 1, the number of the spectrum peak value spectral lines is 0, the number of the square spectrum peak value spectral lines is 0, the number of the quartic spectrum peak value spectral lines is 0, and the number of the octave spectrum peak value spectral lines is 1, the signal is judged to be an 8PSK signal; if the number of the outburst peak spectral lines in the frequency spectrum and the square spectrum is 2, judging the frequency spectrum and the square spectrum to be 2 FSK; if the number of the spectral lines of the outburst peak values in the frequency spectrum and the square spectrum is more than 2 and the position difference of the spectral peaks is equal, judging the frequency spectrum to be 4FSK, identifying a signal set after the identification is finished through the second-stage characteristic quantity, and then identifying by adopting the third-stage characteristic quantity;

the third step: identifying MSK, 2FSK and 4FSK by combining the identification signal set after the identification of the second-stage characteristic quantity with the number of spectral peak highlighting spectral lines and the number of peak values of envelopes after each time of spectral filtering, and judging MSK if the number of spectral peak highlighting spectral lines is 0, the number of square spectral peak highlighting spectral lines is 2 and the peak value of spectral filtering envelopes is 1; if the peak value of the spectrum and square spectrum filtering envelope is 2 or the peak value of the square spectrum filtering and fourth power spectrum envelope is 2, judging the frequency spectrum and square spectrum filtering envelope is 2 FSK; if the envelope peak value of the frequency spectrum and the square spectrum filtering is more than 2, or the envelope peak value of the square spectrum and the quartic spectrum filtering is more than 2, and the position difference between the envelope peak values is equal, judging the frequency spectrum and the square spectrum filtering to be 4FSK, otherwise judging the frequency spectrum and the square spectrum filtering to be 2 FSK; identifying the signal set after the identification is finished through the second-stage characteristic quantity, and then identifying through the fourth-stage characteristic quantity;

the fourth step: and identifying QPSK, 16QAM, SSB and AM signals by combining the identification signal set after the identification of the third-level characteristic quantity with the spectral peak highlighting value spectral line number and the signal amplitude distribution characteristic, if the amplitude square spectral highlighting peak value pulse spectral line peak value number is 1, the spectral peak value spectral line number and the square spectral highlighting pulse peak value spectral line number are 0, the quartic spectral highlighting pulse peak value spectral line number is 1, and the signal amplitude parameter is smaller than the threshold, judging the QPSK signal to be QPSK and judging the QPSK signal to be 16QAM when the signal amplitude parameter exceeds the threshold. And for the SSB and AM signals, if the number of spectral peak values of the spectrum highlighting pulses is greater than 0, judging the signals to be AM signals, otherwise, judging the signals to be SSB signals, and finishing the identification of all the signals in the signal identification set.

Compared with the prior art, the invention has the following beneficial effects:

on one hand, the method carries out multiple identification on the signal to be identified by grading the characteristic quantity, does not influence the identification probability of a single characteristic quantity, and can improve the correct signal identification probability of the whole system under the conditions of low signal-to-noise ratio and multiple signal identification types; on the other hand, by classifying the feature quantities according to complexity, since the feature quantity values are classified for calculation, all the feature quantities do not need to be calculated for most signals, and therefore, the calculation amount for realizing the system is reduced.

Drawings

FIG. 1 is a flow chart of a modulation pattern recognition method based on characteristic quantity grading according to the present invention;

FIG. 2 is a schematic diagram of a first signal identification process according to the present invention;

FIG. 3 is a schematic diagram of a second signal identification process according to the present invention;

FIG. 4 is a schematic diagram of a third step of signal identification according to the present invention;

FIG. 5 is a schematic diagram illustrating a fourth signal identification process according to the present invention;

fig. 6 shows the signal identification results of the present invention under different signal-to-noise ratios.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In the description of the present invention, it should be understood that the terms "front", "back", "left", "right", "up", "down", and the like indicate orientations or positional relationships based on those shown in the drawings, and are used merely for convenience in describing the present invention and simplifying the description, but do not indicate or imply that the devices or elements indicated by the terms must have specific orientations, be constructed and operated in specific orientations, and therefore, should not be construed as limiting the present invention.

Referring to fig. 1 to 6, the present invention provides the following technical solutions: a modulation pattern recognition method based on characteristic quantity grading comprises the following steps:

grading the characteristic quantity according to the sensitivity and complexity of the characteristic quantity to signal-to-noise ratio, frequency offset and the like;

And classifying the characteristic quantity according to factors such as calculated quantity, identification performance and the like, and then carrying out multiple times of judgment based on the decision tree to realize signal identification. It is considered that the feature quantities are divided into a plurality of stages,

the first stage is that the characteristic quantity of a signal modulation mode can be definitely determined through the characteristic quantity, the sensitivity degree to signal-to-noise ratio, frequency offset and the like is low, and the realization complexity is minimum.

The second level is the characteristic quantity which needs to be combined with other characteristic quantities to determine a signal modulation mode, is influenced by factors such as a signal-to-noise ratio, frequency offset and the like to a certain extent, and has higher implementation complexity.

The third level is the characteristic quantity which can only judge the signal type by comparing and excluding several signals based on the characteristic quantity value, and has larger influence by factors such as signal-to-noise ratio, frequency offset and the like and larger realization complexity.

By analogy, the selection of the hierarchical progression is related to the set of signals to be identified, the selected feature quantities, and the like.

And during classification, the first-level characteristic quantity is firstly judged, the second-level characteristic quantity is adopted to judge the unidentified signals, if the unidentified signals are not identified, the third-level characteristic quantity is used for identification, and the like is repeated until all the signals to be identified are identified.

The set of signal identifications is set as: AM, SSB, CW3 analog signals and 2ASK, MSK, 2FSK, 4FSK, BPSK, QPSK, 8PSK, pi/4 DQPSK, 16QAM9 digital signals.

For a clear description of the signal pattern grading process, the characteristics of the various modulation patterns are first analyzed below. The identification feature quantities to be employed are:

(1) and the signal bandwidth can be used for distinguishing a single carrier signal from a modulation signal, and if the bandwidth is less than a threshold, the signal bandwidth is directly judged to be the single carrier signal. The characteristic quantity is easy to obtain, is not influenced by frequency deviation and is slightly influenced by a signal-to-noise ratio.

(2) The number of the single-peak spectral lines of the salient pulses in each sub-frequency spectrum comprises the number of the single-peak spectral lines of the salient pulses in various frequency spectrums such as an amplitude square spectrum, a frequency spectrum, a square spectrum, a quartic spectrum, an octave spectrum and the like. The characteristic quantity is not influenced by frequency deviation and is less influenced by a signal-to-noise ratio. The spectrum is obtained by calculating each sub-spectrum of the signal and performing peak search, the complexity is low, and the number of the prominent peak spectral lines existing in each sub-spectrum of each modulation signal is shown in table 1.

TABLE 1 number of prominent peak lines present in each sub-spectrum of each modulated signal

The peak number of 2FSK, 4FSK and AM signals is related to signal parameters, so the number of the prominent peak spectral lines in each signal spectrum is different. However, in the case of 2FSK and 4FSK signals, when the frequency interval is an integer multiple of the symbol rate, the number of prominent peak spectral lines is a binary number thereof, that is, 2 or 4, and when the frequency interval is squared, and octagonally, the frequency interval is increased by 2 times, 4 times, and 8 times, when the increased frequency interval is an integer multiple of the symbol rate, the number of prominent peak spectral lines in each sub-spectrum is a binary number thereof, and when the frequency interval is not an integer multiple of the symbol rate, no prominent peak spectral line exists in each sub-spectrum. The number of the outburst peak spectral lines in each sub-spectrum of the AM signal is not obviously regular.

As can be seen from table 1, the 2ASK signal can be identified by using the amplitude square spectrum and the spectrum characteristic parameters, BPSK can be identified by using the spectrum and the square spectrum characteristic parameters, pi/4 DQPSK can be identified by using the square spectrum and the quartic spectrum characteristic parameters, and 8PSK can be identified by using the quartic spectrum and the octave spectrum characteristic parameters.

The number of the outstanding peak spectral lines of the 2FSK, 4FSK and AM signals in each sub-frequency spectrum is related to signal parameters, the number of the spectral lines of different signals is different, but the number of the lines is the same, when the number of the outstanding peak spectral lines of each sub-frequency spectrum of the signals is 2, the signals can be identified as 2FSK, and when the number of the outstanding peak spectral lines of each sub-frequency spectrum is 4, the signals are judged as 4FSK when the spectral line position difference is equal.

(3) The number of envelope peaks after smooth filtering of each secondary frequency spectrum can be obtained by performing smooth filtering on each secondary frequency spectrum, and then the number of envelope peaks can be obtained. The feature quantity needs to be subjected to smoothing filtering and peak value searching on each frequency spectrum of the signal, and the complexity is relatively high, but the influence of noise is small. The number of envelope peaks after spectrally smooth filtering for each of the MSK, 2FSK and 4FSK signals is shown in table 2.

TABLE 2 number of envelope peaks after respective spectral filtering of different modulated signals

If the frequency interval of the 2FSK signal and the 4FSK signal is less than 2, the frequency spectrum filtering only has 1 envelope peak value, otherwise, 2 or 4 envelope peak values exist. Since the frequency separation should be at least larger than the symbol rate, the number of envelope peaks after filtering for the squared spectrum and the fourth power spectrum is determined.

Identification of MSK, 2FSK and 4FSK signals may be accomplished by the number of spectral filtering envelope peaks and the number of preceding prominent peak spectral lines. For MSK signals, the number of peaks present in the spectrum combined with the number of peaks of the envelope can be identified. For the 2FSK signal and the 4FSK signal, considering that the number of envelope peaks may be affected by noise, the identification as the 4FSK signal may be performed uniformly in combination with the interval between the envelope peaks, otherwise, the signal is identified as the 2FSK signal.

(4) The signal amplitude characteristic description parameter can be obtained by describing the distribution condition of the signal amplitude and adopting the ratio of the variance and the mean value of the amplitude. The feature quantity is less complex to implement, but is affected to some extent by noise. For QPSK and 16QAM signals, the amplitudes of signal constellation points are different, QPSK has only one amplitude, 16QAM has multiple amplitudes, and QPSK can be identified by the ratio of the variance and the mean value of the signal amplitudes and is recorded as a signal amplitude parameter. The identification of the AM and SSB signals is identified in combination with the number of spectral peak peaks.

From the above analysis, it can be seen that the classification and identification of 12 signals can be completed by the above four parameters. Therefore, the feature quantities to be employed are classified as follows:

a first stage: bandwidth, single carrier and modulated signals can be distinguished.

And a second stage: the number of single-peak spectral lines of the pulse in the frequency spectrum comprises the number of single-peak spectral lines of the pulse in various frequency spectrums, such as an amplitude square spectrum, a frequency spectrum, a square spectrum, a quartic spectrum and the like. The 2ASK, BPSK, pi/4 DQPSK, 8PSK signals and the partial 2FSK, 4FSK signals may be identified, and other signals may be identified in combination with other parameters.

And a third stage: and the peak value number after the frequency spectrum smoothing filtering comprises the peak value number after the smoothing filtering of the frequency spectrum, the square spectrum, the fourth power spectrum and the like of the signal. And combining the number of single-peak spectral lines of the burst pulse in the frequency spectrum to complete the identification of the MSK signal and the identification of unidentified 2FSK and 4FSK signals.

Fourth stage: the identification of QPSK, 16QAM, SSB and AM signals can be completed by combining the signal amplitude distribution parameters with the number of single-peak spectral lines of the salient pulses in each frequency spectrum.

The signal identification process comprises the following steps:

the first step is as follows: the flow is as shown in fig. 2, and single carrier and modulation signal identification is performed. And when the bandwidth of the signal after the preprocessing is smaller than the threshold, judging the signal to be single carrier CW, otherwise, entering the next step and adopting the next-stage characteristic parameter for identification.

The second step is that: the flow is shown in figure 3, and 2ASK, BPSK, pi/4 QPSK and 8PSK signals are identified according to the number of single peak spectral lines of the salient pulses in each frequency spectrum. If the number of peaks of the amplitude squared spectrum peak-highlighting spectral line is 1 and the number of peaks of the spectrum peak-highlighting spectral line is 1, judging the signal to be a 2ASK signal; if the number of peaks of the amplitude square spectrum peak value spectral line is 1, the number of lines of the spectrum peak value spectral line is 0, and the number of lines of the square spectrum peak value spectral line is 1, determining the signal as a BPSK signal; if the peak number of the amplitude square spectrum peak value spectral line is 1, the number of the spectrum peak value spectral lines is 0, the number of the square spectrum peak value spectral lines is 0, and the number of the quartic spectrum peak value spectral lines is 2, judging the signal to be a pi/4 DQPSK signal; if the peak number of the amplitude square spectrum peak value spectral line is 1, the number of the spectrum peak value spectral lines is 0, the number of the square spectrum peak value spectral lines is 0, the number of the quartic spectrum peak value spectral lines is 0, and the number of the octave spectrum peak value spectral lines is 1, the signal is judged to be an 8PSK signal; if the number of the outburst peak spectral lines in the frequency spectrum and the square spectrum is 2, judging the frequency spectrum and the square spectrum to be 2 FSK; and if the number of the spectral lines of the outburst peak values in the frequency spectrum and the square spectrum is more than 2 and the position difference of the spectral peaks is equal, judging the frequency spectrum to be 4 FSK. If not, the next stage of characteristic parameters are adopted for identification.

The third step: the flow is shown in fig. 4, and MSK, 2FSK and 4FSK are identified by combining the number of spectral peak highlighting lines and the number of peaks of envelopes after each spectral filtering. If the peak number of the spectral line of the spectral peak protrusion is 0, the number of the spectral line of the spectral peak protrusion of the square spectrum is 2, and the peak value of the spectral filtering envelope is 1, judging the MSK; if the peak value of the spectrum and square spectrum filtering envelope is 2 or the peak value of the square spectrum filtering and fourth power spectrum envelope is 2, judging the frequency spectrum and square spectrum filtering envelope is 2 FSK; if the envelope peak value of the frequency spectrum and the square spectrum filtering is larger than 2, or the envelope peak value of the square spectrum and the quartic spectrum filtering is larger than 2, and the position difference between the envelope peak values is equal, judging the frequency spectrum is 4FSK, otherwise judging the frequency spectrum is 2 FSK. If not, the next stage of characteristic parameters are adopted for identification.

The fourth step: the flow chart is shown in fig. 5, and QPSK, 16QAM, SSB and AM signals are identified by combining the spectral peak line number and the signal amplitude distribution characteristics. If the peak value number of the amplitude square spectrum peak pulse spectral line is 1, the peak value spectral line number of the frequency spectrum and the square spectrum peak pulse is 0, the peak value spectral line number of the quartic spectrum peak pulse is 1, and the signal amplitude parameter is smaller than the threshold, the signal amplitude parameter is judged to be QPSK and exceeds the threshold, and the signal amplitude parameter is judged to be 16 QAM. And for the SSB and AM signals, if the number of the spectrum peak value spectral lines of the spectrum highlighting pulse is more than 0, judging the signals to be AM signals, otherwise, judging the signals to be SSB signals. Thus, all signals are identified.

The method of the invention firstly grades the characteristic quantity selected by the signal to be identified according to the signal-to-noise ratio performance, the realization complexity and other conditions, and then calculates the size of the characteristic quantity according to the grading result and identifies the signal characteristic. And if the signal is identified, outputting an identification result, otherwise, adopting the next-stage characteristic quantity to continue identification until the signal is identified.

The method of the invention identifies the signal for many times by selecting the characteristic quantity with good signal-to-noise ratio performance and low complexity, increases the identification effect of the signal under low signal-to-noise ratio under the condition of larger signal set, and reduces the realization complexity.

Fig. 6 shows the signal identification results at different signal-to-noise ratios by the method of the present invention.

Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims

1. A modulation pattern recognition method based on characteristic quantity grading is characterized by comprising the following steps:

2. The method according to claim 1, wherein the main factors of the classification level of the feature vector classification are the signal set to be identified and the selected feature vector, and the classification level comprises a first level feature vector, a second level feature vector and a third level feature vector.

3. The method according to claim 2, wherein the step of determining the classification of the feature amount comprises:

firstly, judging a first-level characteristic quantity;

4. The method according to claim 1, wherein the signal identification set comprises 3 analog signals: AM, SSB, CW, and 9 digital signals: 2ASK, MSK, 2FSK, 4FSK, BPSK, QPSK, 8PSK, pi/4 DQPSK, 16 QAM.

5. The method according to claim 4, wherein the feature quantity to be used selected according to the signal identification set comprises the following hierarchical feature quantities:

6. The method according to claim 5, wherein the signal identification comprises the following steps:

the fourth step: identifying QPSK, 16QAM, SSB and AM signals by combining the identification signal set after the identification of the third-level characteristic quantity with the spectral peak highlighting value spectral line number and the signal amplitude distribution characteristic, if the peak value number of the amplitude square spectral highlighting peak pulse spectral line is 1, the peak value spectral line number of the spectral and square spectral highlighting pulse is 0, the peak value spectral line number of the quartic spectral highlighting pulse is 1, the signal amplitude parameter is less than the threshold, judging the signal to be QPSK and judging the signal to be 16QAM if the signal amplitude parameter exceeds the threshold,

and for the SSB and AM signals, if the number of spectral peak values of the spectrum highlighting pulses is greater than 0, judging the signals to be AM signals, otherwise, judging the signals to be SSB signals, and finishing the identification of all the signals in the signal identification set.