CN112307927A - BP network-based identification research for MPSK signals in non-cooperative communication - Google Patents

BP network-based identification research for MPSK signals in non-cooperative communication Download PDF

Info

Publication number
CN112307927A
CN112307927A CN202011153288.8A CN202011153288A CN112307927A CN 112307927 A CN112307927 A CN 112307927A CN 202011153288 A CN202011153288 A CN 202011153288A CN 112307927 A CN112307927 A CN 112307927A
Authority
CN
China
Prior art keywords
signal
neural network
network model
model
algorithm
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.)
Pending
Application number
CN202011153288.8A
Other languages
Chinese (zh)
Inventor
张承畅
余洒
徐余
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN202011153288.8A priority Critical patent/CN112307927A/en
Publication of CN112307927A publication Critical patent/CN112307927A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physiology (AREA)
  • Signal Processing (AREA)
  • Genetics & Genomics (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a BP-GA neural network model for in-class identification of MPSK signals in a non-cooperative communication system. Firstly, selecting six characteristics based on time domain and frequency domain as input samples of a model according to the characteristics of an MPSK signal; designing a BP neural network containing two hidden layers as a classifier for modulation recognition, and optimizing parameters of the neural network by means of a genetic algorithm; and finally, in order to reduce the sensitivity of the model to the signal-to-noise ratio, the signal-to-noise ratio order of the training samples is disturbed and then the training samples are used as the input of the network model for network training. Compared with the existing modulation identification algorithm based on the BP neural network, the identification accuracy of the MPSK signal under the low signal-to-noise ratio is improved. In addition, the method has strong realizability and high identification accuracy, and can be well applied to related projects of non-cooperative communication systems.

Description

BP network-based identification research for MPSK signals in non-cooperative communication
Technical Field
The invention relates to a neural network correlation algorithm and a signal processing correlation theory, and belongs to the field of communication signal processing and artificial intelligence.
Background
With the starting of research on commercial and next-generation mobile communication technologies in 5G, and the wide use of new communication technologies such as satellite communication, spread spectrum communication, frequency hopping communication, etc., wireless communication technology has become one of the most rapidly and widely developed communication technologies, and with the accompanying increasingly complex electromagnetic environment, the modulation mode of signals also presents a diversified development trend, and the technology of automatic modulation recognition of signals is a very critical step in non-cooperative communication systems such as radio spectrum resource supervision and modern electronic wars.
Early signal modulation identification was achieved by manual interpretation of measured parameters and therefore was heavily dependent on the skill level and working experience of the operator. In 4 months of 1969, Weaver C S et al published a first paper on the technical report of stanford university, which studies the automatic identification of modulation schemes — that is, "automatic classification of modulation types is achieved by using pattern recognition technology", thereby opening the door to automatic modulation identification of signals.
Classical modulation recognition techniques can be divided into two broad categories: 1. maximum Likelihood (ML) method based on hypothesis testing, 2. pattern recognition method based on feature extraction. The former can be seen as a multiple hypothesis testing problem. The method is to make theoretical derivation on the test statistic (usually adopting likelihood ratio function) of the intercepted signal under the condition of background interference, find a proper threshold, and then make judgment under the Bayes cost minimum criterion. The latter method uses feature extraction to realize modulation mode identification, which needs to select proper classifier for classification identification. The task of the classifier is: and classifying a given input mode represented by the feature vector into a proper mode class according to a certain criterion, completing the mapping from the feature space to the decision space, and finally giving a recognition result. Commonly used classifiers are: decision tree classifiers, nearest neighbor (KNN) classifiers, bayesian classifiers, Support Vector Machines (SVMs), random forests, and the like.
Nowadays, with the continuous development of artificial intelligence technologies such as deep learning and machine learning, it is a great trend of research to apply neural networks as classifiers in the field of modulation recognition. The neural network applied to the modulation recognition field has two main types: 1. convolutional Neural Network (CNN), 2. Back Propagation (BP) neural network. The CNN is mainly composed of a convolution layer, a pooling layer and a full-link layer. The convolution layer mainly plays a role in extracting the features of the image, the pooling layer mainly plays a role in downsampling, and the fully-connected layer mainly plays a role in classification. Convolutional neural networks are mainly used in the field of image recognition, and therefore are usually used together with relevant pattern features (such as constellation) of signals when CNN is applied to modulation scheme recognition. In 2019, Umbelliferal and the like design a CNN-LSTM parallel network, the same-direction component and the orthogonal component are directly used as input data, characteristic parameters do not need to be designed artificially, and the algorithm has better identification performance under low signal-to-noise ratio due to the influence of artificial factors is reduced. In 2019, Siyang Zhou et al propose a robust automatic modulation recognition method based on a convolutional neural network aiming at the defect that the current modulation recognition model lacks generalization, can recognize 15 signals, and has good recognition accuracy under low signal-to-noise ratio. In 2020, Chenchanmei et al propose an improved convolutional neural network structure which can classify seven different modulation signals, and when the signal-to-noise ratio is not less than 5dB, the recognition rate can reach 97.99%, and when the signal-to-noise ratio is not less than 9dB, the recognition rate can reach 100%. And the BP neural network comprises an input layer, a hidden layer and an output layer. The method mainly solves the weight learning problem of a hidden layer of a multilayer neural network by back propagation of errors. Application of BP neural networks to the field of modulation recognition is often used in conjunction with temporal characteristics of the signal. In 2016, Wangyi et al trained the neural network by a variable gradient BP correction algorithm to improve convergence rate and shorten training time, and when the signal-to-noise ratio is 10dB, the recognition rate reaches 95%. In 2019, Wu Xiqian et al propose a signal modulation identification algorithm based on a BP neural network, and when the signal-to-noise ratio is 0dB, the identification rate can reach more than 85%. In 2019, Yuanmeng et al used a BP neural network algorithm to automatically identify six common digital modulation signals, and when the signal-to-noise ratio was greater than 10dB, the accuracy was more than 98%.
Based on the application of the current neural network in modulation recognition, it can be seen that: the current applications of ANN to modulation recognition can be divided into two broad categories: one type is a modulation recognition algorithm based on a convolutional neural network, the method converts the modulation recognition problem of the signal into an image recognition problem, no prior information of any signal is needed, and the important difficulty is the extraction of the graphic characteristics of the signal; the other type is a modulation recognition algorithm based on a deep neural network (such as a BP neural network), the method is to recognize ANN as a classifier, and the important difficulty is parameter optimization during network training.
Disclosure of Invention
Based on the problem that the identification precision of the common neural network model is not high enough under the condition of low signal to noise ratio during signal modulation identification, the invention provides a method for optimizing network parameters by using a BP neural network as a basic network model and using a Genetic Algorithm (GA) to finally improve the accuracy of phase shift keying (MPSK) signal identification under the condition of low signal to noise ratio. The system flow design diagram of the invention is shown in figure 1.
The Genetic Algorithm (GA) based BP neural network of the present invention is characterized as follows:
in order to overcome the defect that a BP neural network model is easy to fall into a local minimum value and the defect of the existing neural network algorithm aiming at signal modulation identification, the invention provides a novel BP-GA network model aiming at identification in MPSK signals by optimizing the BP neural network by using a GA algorithm. The model is used for training after preprocessing the characteristics of signals by using a BP neural network comprising two hidden layers, and solves the problems that the BP neural network is low in convergence speed and easy to fall into local optimum by using a GA algorithm. The invention greatly reduces the sensitivity of the network model to the signal-to-noise ratio of the signal and improves the identification rate of the PSK signal under low signal-to-noise ratio. The method is already applied to scientific research projects.
The invention discloses an algorithm for identifying MPSK signals in classes based on a BP neural network, which comprises the following steps:
1) extracting three instantaneous characteristics and three characteristics based on high-order cumulant of the MPSK signal as the input of a network model;
2) constructing a BP neural network model containing two hidden layers as a classifier for identifying MPSK signals;
3) optimizing BP network parameters by using a GA algorithm, and obtaining optimal BP neural network parameters by taking the output error of a network model as a fitness function;
4) transmitting the optimal BP neural network parameters to a network model, and training the BP network model by using six signal characteristics as the input of the network model;
5) for the output result of the network model, inverse normalization processing is needed, and then the final recognition result can be obtained through judgment;
6) the modules in the steps are integrated into a program, and when a group of MPSK signal values are input, the identification result can be directly output.
The above step 1) uses three transient characteristics and three characteristics based on high-order cumulants of the MPSK signal, and specifically includes: the instantaneous characteristics of the signal mainly include instantaneous frequency, instantaneous amplitude and instantaneous phase, and based on the three instantaneous characteristics of the signal, five common instantaneous characteristics of the signal are: a maximum value of zero-center normalized instantaneous amplitude spectral density, a standard deviation of zero-center normalized instantaneous amplitude absolute value, a standard deviation of zero-center non-weak signal segment instantaneous phase nonlinear component absolute value, and a standard deviation of zero-center non-weak signal segment normalized instantaneous frequency absolute value. In view of the characteristics of the MPSK signal, the standard deviation of the instantaneous phase nonlinear component of the zero-center non-weak signal segment, the standard deviation of the absolute value of the instantaneous phase nonlinear component of the zero-center non-weak signal segment and the standard deviation of the absolute value of the normalized instantaneous frequency of the zero-center non-weak signal segment are selected as part of characteristic samples.
The high-order cumulant of the signal belongs to the frequency domain characteristics of the signal, can characterize the high-order statistical characteristics of the signal, has relatively high computational complexity, but because some high-order cumulants of Gaussian white noise are constantly equal to zero, and the high-order statistics of the modulation signal has good anti-fading characteristics. The method selects three features based on high-order cumulant as part of feature samples to train the network.
Designing a corresponding BP neural network model for the features extracted from the MPSK signal in the step 2). Because the selected group of characteristic values has 6 characteristics, the input layer of the BP neural network comprises 6 nodes; since the model owner performs in-class identification on MPSK (M is 2, 4, 8, 16) signals, the output layer of the BP neural network includes 4 nodes; and the number of nodes of the middle two hidden layers is found through experimental simulation analysis: when the number of nodes of the first hidden layer is set to be 40 and the number of nodes of the second hidden layer is set to be 10, the recognition effect is best, and a specific BP neural network model design invention is shown in figure 2.
In the step 3), in order to accelerate the convergence rate of the BP neural network and avoid the trapping of local optima, a genetic algorithm is used for optimizing the parameters of the network, and a specific algorithm optimization strategy is as follows:
the genetic algorithm selected by the invention is a parallel random search optimization method for simulating a natural genetic mechanism and a biological evolution theory. The algorithm selects the output error of the neural network as a fitness function, firstly, network parameters are used as population individuals to be coded, and then, the individuals are screened through selection, crossing and variation operations in inheritance, so that the individuals with good fitness values are reserved, the individuals with poor fitness values are eliminated, and a new population inherits the information of the previous generation and is superior to the information of the previous generation. And the process is repeatedly circulated until the condition is met.
The selection operation is to select an individual from an old population to a new population, the probability of the individual being selected is related to the fitness value, and the smaller the fitness value, the higher the probability of the individual being selected. The crossover operation refers to the process of randomly selecting one point or multiple points of chromosomes of two individuals in a population to exchange to generate new individuals, and a specific cross operation schematic diagram is shown in figure 3. The mutation operation refers to selecting a chromosome of an individual from a population, and mutating one point of the chromosome to generate a better individual, and the specific mutation operation is shown in fig. 4. The cross probability and the mutation probability set in the invention are 0.3 and 0.1 respectively.
And 4) introducing the BP neural network parameters obtained by genetic algorithm optimization in the step 3) into a BP neural network model, and training the BP neural network model by using the six characteristic values extracted in the step 1) under the condition that the BP neural network parameters are optimal.
And step 5) further processing the result obtained by the network model, namely performing inverse normalization and threshold judgment, so as to obtain a final recognition result.
And 6) obtaining a trained BP neural network model by means of the four steps, and testing the MPSK signal identification effect of the BP neural network classifier by inputting a test sample into the network model.
The method has the main effects of carrying out in-class identification on the MPSK signals in the non-cooperative communication system under the low signal-to-noise ratio, reducing the sensitivity of a network model to the signal-to-noise ratio and improving the accuracy of signal identification under the low signal-to-noise ratio. The method comprises the following specific steps:
the performance measurement indexes of the model are evaluated to be P (accuracy), R (recall rate) and F (comprehensive performance index F value), and the accuracy and the recall rate are widely applied in the fields of information retrieval and statistical classification and play an important role in measuring the quality of results. P (accuracy) represents the proportion of the accurate number of the predictions in the system and measures the precision ratio in the system. R (recall) represents the ratio of true prediction correct to all positive samples in the system, and measures the recall ratio in the system. Let TP denote the number of positive samples predicted from positive samples and FP denote the number of false positives for positive samples predicted from negative samples. FN represents the number of false positives that predict positive samples as negative samples. The relationship between the three and P and R is shown in formulas (1) and (2). However, in general, there is a certain contradiction between the accuracy rate and the recall rate, and the phenomenon of high accuracy rate but low recall rate occurs, or vice versa. Therefore, in this time, the situation of the accuracy and the recall index needs to be considered comprehensively, namely, an F-measure evaluation method is adopted. The relationship between F-measure and P (accuracy rate) and R (recall rate) is shown in formula 3.
Figure BDA0002741885250000041
Figure BDA0002741885250000042
Figure BDA0002741885250000043
Drawings
FIG. 1 is a flow chart of the system of the present invention
FIG. 2 is a schematic diagram of a BP neural network model designed by the present invention
FIG. 3 is a schematic diagram of the crossover operation of the genetic algorithm in the present invention
FIG. 4 is a schematic diagram of the operation of genetic algorithm mutation in the present invention
Detailed Description
The invention is called BP-GA model by combining genetic algorithm and BP neural network. The identification model of the invention is mainly divided into: a feature extraction module and a classifier module formed by a BP-GA model. Firstly, analysis is carried out according to the modules, and the specific steps are as follows:
the method comprises the following steps: performing feature extraction of the signal by using a mathematical calculation function;
a pretreatment part: the specific calculation formula for calculating the instantaneous amplitude, the instantaneous phase, the instantaneous frequency and the high-order cumulant of the received signal is as follows:
assuming that the signal is x (t), a complex analytic formula of the signal can be obtained as
s(t)＝x(t)+jy(t) (1)
Where y (t) is the Hilbert transform of signal x (t), i.e.,
Figure BDA0002741885250000051
the instantaneous amplitude and the instantaneous phase can be obtained by a complex analytic expression of the signal, and are respectively:
Figure BDA0002741885250000052
Figure BDA0002741885250000053
the instantaneous frequency of a signal is the differential of the instantaneous phase, so for a discrete signal, its instantaneous frequency can be calculated as the difference of the instantaneous phases, i.e.:
Figure BDA0002741885250000054
for a stationary complex random process x (k) with a mean value of zero, the mixing distance is defined as:
Mpq＝E{[x(k)]p-q[x*(t)]q} (6)
where denotes the complex conjugate operation and E {. is the mathematical expectation. The various higher order cumulative amounts of the signal may be expressed as:
Figure BDA0002741885250000061
a feature calculation section: the six characteristic values of the signal used in the invention are obtained based on the calculation of the instantaneous characteristic and the high-order cumulative amount of the signal as follows:
(1) feature 1
Figure BDA0002741885250000062
Wherein, C represents non-weak signal (signal amplitude is larger than decision threshold a)t) The number of the sequences is such that,
Figure BDA0002741885250000063
zero center instantaneous nonlinear phase.
(2) Feature 2
Figure BDA0002741885250000064
(3) Feature 3
Figure BDA0002741885250000065
Wherein,
Figure BDA0002741885250000066
normalizing the instantaneous amplitude, R, for zero centerbIn order to be the rate of the symbols,
Figure BDA0002741885250000067
is zero center frequency.
(4) Feature 4
Figure BDA0002741885250000068
(5) Feature 5
Figure BDA0002741885250000069
(6) Feature 6
Figure BDA00027418852500000610
A data set processing section: the feature value group number of the six feature values of each signal under different system numbers and different signal-to-noise ratios is 2600, and the input dimension of the data set is 10400 × 6. Randomly scrambling input samples to reduce the sensitivity of a network model to signal noise; the method comprises the steps of reasonably dividing a data set, carrying out unified normalization and standardization on data, unifying input dimensionality, and simultaneously specifying the ratio of a training sample to a test sample to be 9: 1.
Step two: optimizing the BP network model by a genetic algorithm;
(1) setting related parameters of a BP neural network model and a genetic algorithm;
the BP neural network is provided with two hidden layers, and the number of nodes of the hidden layers is respectively set to be 40 and 10; the number of network training times is set to 1000, the learning rate is set to 0.01, and the training target is set to 0.0001; since the input samples and the desired output are determined, the number of nodes of the input layer and the output layer of the neural network is determined, 6 and 4 respectively.
The relevant parameters of the genetic algorithm in the invention are set as follows: the population scale is set to be 30, and the evolution algebra is set to be 50; the upper and lower boundary values of the population are set to-3 and 3 respectively; the probability of intersection and mutation is respectively set as: 0.3, 0.1; setting a fitness function of a genetic algorithm as an error value between actual output and expected output of a BP neural network, wherein the expression is as follows:
Figure BDA0002741885250000071
wherein, youtTo desired output, yrealIs the actual output. In the present invention, the smaller the value of the fitness function, the better.
(2) Optimizing BP neural network parameters
After a BP neural network model is built, initial parameters (a threshold value and a weight value of each layer) of the network are input into a genetic algorithm, the parameters are encoded, iteration is performed by means of selection, intersection and variation operation of the genetic algorithm to continuously change the values of the parameters, and the fitness value of each individual is calculated after each iteration to determine eliminated and left individuals. And finally, obtaining the optimal BP neural network parameter after the maximum iteration number is reached or the fitness value meets the requirement.
Step three: training a network model;
the method comprises the steps that firstly, two modules of feature extraction and network parameter optimization are respectively completed, and then extracted features can be used as input samples to train the BP neural network using the optimal network model parameters.
Because the extracted samples are arranged according to the sequence of the signal-to-noise ratios from low to high under different signal-to-noise ratios, the obtained samples need to be preprocessed before training, namely, the sequence of the samples is randomly disturbed to reduce the sensitivity of the network model to the signal noise. And then, selecting 90% of characteristic values as training samples to train the model. Network training is stopped when the maximum number of iterations is reached or the error reaches a desired target. The obtained model is a trained BP neural network model, and the model can realize the in-class identification of the MPSK signal.
Step four: and processing the output result of the network model.
Since the normalization processing is performed on the input data set when the data set is processed, the inverse normalization processing needs to be performed on the output result after the output of the BP neural network model is obtained, and then the obtained model output result is judged, and the obtained result is the final recognition result. The judgment threshold value set in the invention is 0.5, when the result is greater than 0.5, the identification result is judged to be 1, otherwise, the identification result is judged to be 0.
Step five: and (6) testing the model.
And inputting the test sample into the trained BP neural network model to obtain a recognition result. And then, calculating specific values according to the three indexes of the performance measurement indexes of the model, namely the accuracy, the recall rate and the comprehensive performance index so as to evaluate the performance of the model.
The invention mainly realizes the in-class identification of MPSK signals in a non-cooperative communication system by means of a BP neural network and a genetic algorithm. The invention provides a BP-GA network model architecture, which improves the convergence speed of a network model by means of a genetic algorithm and simultaneously avoids the network from falling into local optimization. The invention can improve the identification accuracy of the MPSK signal under low signal-to-noise ratio and simultaneously reduce the sensitivity of the network model to the signal noise.

Claims (5)

1. The patent proposes a BP neural network model for intra-class identification of MPSK signals in a non-cooperative communication system. The model has more excellent performance in the aspect of accuracy under low signal-to-noise ratio compared with the prior model. Are currently used in real projects.
The patent algorithm model structure mainly comprises the following steps:
1) extracting three instantaneous characteristics and three characteristics based on high-order cumulant of the MPSK signal as the input of a network model;
2) constructing a BP neural network model containing two hidden layers as a classifier for identifying MPSK signals;
3) optimizing BP network parameters by using a GA algorithm, and obtaining optimal BP neural network parameters by taking the output error of a network model as a fitness function;
4) transmitting the optimal BP neural network parameters to a network model, and training the BP network model by using six signal characteristics as the input of the network model;
5) for the output result of the network model, inverse normalization processing is needed, and then the final recognition result can be obtained through judgment;
6) the modules in the steps are integrated into a program, and when a group of MPSK signal values are input, the identification result can be directly output.
2. The feature extraction module of claim 1, wherein: based on the step 1), three characteristics based on instantaneous characteristics and three characteristics based on high-order cumulant are selected as characteristic samples by analyzing the characteristics of the MPSK signals. The extracted features are only simple superposition of single type features or features obtained through complex operation, the features are not representative enough, and the time complexity is too high. The characteristics used by the method are combined with the time domain characteristics and the frequency domain characteristics of the target signal, and the characteristics are more representative and can reflect the characteristics of PSK signals with different systems. The specific characteristic extraction comprises the following steps:
transient characteristics: the commonly used transient characteristics of a signal mainly include three of instantaneous frequency, instantaneous amplitude and instantaneous phase. Because the phase difference is mainly between the PSK signals with different binary numbers, the method selects the characteristics of the PSK signals based on the instantaneous phase and the instantaneous frequency to calculate the characteristics. The three temporal characteristics are as follows:
(1) feature 1
Figure FDA0002741885240000011
Wherein, C represents non-weak signal (signal amplitude is larger than decision threshold a)t) The number of the sequences is such that,
Figure FDA0002741885240000012
zero center instantaneous nonlinear phase.
(2) Feature 2
Figure FDA0002741885240000013
(3) Feature 3
Figure FDA0002741885240000021
Wherein,
Figure FDA0002741885240000022
normalizing the instantaneous amplitude, R, for zero centerbIn order to be the rate of the symbols,
Figure FDA0002741885240000023
is zero center frequency.
High order cumulant: the high-order cumulant of the signal is the frequency domain characteristic of the signal, and in view of the insensitivity of the high-order cumulant to certain white Gaussian noise, the three frequency domain characteristics based on the high-order cumulant are obtained through the operation of the high-order cumulant. The specific characteristic calculation formula is as follows:
(4) feature 4
Figure FDA0002741885240000024
(5) Feature 5
Figure FDA0002741885240000025
(6) Feature 6
Figure FDA0002741885240000026
The above steps can obtain the characteristics of the target signal, and a certain processing is required to obtain the final data set. Firstly, in order to reduce the sensitivity of a network model to noise, the signal-to-noise ratio disorder processing needs to be carried out on the obtained features; then, normalization is required for the samples to obtain the final input data set. The characteristic value group number of the six characteristic values of each signal under different system numbers and different signal-to-noise ratios is 2600, and the input dimensionality of the finally obtained data set of the patent is 10400 multiplied by 6.
3. The BP neural network model of claim 1, wherein: the BP neural network model based on the step 2) is composed of an input layer, two hidden layers and an output layer. The number of nodes of an input layer of a BP neural network model in the patent can be determined to be 6 according to the characteristic data extracted by the signal; the number of output layer nodes of the BP neural network model in the patent can be determined to be 4 according to the number of types of target signals; the node numbers of the remaining two hidden layers are compared and measured with model performance indexes of different node numbers through an earlier simulation experiment, and finally the node numbers of the hidden layers of the BP neural network in the patent are determined to be 40 and 10.
The activation function of the BP neural network hidden layer uses a hyperbolic tangent function, and the expression of the hyperbolic tangent function is as follows:
Figure FDA0002741885240000027
the activation function of the output layer uses a linear function purelin, and the expression of the linear function purelin is as follows:
y＝purelin(x)＝x (8)
the training function uses a trainlm function, and for a medium-scale BP neural network convergence rate block, the algorithm avoids directly calculating a Hessian matrix, so that the training calculation amount is small. The function uses a Levenberg-Marquardt algorithm, and the algorithm can achieve the advantages of combining the Gauss-Newton algorithm and the gradient descent method by modifying parameters when being executed, namely reducing the parameter value to be closer to the Gauss-Newton method when the descent is too fast; increasing the parameter value when the drop is too slow brings it closer to the gradient descent method.
4. The invention adds genetic algorithm in BP neural network to optimize network parameter, which is characterized in that: based on the genetic algorithm used in the step 3), the genetic algorithm mainly comprises three parts, namely individual coding, selection, crossing and mutation, and fitness function calculation.
Firstly, inputting the initialization parameters of the BP neural network model into an algorithm to encode the parameters. The coding method selected in the patent is real number coding; then, the fitness value of each individual is calculated, and the fitness function selected in the patent is the output error of the BP neural network, and the expression of the fitness function is as follows:
Figure FDA0002741885240000031
wherein, youtTo desired output, yrealIs the actual output. In the present invention, the smaller the value of the fitness function, the better. The best individual fitness value is found as the fitness value of the population after each iteration. Selecting operation by adopting a roulette method during genetic operation, wherein the fitness value is an important selection metric standard; in the crossing operation and the mutation operation, individuals are selected to cross or mutate according to the set probability. Meanwhile, boundary values of individuals in the population are set in the patent, and whether the operation is effective or not can be checked when each genetic operation is carried out, so that the effectiveness of the individuals in the population is ensured.
5. The BP neural network model for class-in identification of MPSK signals in a non-cooperative communication system as claimed in claims 1, 2, 3 and 4, wherein the steps 4) and 5) are core steps of network model training, specifically, the input data set obtained in the step 1) is put into the BP-GA network model obtained in the steps 2) and 3) to train the network model. And further processing the output result of the network model through steps of inverse normalization, threshold judgment and the like to finally obtain an identification result. And finally, putting the test sample into the trained sample according to the step 6) to obtain a recognition result, calculating related parameters according to the performance indexes of the model, and measuring the performance of the model. The invention provides a BP network model architecture, and the convergence speed of the network model is improved by means of a genetic algorithm, and meanwhile, the network is prevented from falling into local optimization. The invention can improve the identification accuracy of the MPSK signal under low signal-to-noise ratio and simultaneously reduce the sensitivity of the network model to the signal noise.
CN202011153288.8A 2020-10-26 2020-10-26 BP network-based identification research for MPSK signals in non-cooperative communication Pending CN112307927A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011153288.8A CN112307927A (en) 2020-10-26 2020-10-26 BP network-based identification research for MPSK signals in non-cooperative communication

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011153288.8A CN112307927A (en) 2020-10-26 2020-10-26 BP network-based identification research for MPSK signals in non-cooperative communication

Publications (1)

Publication Number Publication Date
CN112307927A true CN112307927A (en) 2021-02-02

Family

ID=74331215

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011153288.8A Pending CN112307927A (en) 2020-10-26 2020-10-26 BP network-based identification research for MPSK signals in non-cooperative communication

Country Status (1)

Country Link
CN (1) CN112307927A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112887239A (en) * 2021-02-15 2021-06-01 青岛科技大学 Method for rapidly and accurately identifying underwater sound signal modulation mode based on deep hybrid neural network
CN113702317A (en) * 2021-08-30 2021-11-26 中国农业科学院农业信息研究所 Drainage basin non-point source pollution component sensor, monitoring system and method
CN113792852A (en) * 2021-09-09 2021-12-14 湖南艾科诺维科技有限公司 Signal modulation mode identification system and method based on parallel neural network
CN114826850A (en) * 2022-04-21 2022-07-29 中国人民解放军国防科技大学 Modulation identification method, device and equipment based on time-frequency diagram and deep learning
CN118413289A (en) * 2024-07-01 2024-07-30 成都安则科技有限公司 Radio signal interference method of traversing machine

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030204380A1 (en) * 2002-04-22 2003-10-30 Dishman John F. Blind source separation utilizing a spatial fourth order cumulant matrix pencil
CN108540202A (en) * 2018-03-15 2018-09-14 西安电子科技大学 A kind of satellite communication signals Modulation Mode Recognition method, satellite communication system
CN108768907A (en) * 2018-01-05 2018-11-06 南京邮电大学 A kind of Modulation Identification method based on temporal characteristics statistic and BP neural network
CN110120926A (en) * 2019-05-10 2019-08-13 哈尔滨工程大学 Modulation mode of communication signal recognition methods based on evolution BP neural network
CN111049770A (en) * 2019-12-06 2020-04-21 西安电子科技大学 Modulation signal identification method based on high-order cumulant
CN111259750A (en) * 2020-01-10 2020-06-09 西北工业大学 Underwater sound target identification method for optimizing BP neural network based on genetic algorithm

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030204380A1 (en) * 2002-04-22 2003-10-30 Dishman John F. Blind source separation utilizing a spatial fourth order cumulant matrix pencil
CN108768907A (en) * 2018-01-05 2018-11-06 南京邮电大学 A kind of Modulation Identification method based on temporal characteristics statistic and BP neural network
CN108540202A (en) * 2018-03-15 2018-09-14 西安电子科技大学 A kind of satellite communication signals Modulation Mode Recognition method, satellite communication system
CN110120926A (en) * 2019-05-10 2019-08-13 哈尔滨工程大学 Modulation mode of communication signal recognition methods based on evolution BP neural network
CN111049770A (en) * 2019-12-06 2020-04-21 西安电子科技大学 Modulation signal identification method based on high-order cumulant
CN111259750A (en) * 2020-01-10 2020-06-09 西北工业大学 Underwater sound target identification method for optimizing BP neural network based on genetic algorithm

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
M.L.D.WONG 等: "Automatic digital modulation Recognition using artificial neural network and genetic algorithm", 《SIGNAL PROCESSING》 *
WENZHE SHI 等: "Particle Swarm Optimization-Based Deep Neural Network for Digital Modulation Recognition", 《IEEE ACCESS》 *
吴喜权 等: "利用遗传BP神经网络的调制识别新算法", 《通信技术》 *
徐志超: "基于改进神经网络的数字调制信号的综合识别", 《中国优秀博硕士学位论文全文数据库（硕士）信息科技辑》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112887239A (en) * 2021-02-15 2021-06-01 青岛科技大学 Method for rapidly and accurately identifying underwater sound signal modulation mode based on deep hybrid neural network
CN112887239B (en) * 2021-02-15 2022-04-26 青岛科技大学 Method for rapidly and accurately identifying underwater sound signal modulation mode based on deep hybrid neural network
CN113702317A (en) * 2021-08-30 2021-11-26 中国农业科学院农业信息研究所 Drainage basin non-point source pollution component sensor, monitoring system and method
CN113702317B (en) * 2021-08-30 2023-10-27 中国农业科学院农业信息研究所 River basin non-point source pollution component sensor, monitoring system and method
CN113792852A (en) * 2021-09-09 2021-12-14 湖南艾科诺维科技有限公司 Signal modulation mode identification system and method based on parallel neural network
CN113792852B (en) * 2021-09-09 2024-03-19 湖南艾科诺维科技有限公司 Signal modulation mode identification system and method based on parallel neural network
CN114826850A (en) * 2022-04-21 2022-07-29 中国人民解放军国防科技大学 Modulation identification method, device and equipment based on time-frequency diagram and deep learning
CN118413289A (en) * 2024-07-01 2024-07-30 成都安则科技有限公司 Radio signal interference method of traversing machine

Similar Documents

Publication Publication Date Title
CN112307927A (en) BP network-based identification research for MPSK signals in non-cooperative communication
CN108830308B (en) Signal-based traditional feature and depth feature fusion modulation identification method
CN109450834B (en) Communication signal classification and identification method based on multi-feature association and Bayesian network
CN108696331B (en) Signal reconstruction method based on generation countermeasure network
Wang et al. Fault recognition using an ensemble classifier based on Dempster–Shafer Theory
Aydogan et al. hGA: Hybrid genetic algorithm in fuzzy rule-based classification systems for high-dimensional problems
CN106656357B (en) Power frequency communication channel state evaluation system and method
CN115018193A (en) Time series wind energy data prediction method based on LSTM-GA model
CN114143210A (en) Deep learning-based command control network key node identification method
CN101901251A (en) Method for analyzing and recognizing complex network cluster structure based on markov process metastability
CN114298166B (en) Spectrum availability prediction method and system based on wireless communication network
Pan et al. Specific radar emitter identification using 1D-CBAM-ResNet
CN110516792A (en) Non-stable time series forecasting method based on wavelet decomposition and shallow-layer neural network
CN108631817B (en) Method for predicting frequency hopping signal frequency band based on time-frequency analysis and radial neural network
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
CN117873837A (en) Analysis method for capacity depletion trend of storage device
Wu et al. Least-squares support vector machine-based learning and decision making in cognitive radios
CN117829054A (en) Circuit layout optimization method and system based on programmable logic controller
CN115033893B (en) Information vulnerability data analysis method of improved clustering algorithm
CN115174421B (en) Network fault prediction method and device based on self-supervision unwrapping hypergraph attention
CN116822742A (en) Power load prediction method based on dynamic decomposition-reconstruction integrated processing
CN114118151B (en) Intelligent spectrum sensing method with environment adaptation capability
CN112529035A (en) Intelligent identification method for identifying individual types of different radio stations
CN118445627B (en) Slope stability analysis method based on artificial intelligence
Bai et al. Cooperative spectrum sensing method based on channel attention and parallel CNN-LSTM

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210202

WD01 Invention patent application deemed withdrawn after publication