CN116405355A - Signal demodulation method based on mode selection - Google Patents
Signal demodulation method based on mode selection Download PDFInfo
- Publication number
- CN116405355A CN116405355A CN202310443259.2A CN202310443259A CN116405355A CN 116405355 A CN116405355 A CN 116405355A CN 202310443259 A CN202310443259 A CN 202310443259A CN 116405355 A CN116405355 A CN 116405355A
- Authority
- CN
- China
- Prior art keywords
- signal
- modulation
- individual
- mode
- updated
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 239000013598 vector Substances 0.000 claims description 31
- 239000011159 matrix material Substances 0.000 claims description 25
- 238000005457 optimization Methods 0.000 claims description 22
- 230000000739 chaotic effect Effects 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 19
- 230000004913 activation Effects 0.000 claims description 15
- 238000011176 pooling Methods 0.000 claims description 12
- 238000013507 mapping Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 3
- 238000004891 communication Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 230000010363 phase shift Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010295 mobile communication Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 101100083446 Danio rerio plekhh1 gene Proteins 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/0008—Modulated-carrier systems arrangements for allowing a transmitter or receiver to use more than one type of modulation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a signal demodulation method based on mode selection, which relates to the technical field of signal processing.
Description
Technical Field
The invention relates to the technical field of signal processing, in particular to a signal demodulation method based on mode selection.
Background
Wireless communication plays an extremely important role in the field of modern communications, and is widely used in various fields such as: mobile communications, microwave communications, wireless relay, cellular network communications, satellite communications, and translational layer communications, among others. As new communication protocols and standards continue to be proposed, the rational allocation of spectrum resources presents a significant challenge for the convergence of various communication regimes.
In recent years, wireless communication technology is widely used in the commercial field, the communication environment of the wireless communication technology becomes increasingly complex, and the requirements on communication equipment are becoming more and more stringent, particularly, the wide use of 4G and 5G mobile communication technology, and how to reasonably and efficiently allocate wireless spectrum is important when communication signals transmit information by using various modulation types on a high bandwidth. In order for signals of different modulation types to be effectively identified and monitored, automatic modulation mode identification (Automatic Modulation Recognition, AMR) is a difficult and hot spot technique for research in the field of wireless communications.
The conventional AMR algorithm mainly comprises Likelihood Based estimation (LB). In the likelihood estimation method, the modulation recognition problem is expressed as a multiple hypothesis test problem. The signals are processed through a Bayesian estimation method or a maximum likelihood estimation method, likelihood ratios of all signal hypotheses are compared with threshold values, and classification of signal modulation modes is achieved. Although the method based on likelihood estimation can classify modulation types well, the method has the problems of high algorithm complexity, more priori knowledge, difficult realization and the like.
Disclosure of Invention
The invention aims to provide a signal demodulation method based on mode selection, which solves the problems existing in the prior art.
The invention is realized by the following technical scheme:
a signal demodulation method based on mode selection, comprising:
acquiring a modulation signal transmitted by external equipment, and extracting characteristics of the modulation signal to obtain mode characteristics corresponding to the modulation signal;
constructing a modulation signal mode selection model, and training the modulation signal mode selection model by adopting a parameter optimization algorithm to obtain a trained modulation signal mode selection model;
And identifying the mode characteristics by adopting the trained modulation signal mode selection model to obtain a target modulation mode corresponding to the modulation signal, and demodulating the modulation signal by modulating a demodulation model corresponding to the target modulation mode to obtain a signal demodulation result.
In one possible implementation manner, obtaining a modulated signal transmitted by an external device, and extracting features of the modulated signal to obtain mode features corresponding to the modulated signal, where the method includes:
acquiring an AM signal, an ASK signal, an FSK signal, a BPSK signal, a QPSK signal or an MSK signal transmitted by external equipment to obtain a modulation signal;
discretizing the modulation signal, dividing the discretized modulation signal into M frames according to a preset frame division length L, and obtaining M frame sub-signals, wherein the frame division length L is used for representing that each frame of sub-signals comprises L signal points;
performing Fourier transform on each frame of sub-signal to obtain a first target sub-signal, and determining the frequency of the highest point in the first target sub-signal to obtain a first characteristic;
combining all the first features according to the sequence of the sub-signals to obtain a feature vector of the highest point of the short-time Fourier;
short-time energy of each frame of sub-signals is obtained, second characteristics are obtained, all the second characteristics are combined into characteristic vectors according to the sequence of the sub-signals, and short-time energy characteristic vectors are obtained;
Obtaining the similarity between each frame of sub-signal and the next frame of sub-signal to obtain third characteristics, and combining all the third characteristics into characteristic vectors according to the sequence of the sub-signals to obtain frame similarity characteristic vectors;
acquiring the instantaneous phase of a sub-signal intermediate point of each frame to obtain a fourth characteristic, and combining the fourth characteristic into a characteristic vector according to the sequence of the sub-signals to obtain an instantaneous phase characteristic vector of the intermediate point of the frame;
and combining the short-time Fourier highest point feature vector, the short-time energy feature vector, the frame similarity feature vector and the frame intermediate point instantaneous phase feature vector as feature matrixes to obtain mode features corresponding to the modulation signals.
In one possible implementation, constructing the modulation signal mode selection model includes:
constructing an input layer to receive mode features corresponding to the modulation signals;
constructing a first convolution layer, a first Relu activation function layer, a first maximum pooling layer, a second convolution layer, a second Relu activation function layer, a second maximum pooling layer, a third convolution layer, a third Relu activation function layer, a third maximum pooling layer, a fourth Relu activation function layer, a full connection layer and a softmax activation function layer which are sequentially connected, so as to process mode characteristics corresponding to a modulation signal and determine a modulation mode corresponding to the modulation signal;
And constructing an output layer to output a modulation mode corresponding to the modulation signal.
In one possible implementation, training the modulated signal pattern selection model using a parameter optimization algorithm includes:
introducing a chaotic sequence to initialize network parameters of a modulation signal mode selection model to obtain an initial population, wherein the initial population comprises a plurality of individuals, and each individual comprises all network parameters to be optimized in the modulation signal mode selection model;
acquiring an fitness value of each individual in the initial population, and determining the individual with the largest fitness value as an optimal individual;
dividing the individuals except the optimal individuals into two parts according to a preset proportion, wherein one part forms a first optimized population, and the other part forms a second optimized population;
updating individuals in the first optimized population by adopting a local searching and updating method aiming at the first optimized population to obtain an updated first optimized population, and updating optimal individuals according to the updated first optimized population to obtain updated optimal individuals;
aiming at the second optimized population, guiding and updating the individuals in the second optimized population according to the updated optimal individuals to obtain an updated second optimized population;
Aiming at the updated first optimized population and second optimized population, carrying out secondary updating on all individuals except the optimal individuals based on the overall position to obtain the second updated first optimized population and second optimized population;
re-determining optimal individuals and at least one worst individual with the worst fitness value according to the second updated first optimized population and the updated second optimized population;
for the optimal individual, global searching and updating method is adopted to conduct global optimization on the optimal individual; generating a new individual for the worst individual, and replacing the worst individual with the new individual;
judging whether the iteration termination condition is met, if so, ending the training, taking the individual with the largest fitness value as the final network parameter of the modulation signal mode selection model, otherwise, repeatedly updating the first optimized population and the second optimized population until the iteration termination condition is met, ending the training, and taking the individual with the largest fitness value as the final network parameter of the modulation signal mode selection model.
In one possible implementation, the network parameter initialization that introduces the chaotic sequence to the modulated signal pattern selection model, obtains an initial population, includes:
Randomly generating a network parameter of a modulation signal mode selection model between an upper limit and a lower limit of the network parameter to obtain a target network parameter;
generating a chaotic sequence by adopting a chaotic mapping strategy based on the target network parameters, and taking the chaotic sequence as an individual, wherein the individual is used for representing a network parameter matrix of a modulation signal mode selection model;
the chaotic mapping strategy is as follows:
wherein x is n Representing the nth network parameter, x, in the network parameter matrix n+1 N=1, 2, …, N-1, N represents the total number of network parameters in the network parameter matrix, λ represents a constant number;
and repeatedly generating a plurality of individuals to obtain an initial population.
In one possible implementation manner, for a first optimized population, updating an individual in the first optimized population by using a search updating method to obtain an updated first optimized population, and updating an optimal individual according to the updated first optimized population to obtain an updated optimal individual, including:
determining search directions p=1, 2, …, h, h representing the total number of search directions;
based on the search direction p, the updated values for the individuals in the first optimized population are:
Wherein x is id Representing individual x i D=1, 2, …, N representing the total number of network parameters in the network parameter matrix,representing individual x i The updated value of the d-th network parameter in the searching direction p, pi represents the circumference ratio, and s1 represents the searching step length;
judging network parameter x id Updated toIndividual x thereafter i Whether the fitness value of (a) increases, if so, an update value is used +.>Replacement network parameter x id Otherwise, the network parameter x is reserved id ;
Judging whether the searching direction p is equal to or greater than h, if so, finishing single updating of the individuals in the first optimized population, otherwise, adding one to the searching direction p, and continuing updating of the individuals in the first optimized population;
judging whether the fitness value of the individual in the first optimized population is larger than that of the optimal individual, if so, finishing updating of the first population, obtaining an updated first optimized population, putting the optimal individual into the first optimized population, taking out the individual with the largest fitness value in the first optimized population, taking the individual as a new optimal individual, and otherwise, judging the updating times;
judging whether the number of times of single updating is larger than the set maximum number of times of single updating, if so, finishing updating the first population, obtaining an updated first optimized population, keeping the optimal individuals unchanged, and otherwise, carrying out single updating on the individuals in the first optimized population again.
In one possible implementation manner, for the second optimized population, guiding and updating the individuals in the second optimized population according to the updated optimal individuals to obtain an updated second optimized population, including:
wherein,,representing individuals x in the second optimized population at the kth global update i D-th network parameter of->Representing updated->The d-th network parameter representing the optimal individual at the kth global update.
In one possible embodiment, for the updated first optimized population and second optimized population, based on the overall location, the second updating is performed for all the individuals except the optimal individual as:
according to the overall positionThe updated values for all individuals except the optimal individual are determined as:
wherein,,representing individual x at the kth global update j Is the first of (2)d network parameters, ">Representing updated x j Lambda represents [ -1,1]Random numbers uniformly distributed among them, ">Representing the average of all individual d-th network parameters at the kth global update;
judging updated individual x j Whether the fitness value of (a) is greater than x before update j And (3) if so, receiving the secondary update, otherwise, refusing the secondary update.
In one possible implementation manner, for an optimal individual, a global search update method is used to perform global optimization on the optimal individual, including:
the selection probability P is determined as follows:
P=-exp(1-k/T max ) 20 +θ 1
wherein k represents the current overall update times, T max Representing a preset maximum overall update number, θ 1 Represents the adjustment parameter, θ 1 ∈(0,0.1);
Generating a decision factor gamma between (0 and 1) and judging whether the decision factor gamma is greater than or equal to the selection probability P, if so, performing global optimization on the optimal individual by adopting a first global search strategy, otherwise, performing global optimization on the optimal individual by adopting a second global search strategy;
the first global search strategy includes:
the updated value of the optimal individual is determined as follows:
η=(θ 2 (f best -f worst ))/T max
wherein,,the (d) th network parameter representing the optimal individual at the kth global update,/th>Representing updated->Cauchy (0, 1) represents a standard Cauchy random distribution, η represents variation intensity, θ 2 Represents the control coefficient of the variable pressure intensity, f best Indicating the fitness value of the optimal individual, f worst Representing the worst fitness value among all individuals;
judging whether the fitness value corresponding to the updated optimal individual is larger than the fitness value corresponding to the optimal individual before updating, if so, accepting the updating of the optimal individual, otherwise, refusing the updating of the optimal individual;
The second global search strategy includes:
the updated value of the optimal individual is determined as follows:
wherein,,representing the optimal individual at the kth global update, ub representing a 1×n upper network parameter limit matrix, lb representing a 1×n lower network parameter limit matrix, r representing a 1×n random number matrix uniformly distributed in compliance with the (0, 1) standard, b representing a 1×n ac coefficient matrix, and b each element being (T) max -k/T max ),/>Representing a point-to-point multiplication;
judging whether the fitness value corresponding to the updated optimal individual is larger than the fitness value corresponding to the optimal individual before updating, if so, accepting the updating of the optimal individual, otherwise, refusing the updating of the optimal individual.
In one possible implementation manner, the method includes identifying a mode feature by using a trained modulation signal mode selection model to obtain a target modulation mode corresponding to a modulation signal, and demodulating the modulation signal by modulating a demodulation model corresponding to the target modulation mode to obtain a signal demodulation result, including:
identifying mode characteristics by adopting a trained modulation signal mode selection model to obtain a target modulation mode corresponding to a modulation signal, wherein the target modulation mode is an AM modulation mode, an ASK modulation mode, an FSK modulation mode, a BPSK modulation mode, a QPSK modulation mode or an MSK modulation mode;
When the target modulation mode is an AM modulation mode, an AM demodulation model is modulated to demodulate a modulation signal, and a signal demodulation result is obtained;
when the target modulation mode is an ASK modulation mode, modulating an ASK demodulation model to demodulate a modulation signal to obtain a signal demodulation result;
when the target modulation mode is the FSK modulation mode, modulating the FSK demodulation model to demodulate the modulation signal, and obtaining a signal demodulation result;
when the target modulation mode is a BPSK modulation mode, a BPSK demodulation model is modulated to demodulate a modulation signal, and a signal demodulation result is obtained;
when the target modulation mode is QPSK modulation mode, modulating a QPSK demodulation model to demodulate a modulation signal to obtain a signal demodulation result;
when the target modulation mode is the MSK modulation mode, the MSK demodulation model is modulated to demodulate the modulation signal, and a signal demodulation result is obtained.
According to the signal demodulation method based on mode selection, the modulation signal mode selection model is built, and the built modulation signal mode selection model is trained, so that the type of a modulation signal is identified through the modulation signal mode selection model which is completed through training, and then the corresponding demodulation model is selected to demodulate the modulation signal, the complexity of an automatic demodulation process is effectively reduced, and the demodulation efficiency is improved.
Drawings
In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings that are needed in the examples will be briefly described below, it being understood that the following drawings only illustrate some examples of the present invention and therefore should not be considered as limiting the scope, and that other related drawings may be obtained from these drawings without inventive effort for a person skilled in the art. In the drawings:
fig. 1 is a flowchart of a signal demodulation method based on mode selection according to an embodiment of the present invention.
Detailed Description
For the purpose of making apparent the objects, technical solutions and advantages of the present invention, the present invention will be further described in detail with reference to the following examples and the accompanying drawings, wherein the exemplary embodiments of the present invention and the descriptions thereof are for illustrating the present invention only and are not to be construed as limiting the present invention.
Examples
As shown in fig. 1, a signal demodulation method based on mode selection includes:
s1, acquiring a modulation signal transmitted by external equipment, and extracting characteristics of the modulation signal to obtain mode characteristics corresponding to the modulation signal.
The characteristics corresponding to different modulation signals are different, so that the mode characteristics corresponding to the modulation signals can be extracted for identification, and the modulation mode corresponding to the modulation signals can be determined.
S2, constructing a modulation signal mode selection model, and training the modulation signal mode selection model by adopting a parameter optimization algorithm to obtain a trained modulation signal mode selection model.
The mode selection model of the modulation signal can be constructed by adopting the deep learning network, and then the mode characteristics are learned by the deep learning network, so that the modulation mode corresponding to the modulation signal can be accurately calculated.
S3, identifying mode characteristics by using a trained modulation signal mode selection model to obtain a target modulation mode corresponding to the modulation signal, and demodulating the modulation signal by modulating a demodulation model corresponding to the target modulation mode to obtain a signal demodulation result.
In one possible implementation manner, obtaining a modulated signal transmitted by an external device, and extracting features of the modulated signal to obtain mode features corresponding to the modulated signal, where the method includes:
an AM signal, an ASK signal, an FSK signal, a BPSK signal, a QPSK signal or an MSK signal transmitted by an external device is acquired to obtain a modulated signal.
It should be noted that, in addition to the above-mentioned several modulation signals, other modulation signals may be identified, and the present invention mainly performs pattern recognition and demodulation for the several modulation signals.
Discretizing the modulation signal, dividing the discretized modulation signal into m frames according to a preset frame division length L, and obtaining m frame sub-signals, wherein the frame division length L is used for representing that each frame of sub-signals comprises L signal points.
For example, the modulated signal may be discretized and then the frame length is set to 128, and one frame is taken every 128 points, so that the modulated signal may be framed into m-frame sub-signals x 1 (n),x 2 (n),...,x m (n),n=1,2,...,128,x m (n) represents an nth point in the mth frame sub-signal, and zero padding is performed on the last frame sub-signal for which the possible length is less than 128.
And carrying out Fourier transform on each frame of sub-signal to obtain a first target sub-signal, and determining the frequency of the highest point in the first target sub-signal to obtain a first characteristic.
Optionally, the fourier transform of the mth frame sub-signal is:
wherein,,X m (k) Represents the fourier transform of the kth point in the mth frame sub-signal, k=1, 2.
The highest point frequency of the m-th frame sub-signal after Fourier transformation is obtained is as follows:
max1=argmax(X m (k))
and combining all the first features according to the sequence of the sub-signals to obtain a feature vector of the highest point of the short-time Fourier.
And obtaining short-time energy of each frame of sub-signals, obtaining second features, and combining all the second features according to the sequence of the sub-signals to obtain feature vectors of the short-time energy.
Short-time energy refers to energy in a frame of a sub-signal, specifically:
and obtaining the similarity between each frame of sub-signal and the next frame of sub-signal to obtain third features, and combining all the third features according to the sequence of the sub-signals to obtain a feature vector of the frame similarity.
The similarity between each frame of sub-signal and the next frame of sub-signal is obtained as follows:
wherein x is m Representing the m-th frame sub-signal, x m+1 Represents the m+1st frame sub-signal, cov (x m ,x m+1 ) Representing the covariance of the mth frame sub-signal and the (m+1) th frame sub-signal, delta xm Representing the variance, delta, of the m-th frame sub-signal xm+1 Representing the variance of the m+1st frame sub-signal.
And obtaining the instantaneous phase of the intermediate point of each frame of sub-signals to obtain a fourth characteristic, and combining the fourth characteristic into a characteristic vector according to the sequence of the sub-signals to obtain the instantaneous phase characteristic vector of the intermediate point of the frame.
The instantaneous phase of the intermediate point of each frame of sub-signal is obtained as follows:
wherein x is m (z) represents an intermediate point of the mth frame sub-signal.
And combining the short-time Fourier highest point feature vector, the short-time energy feature vector, the frame similarity feature vector and the frame intermediate point instantaneous phase feature vector as feature matrixes to obtain mode features corresponding to the modulation signals.
Alternatively, the mode feature a may be:
wherein maxm represents a short-time Fourier highest point of the mth frame sub-signal, em represents short-time energy of the mth frame sub-signal, ρm represents frame similarity of the mth frame sub-signal,representing the instantaneous phase of the frame midpoint of the mth frame sub-signal.
Alternatively, a constellation diagram of the modulation signal may be selected as the corresponding mode feature, and then a modulation signal mode selection model is constructed and trained, so as to perform modulation mode selection.
In one possible implementation, constructing the modulation signal mode selection model includes:
the input layer is constructed to receive pattern features corresponding to the modulated signal.
The method comprises the steps of constructing a first convolution layer, a first Relu activation function layer, a first maximum pooling layer, a second convolution layer, a second Relu activation function layer, a second maximum pooling layer, a third convolution layer, a third Relu activation function layer, a third maximum pooling layer, a fourth Relu activation function layer, a full connection layer and a softmax activation function layer which are sequentially connected, and processing mode characteristics corresponding to a modulation signal to determine a modulation mode corresponding to the modulation signal.
And constructing an output layer to output a modulation mode corresponding to the modulation signal.
It should be noted that, in addition to the modulation signal mode selection model described in this embodiment, other neural networks may be used to construct the modulation signal mode selection model. After the modulation signal mode selection model is constructed, training is carried out on the modulation signal mode selection model, so that the recognition of the modulation mode can be realized.
Alternatively, the first convolution layer may be 5×5 in size, with 32 filters; the size of the first maximum pooling layer may be 3×3; the second convolution layer may be 5 x 5 in size with 32 filters; the second maximum pooling layer may be 3 x 3 in size; the third convolution layer may be 5 x 5 in size with 64 filters; the third maximum pooling layer may be 3 x 3 in size.
In one possible implementation, training the modulated signal pattern selection model using a parameter optimization algorithm includes:
and introducing a chaotic sequence to initialize network parameters of the modulation signal mode selection model to obtain an initial population, wherein the initial population comprises a plurality of individuals, and each individual comprises all network parameters to be optimized in the modulation signal mode selection model. The network parameters to be optimized may be weights and offsets.
And acquiring the fitness value of each individual in the initial population, and determining the individual with the largest fitness value as the optimal individual.
The fitness value of each individual in the initial population may be obtained by:
wherein,,representing the expected output of the first neuron of the output layer when the p-th input sample is input, y pl Representing the p-th input sample input-time output layerThe actual output of the first neuron, P represents the total number of samples, and L represents the total number of neurons in the output layer.
Alternatively, training data may be obtained, where the training data includes mode features corresponding to the training modulation signal and corresponding truth labels (i.e., modulation modes), and then the mode features corresponding to the training modulation signal are used as inputs to the modulation signal mode selection model, and the corresponding truth labels are used as desired outputs to the modulation signal mode selection model, so that fitness values of each individual may be obtained.
Dividing the individuals except the optimal individuals into two parts according to a preset proportion, wherein one part forms a first optimized population, and the other part forms a second optimized population.
And updating the individuals in the first optimized population by adopting a local searching and updating method aiming at the first optimized population to obtain an updated first optimized population, and updating the optimal individuals according to the updated first optimized population to obtain updated optimal individuals.
And aiming at the second optimized population, guiding and updating the individuals in the second optimized population according to the updated optimal individuals to obtain the updated second optimized population.
And aiming at the updated first optimized population and second optimized population, carrying out secondary updating on all the individuals except the optimal individuals based on the overall position to obtain the second updated first optimized population and second optimized population.
And re-determining the optimal individuals and at least one worst individual with the worst fitness value according to the second updated first optimized population and the updated second optimized population.
And for the optimal individual, performing global optimization on the optimal individual by adopting a global search updating method. A new individual is generated for the worst individual and the worst individual is replaced with the new individual.
Judging whether the iteration termination condition is met, if so, ending the training, taking the individual with the largest fitness value as the final network parameter of the modulation signal mode selection model, otherwise, repeatedly updating the first optimized population and the second optimized population until the iteration termination condition is met, ending the training, and taking the individual with the largest fitness value as the final network parameter of the modulation signal mode selection model.
According to the parameter optimization algorithm provided by the embodiment, through searching in all directions, the parameter optimization algorithm can quickly approach to a local optimal value, has global optimization capability, can quickly converge, and can avoid sinking into the local optimal value.
Alternatively, the iteration termination condition may be: when the individual has a fitness value greater than a set threshold or the overall number of iterations reaches a set maximum number.
In one possible implementation, the network parameter initialization that introduces the chaotic sequence to the modulated signal pattern selection model, obtains an initial population, includes:
and randomly generating a network parameter of a modulation signal mode selection model between the upper limit and the lower limit of the network parameter to obtain the target network parameter.
Based on the target network parameters, a chaotic mapping strategy is adopted to generate a chaotic sequence, and the chaotic sequence is used as an individual which is used for representing a network parameter matrix of a modulation signal mode selection model.
The chaotic mapping strategy is as follows:
wherein x is n Representing the nth network parameter, x, in the network parameter matrix n+1 Represents the n+1th network parameter in the network parameter matrix, n=1, 2, …, N-1, N represents the total number of network parameters in the network parameter matrix, and λ represents a constant number.
And repeatedly generating a plurality of individuals to obtain an initial population.
The chaotic sequence has the advantages of regularity, randomness, ergodic property and the like, the global optimizing capability of an algorithm can be effectively improved, and compared with complete randomization, the chaotic sequence has better diversity for individuals.
In one possible implementation manner, for a first optimized population, updating an individual in the first optimized population by using a search updating method to obtain an updated first optimized population, and updating an optimal individual according to the updated first optimized population to obtain an updated optimal individual, including:
the search direction p=1, 2, …, h, h representing the total number of search directions is determined.
Based on the search direction p, the updated values for the individuals in the first optimized population are:
wherein x is id Representing individual x i D=1, 2, …, N representing the total number of network parameters in the network parameter matrix,representing individual x i The updated value of the d-th network parameter in the search direction p, pi represents the circumference ratio, s1 represents the search step.
Judging network parameter x id Updated toIndividual x thereafter i Whether the fitness value of (a) increases, if so, an update value is used +.>Replacement network parameter x id Otherwise, the network parameter x is reserved id 。
And judging whether the searching direction p is equal to or greater than h, if so, finishing single updating of the individuals in the first optimized population, otherwise, adding one to the searching direction p, and continuing updating of the individuals in the first optimized population.
Judging whether the fitness value of the individual in the first optimized population is larger than that of the optimal individual, if so, finishing updating the first population, obtaining an updated first optimized population, putting the optimal individual into the first optimized population, taking out the individual with the largest fitness value in the first optimized population, taking the individual as a new optimal individual, and otherwise, judging the updating times.
Judging whether the number of times of single updating is larger than the set maximum number of times of single updating, if so, finishing updating the first population, obtaining an updated first optimized population, keeping the optimal individuals unchanged, and otherwise, carrying out single updating on the individuals in the first optimized population again.
The optimal individuals represent the individuals with the best fitness value in the current population, and other individuals can quickly approach the optimal individuals by searching in different directions, so that the local optimal points are found.
In one possible implementation manner, for the second optimized population, guiding and updating the individuals in the second optimized population according to the updated optimal individuals to obtain an updated second optimized population, including:
Wherein,,representing individuals x in the second optimized population at the kth global update i D-th network parameter of->Representing updated->The d-th network parameter representing the optimal individual at the kth global update.
In one possible embodiment, for the updated first optimized population and second optimized population, based on the overall location, the second updating is performed for all the individuals except the optimal individual as:
according to the whole bodyPosition ofThe updated values for all individuals except the optimal individual are determined as:
wherein,,representing individual x at the kth global update j D-th network parameter of->Representing updated x j Lambda represents [ -1,1]Random numbers uniformly distributed among them, ">Represents the average of all individual, d-th network parameters at the kth global update.
Judging updated individual x j Whether the fitness value of (a) is greater than x before update j And (3) if so, receiving the secondary update, otherwise, refusing the secondary update.
In the initial stage of the algorithm, the search space is relatively large, so that the search is performed in a relatively large step size, and unnecessary search time can be avoided; in the later period of searching, the searching range becomes narrow, and the final optimizing is performed by using a smaller step length at the moment so as to improve the fineness of searching. If the search step length is too large, the accuracy of algorithm optimization is affected; if the search step is too small, the convergence speed of the algorithm is affected, that is, when the maximum number of iterations is reached, the optimal solution has not been found yet. Therefore, the invention adopts self-adaptive step length to carry out optimizing.
In one possible implementation manner, for an optimal individual, a global search update method is used to perform global optimization on the optimal individual, including:
the selection probability P is determined as follows:
P=-exp(1-k/T max ) 20 +θ 1
wherein k represents the current overall update times, T max Representing a preset maximum overall update number, θ 1 Represents the adjustment parameter, θ 1 ∈(0,0.1)。
And generating a decision factor gamma between the (0 and 1) and judging whether the decision factor gamma is larger than or equal to the selection probability P, if so, performing global optimization on the optimal individual by adopting a first global search strategy, otherwise, performing global optimization on the optimal individual by adopting a second global search strategy.
The first global search strategy includes:
the updated value of the optimal individual is determined as follows:
η=(θ 2 (f best -f worst ))/T max
wherein,,the (d) th network parameter representing the optimal individual at the kth global update,/th>Representing updated->Cauchy (0, 1) represents a standard Cauchy random distribution, η represents variation intensity, θ 2 Represents the control coefficient of the variable pressure intensity, f best Indicating the fitness value of the optimal individual, f worst Representing the worst fitness value among all individuals.
Judging whether the fitness value corresponding to the updated optimal individual is larger than the fitness value corresponding to the optimal individual before updating, if so, accepting the updating of the optimal individual, otherwise, refusing the updating of the optimal individual.
The second global search strategy includes:
the updated value of the optimal individual is determined as follows:
wherein,,representing the optimal individual at the kth global update, ub representing a 1×n upper network parameter limit matrix, lb representing a 1×n lower network parameter limit matrix, r representing a 1×n random number matrix uniformly distributed in compliance with the (0, 1) standard, b representing a 1×n ac coefficient matrix, and b each element being (T) max -k/T max ),/>Representing a point-to-point multiplication.
Judging whether the fitness value corresponding to the updated optimal individual is larger than the fitness value corresponding to the optimal individual before updating, if so, accepting the updating of the optimal individual, otherwise, refusing the updating of the optimal individual.
Through the first global search strategy and the second global search strategy, the efficiency of solving the global optimum can be effectively improved, and the algorithm is prevented from falling into a local optimum solution.
In one possible implementation manner, the method includes identifying a mode feature by using a trained modulation signal mode selection model to obtain a target modulation mode corresponding to a modulation signal, and demodulating the modulation signal by modulating a demodulation model corresponding to the target modulation mode to obtain a signal demodulation result, including:
and identifying the mode characteristics by adopting a trained modulation signal mode selection model to obtain a target modulation mode corresponding to the modulation signal, wherein the target modulation mode is an AM (Amplitude Modulation) modulation mode, an ASK (amplitude modulation, amplitude keying) modulation mode, an FSK (Frequency-shift keying) modulation mode, a BPSK (Binary Phase Shift Keying ) modulation mode, a QPSK (Quadrature Phase Shift Keying, quadrature phase shift keying) modulation mode or an MSK (Minimum Shift Keying, minimum Frequency shift keying) modulation mode.
And when the target modulation mode is an AM modulation mode, modulating an AM demodulation model to demodulate the modulation signal to obtain a signal demodulation result.
And when the target modulation mode is an ASK modulation mode, modulating an ASK demodulation model to demodulate the modulated signal, and obtaining a signal demodulation result.
When the target modulation mode is the FSK modulation mode, modulating the FSK demodulation model to demodulate the modulation signal, and obtaining a signal demodulation result.
And when the target modulation mode is the BPSK modulation mode, a BPSK demodulation model is modulated to demodulate the modulation signal, and a signal demodulation result is obtained.
And when the target modulation mode is QPSK modulation mode, modulating the QPSK demodulation model to demodulate the modulated signal to obtain a signal demodulation result.
When the target modulation mode is the MSK modulation mode, the MSK demodulation model is modulated to demodulate the modulation signal, and a signal demodulation result is obtained.
According to the signal demodulation method based on mode selection, the modulation signal mode selection model is built, and the built modulation signal mode selection model is trained, so that the type of a modulation signal is identified through the modulation signal mode selection model which is completed through training, and then the corresponding demodulation model is selected to demodulate the modulation signal, the complexity of an automatic demodulation process is effectively reduced, and the demodulation efficiency is improved.
The foregoing description of the embodiments has been provided for the purpose of illustrating the general principles of the invention, and is not meant to limit the scope of the invention, but to limit the invention to the particular embodiments, and any modifications, equivalents, improvements, etc. that fall within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. A method for demodulating a signal based on mode selection, comprising:
acquiring a modulation signal transmitted by external equipment, and extracting characteristics of the modulation signal to obtain mode characteristics corresponding to the modulation signal;
constructing a modulation signal mode selection model, and training the modulation signal mode selection model by adopting a parameter optimization algorithm to obtain a trained modulation signal mode selection model;
and identifying the mode characteristics by adopting the trained modulation signal mode selection model to obtain a target modulation mode corresponding to the modulation signal, and demodulating the modulation signal by modulating a demodulation model corresponding to the target modulation mode to obtain a signal demodulation result.
2. The signal modulation method based on mode selection according to claim 1, wherein obtaining a modulated signal transmitted by an external device, and extracting features of the modulated signal to obtain mode features corresponding to the modulated signal, comprises:
Acquiring an AM signal, an ASK signal, an FSK signal, a BPSK signal, a QPSK signal or an MSK signal transmitted by external equipment to obtain a modulation signal;
discretizing the modulation signal, dividing the discretized modulation signal into M frames according to a preset frame division length L, and obtaining M frame sub-signals, wherein the frame division length L is used for representing that each frame of sub-signals comprises L signal points;
performing Fourier transform on each frame of sub-signal to obtain a first target sub-signal, and determining the frequency of the highest point in the first target sub-signal to obtain a first characteristic;
combining all the first features according to the sequence of the sub-signals to obtain a feature vector of the highest point of the short-time Fourier;
short-time energy of each frame of sub-signals is obtained, second characteristics are obtained, all the second characteristics are combined into characteristic vectors according to the sequence of the sub-signals, and short-time energy characteristic vectors are obtained;
obtaining the similarity between each frame of sub-signal and the next frame of sub-signal to obtain third characteristics, and combining all the third characteristics into characteristic vectors according to the sequence of the sub-signals to obtain frame similarity characteristic vectors;
acquiring the instantaneous phase of a sub-signal intermediate point of each frame to obtain a fourth characteristic, and combining the fourth characteristic into a characteristic vector according to the sequence of the sub-signals to obtain an instantaneous phase characteristic vector of the intermediate point of the frame;
And combining the short-time Fourier highest point feature vector, the short-time energy feature vector, the frame similarity feature vector and the frame intermediate point instantaneous phase feature vector as feature matrixes to obtain mode features corresponding to the modulation signals.
3. The mode selection-based signal modulation method of claim 2, wherein constructing a modulation signal mode selection model comprises:
constructing an input layer to receive mode features corresponding to the modulation signals;
constructing a first convolution layer, a first Relu activation function layer, a first maximum pooling layer, a second convolution layer, a second Relu activation function layer, a second maximum pooling layer, a third convolution layer, a third Relu activation function layer, a third maximum pooling layer, a fourth Relu activation function layer, a full connection layer and a softmax activation function layer which are sequentially connected, so as to process mode characteristics corresponding to a modulation signal and determine a modulation mode corresponding to the modulation signal;
and constructing an output layer to output a modulation mode corresponding to the modulation signal.
4. A mode selection based signal modulation method as claimed in claim 3 wherein training the modulated signal mode selection model using a parameter optimization algorithm comprises:
Introducing a chaotic sequence to initialize network parameters of a modulation signal mode selection model to obtain an initial population, wherein the initial population comprises a plurality of individuals, and each individual comprises all network parameters to be optimized in the modulation signal mode selection model;
acquiring an fitness value of each individual in the initial population, and determining the individual with the largest fitness value as an optimal individual;
dividing the individuals except the optimal individuals into two parts according to a preset proportion, wherein one part forms a first optimized population, and the other part forms a second optimized population;
updating individuals in the first optimized population by adopting a local searching and updating method aiming at the first optimized population to obtain an updated first optimized population, and updating optimal individuals according to the updated first optimized population to obtain updated optimal individuals;
aiming at the second optimized population, guiding and updating the individuals in the second optimized population according to the updated optimal individuals to obtain an updated second optimized population;
aiming at the updated first optimized population and second optimized population, carrying out secondary updating on all individuals except the optimal individuals based on the overall position to obtain the second updated first optimized population and second optimized population;
Re-determining optimal individuals and at least one worst individual with the worst fitness value according to the second updated first optimized population and the updated second optimized population;
for the optimal individual, global searching and updating method is adopted to conduct global optimization on the optimal individual; generating a new individual for the worst individual, and replacing the worst individual with the new individual;
judging whether the iteration termination condition is met, if so, ending the training, taking the individual with the largest fitness value as the final network parameter of the modulation signal mode selection model, otherwise, repeatedly updating the first optimized population and the second optimized population until the iteration termination condition is met, ending the training, and taking the individual with the largest fitness value as the final network parameter of the modulation signal mode selection model.
5. The mode selection-based signal modulation method of claim 4 wherein introducing a chaotic sequence to initialize network parameters of a mode selection model of the modulated signal to obtain an initial population comprises:
randomly generating a network parameter of a modulation signal mode selection model between an upper limit and a lower limit of the network parameter to obtain a target network parameter;
Generating a chaotic sequence by adopting a chaotic mapping strategy based on the target network parameters, and taking the chaotic sequence as an individual, wherein the individual is used for representing a network parameter matrix of a modulation signal mode selection model;
the chaotic mapping strategy is as follows:
wherein x is n Representing the nth network parameter, x, in the network parameter matrix n+1 N=1, 2, …, N-1, N represents the total number of network parameters in the network parameter matrix, λ represents a constant number;
and repeatedly generating a plurality of individuals to obtain an initial population.
6. The pattern selection-based signal modulation method of claim 4, wherein for the first optimized population, updating the individuals in the first optimized population by a search update method to obtain an updated first optimized population, and updating the optimal individuals according to the updated first optimized population to obtain updated optimal individuals, comprising:
determining search directions p=1, 2, …, h, h representing the total number of search directions;
based on the search direction p, the updated values for the individuals in the first optimized population are:
wherein x is id Representing individual x i D=1, 2, …, N representing the total number of network parameters in the network parameter matrix, Representing individual x i The updated value of the d-th network parameter in the searching direction p, pi represents the circumference ratio, and s1 represents the searching step length;
judging network parameter x id Updated toIndividual x thereafter i Whether the fitness value of (a) increases, if so, an update value is used +.>Replacement network parameter x id Otherwise, the network parameter x is reserved id ;
Judging whether the searching direction p is equal to or greater than h, if so, finishing single updating of the individuals in the first optimized population, otherwise, adding one to the searching direction p, and continuing updating of the individuals in the first optimized population;
judging whether the fitness value of the individual in the first optimized population is larger than that of the optimal individual, if so, finishing updating of the first population, obtaining an updated first optimized population, putting the optimal individual into the first optimized population, taking out the individual with the largest fitness value in the first optimized population, taking the individual as a new optimal individual, and otherwise, judging the updating times;
judging whether the number of times of single updating is larger than the set maximum number of times of single updating, if so, finishing updating the first population, obtaining an updated first optimized population, keeping the optimal individuals unchanged, and otherwise, carrying out single updating on the individuals in the first optimized population again.
7. The pattern selection-based signal modulation method of claim 4, wherein for the second optimized population, guiding the updating of the individuals in the second optimized population according to the updated optimal individuals to obtain an updated second optimized population, comprising:
8. The pattern selection-based signal modulation method of claim 4, wherein the second updating of all individuals except the optimal individual based on the overall location is performed for the updated first optimized population and the second optimized population as:
according to the overall positionThe updated values for all individuals except the optimal individual are determined as:
wherein,,representing individual x at the kth global update j D-th network parameter of->Representing updated x j Lambda represents [ -1,1]Random numbers uniformly distributed among them, ">Representing the average of all individual d-th network parameters at the kth global update;
judging updated individual x j Whether the fitness value of (a) is greater than x before update j And (3) if so, receiving the secondary update, otherwise, refusing the secondary update.
9. The mode selection-based signal modulation method according to claim 4, wherein the global search updating method is adopted for the optimal individual to perform global optimization on the optimal individual, comprising:
the selection probability P is determined as follows:
P=-exp(1-k/T max ) 20 +θ 1
wherein k represents the current overall update times, T max Representing a preset maximum overall update number, θ 1 Represents the adjustment parameter, θ 1 ∈(0,0.1);
Generating a decision factor gamma between (0 and 1) and judging whether the decision factor gamma is greater than or equal to the selection probability P, if so, performing global optimization on the optimal individual by adopting a first global search strategy, otherwise, performing global optimization on the optimal individual by adopting a second global search strategy;
the first global search strategy includes:
the updated value of the optimal individual is determined as follows:
η=(θ 2 (f best -f worst ))/T max
wherein,,the (d) th network parameter representing the optimal individual at the kth global update,/th>Representing updatedCauchy (0, 1) represents a standard Cauchy random distribution, η represents variation intensity, θ 2 Represents the control coefficient of the variable pressure intensity, f best Indicating the fitness value of the optimal individual, f worst Representing the worst fitness value among all individuals;
Judging whether the fitness value corresponding to the updated optimal individual is larger than the fitness value corresponding to the optimal individual before updating, if so, accepting the updating of the optimal individual, otherwise, refusing the updating of the optimal individual;
the second global search strategy includes:
the updated value of the optimal individual is determined as follows:
wherein,,representing the optimal individual at the kth global update, ub represents the network parameter of 1 XNThe limit matrix, lb, represents a 1×n network parameter lower limit matrix, r represents a 1×n random number matrix uniformly distributed in compliance with the (0, 1) standard, b represents a 1×n ac coefficient matrix, and each element in b is (T max -k/T max ),/>Representing a point-to-point multiplication;
judging whether the fitness value corresponding to the updated optimal individual is larger than the fitness value corresponding to the optimal individual before updating, if so, accepting the updating of the optimal individual, otherwise, refusing the updating of the optimal individual.
10. The method for modulating signals based on mode selection according to claim 2, wherein the step of identifying the mode characteristics by using a trained modulation signal mode selection model to obtain a target modulation mode corresponding to the modulation signal, and modulating a demodulation model corresponding to the target modulation mode to demodulate the modulation signal to obtain a signal demodulation result comprises the steps of:
Identifying mode characteristics by adopting a trained modulation signal mode selection model to obtain a target modulation mode corresponding to a modulation signal, wherein the target modulation mode is an AM modulation mode, an ASK modulation mode, an FSK modulation mode, a BPSK modulation mode, a QPSK modulation mode or an MSK modulation mode;
when the target modulation mode is an AM modulation mode, an AM demodulation model is modulated to demodulate a modulation signal, and a signal demodulation result is obtained;
when the target modulation mode is an ASK modulation mode, modulating an ASK demodulation model to demodulate a modulation signal to obtain a signal demodulation result;
when the target modulation mode is the FSK modulation mode, modulating the FSK demodulation model to demodulate the modulation signal, and obtaining a signal demodulation result;
when the target modulation mode is a BPSK modulation mode, a BPSK demodulation model is modulated to demodulate a modulation signal, and a signal demodulation result is obtained;
when the target modulation mode is QPSK modulation mode, modulating a QPSK demodulation model to demodulate a modulation signal to obtain a signal demodulation result;
when the target modulation mode is the MSK modulation mode, the MSK demodulation model is modulated to demodulate the modulation signal, and a signal demodulation result is obtained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310443259.2A CN116405355A (en) | 2023-04-21 | 2023-04-21 | Signal demodulation method based on mode selection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310443259.2A CN116405355A (en) | 2023-04-21 | 2023-04-21 | Signal demodulation method based on mode selection |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116405355A true CN116405355A (en) | 2023-07-07 |
Family
ID=87007436
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310443259.2A Pending CN116405355A (en) | 2023-04-21 | 2023-04-21 | Signal demodulation method based on mode selection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116405355A (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN113141272A (en) * | 2021-04-20 | 2021-07-20 | 浙江大学 | Network security situation analysis method based on iteration optimization RBF neural network |
CN113762602A (en) * | 2021-08-13 | 2021-12-07 | 中国大唐集团科学技术研究院有限公司西北电力试验研究院 | Short-term wind speed prediction method for wind power plant |
-
2023
- 2023-04-21 CN CN202310443259.2A patent/CN116405355A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN113141272A (en) * | 2021-04-20 | 2021-07-20 | 浙江大学 | Network security situation analysis method based on iteration optimization RBF neural network |
CN113762602A (en) * | 2021-08-13 | 2021-12-07 | 中国大唐集团科学技术研究院有限公司西北电力试验研究院 | Short-term wind speed prediction method for wind power plant |
Non-Patent Citations (11)
Title |
---|
XIN FENG等: ""Infrared and Visible Image Fusion Based on the Total Variational Model and Adaptive Wolf Pack Algorithm"", 《IEEE ACCESS》 * |
刘爱军;杨育;邢青松;陆惠;张煜东;: "含精英策略的小生境遗传退火算法研究及其应用", 中国机械工程, pages 1 - 2 * |
吴虎胜;张凤鸣;吴庐山;: "一种新的群体智能算法――狼群算法", 系统工程与电子技术, no. 11 * |
杨洁;夏卉;: "基于卷积神经网络的通信信号调制识别研究", 计算机测量与控制, no. 07 * |
王威;王培源;彭奇峰;: "基于卷积神经网络STBC调制信号识别", 中国电子科学研究院学报, no. 05 * |
王振东等: ""一种增强型麻雀搜索算法的无线传感器网络覆盖优化研究"", 《传感技术学报》, pages 3 * |
王蒙;于宏毅;胡鹏;代琨;: "基于特征选择和支持向量机的数字调制识别方法", 信息工程大学学报, no. 04 * |
王鹏: ""基于改进麻雀搜索算法的微电网群优化调度研究"", 《中国硕士学位论文全文数据库》, no. 02, pages 4 * |
郭立民;陈鑫;陈涛;: "基于AlexNet模型的雷达信号调制类型识别", 吉林大学学报(工学版), no. 03 * |
韦杏秋;陈碧云;陈绍南;: "基于改进多种群遗传算法的节能环保多目标优化模型", 电网与清洁能源, no. 12, pages 3 * |
黄思嘉: ""基于卷积神经网络的调制模式识别研究"", 《中国硕士学位论文全文数据库》, no. 5, pages 1 - 5 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Hu et al. | Deep neural network for robust modulation classification under uncertain noise conditions | |
Ali et al. | Automatic modulation classification of digital modulation signals with stacked autoencoders | |
CN110113288B (en) | Design and demodulation method of OFDM demodulator based on machine learning | |
Zhang et al. | Automatic modulation classification using a deep multi-stream neural network | |
Ansari et al. | Automatic digital modulation recognition based on genetic-algorithm-optimized machine learning models | |
CN109787926A (en) | A kind of digital signal modulation mode recognition methods | |
CN110659684A (en) | Convolutional neural network-based STBC signal identification method | |
CN112887239A (en) | Method for rapidly and accurately identifying underwater sound signal modulation mode based on deep hybrid neural network | |
Hong et al. | Convolutional neural network aided signal modulation recognition in OFDM systems | |
Huynh-The et al. | MIMO-OFDM modulation classification using three-dimensional convolutional network | |
CN111431825B (en) | Signal automatic classification and identification method based on deep multi-flow neural network | |
CN114615118B (en) | Modulation identification method based on multi-terminal convolution neural network | |
Zhang et al. | Lightweight automatic modulation classification via progressive differentiable architecture search | |
Lin et al. | A real-time modulation recognition system based on software-defined radio and multi-skip residual neural network | |
Almohamad et al. | Dual-determination of modulation types and signal-to-noise ratios using 2D-ASIQH features for next generation of wireless communication systems | |
CN114398931A (en) | Modulation recognition method and system based on numerical characteristic and image characteristic fusion | |
Zhang et al. | High-order convolutional attention networks for automatic modulation classification in communication | |
Ali et al. | Modulation format identification using supervised learning and high-dimensional features | |
Essai et al. | Automatic Modulation Classification: Convolutional Deep Learning Neural Networks Approaches | |
Kumar et al. | Automatic modulation classification for adaptive OFDM systems using convolutional neural networks with residual learning | |
Ya et al. | Modulation recognition of digital signal based on deep auto-ancoder network | |
Indira et al. | Deep learning CNN-based hybrid extreme learning machine with bagging classifier for automatic modulation classification | |
Wang et al. | Underwater communication signal recognition using sequence convolutional network | |
CN116405355A (en) | Signal demodulation method based on mode selection | |
Naderializadeh | Wireless link scheduling via graph representation learning: A comparative study of different supervision levels |
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 |