A signal recognition method combining mixed feature sample expansion and evolved LSTM network

Abstract

The application discloses a signal recognition method of a mixed feature sample expansion combined evolution LSTM network, first, a plurality of mixed features are extracted, and then a sliding window is used for sample expansion to obtain feature samples suitable for an LSTM neural network; in view of the problem that the hyperparameters of a traditional LSTM neural network are difficult to obtain optimal hyperparameters due to dependence on expert experience adjustment, a quantum arctangent mechanism is used for efficient solution, and the application limitation of dependence on expert experience and randomness caused by dependence of the traditional LSTM neural network on artificial selection of hyperparameters is broken through.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of communication signal processing, and particularly relates to a signal recognition method combining mixed feature sample expansion and an evolved LSTM network. BACKGROUND

[0002] Automatic modulation recognition is a technology for automatically distinguishing the modulation types of received signals, and is an intermediate step required for signal detection and information recovery. The automatic modulation recognition technology of communication signals is one of the key technologies in scenarios such as spectrum allocation, electronic countermeasures, and cognitive radio. In the military field, it is necessary to distinguish the modulation types of various communication signals and radar signals emitted by enemy electronic equipment, and then to perform demodulation, even listening and jamming. In the civil field, the modulation recognition technology is applied to the field of cognitive radio, which can cooperate with modules such as parameter estimation and signal demodulation, effectively avoiding radio interference and optimizing spectrum allocation.

[0003] With the continuous development of machine learning, technologies such as decision trees and artificial neural networks have gradually been applied in multiple fields such as image processing and signal processing, including product detection, bankruptcy prediction, medical diagnosis, handwritten character recognition, and automatic modulation recognition. Since machine learning methods do not require manual design of decision thresholds and do not require the calculation of complex likelihood functions, more and more scholars have applied machine learning to modulation signal recognition methods. Evolutionary neural networks have strong pattern recognition capabilities, automatically and adaptively update weights and thresholds for each node, and can better handle complex nonlinear problems to achieve better recognition performance. Therefore, evolutionary neural networks can be used as classifiers to recognize modulation signal types.

[0004] Modulation signal recognition methods based on evolutionary neural networks generally include three basic steps of preprocessing and feature extraction, evolution of network parameters or structure, and classification recognition. Compared with machine learning classifiers such as BP neural networks, K-nearest neighbors, and decision trees, long short-term memory (LSTM) neural networks are good at learning time series information in signals, and therefore can often achieve better results in modulation signal recognition. It is of great significance and value to study a modulation signal recognition method based on quantum polar bear mechanism evolved LSTM neural networks under Gaussian noise.

[0005] Through the search of prior art documents, it is found that Gao Hongyuan et al. in 2023 authorized the invention patent "Modulation signal recognition method of evolutionary long short-term memory neural network" (Heilongjiang Province: CN202111421628.5, 2023-10-03.) directly takes the received signal as the input of the LSTM neural network, proposes a marlin predation search mechanism to optimize the three hyperparameters of the LSTM neural network, including the number of hidden layer neurons, the learning rate and the training round number, and achieves good results, but the process of feature extraction is not carried out, resulting in a large amount of data, and the LSTM neural network still has other hyperparameters that have an important influence on the training results, such as the small batch size that affects the memory consumption and the total weight update times in the training process, and the appropriate dropout probability that can control the complexity of the model to improve the generalization ability of the network; Liu Sheng et al. published "Modulation recognition method based on instantaneous feature parameters" in "Communication Technology" (2022, 55(05): 559-564) extract three kinds of instantaneous features of the signal and use a three-layer fully connected neural network as a classifier for modulation signal recognition, but the single type of features used leads to poor performance under low signal-to-noise ratio; Zhang Wenna et al. published "Digital modulation recognition based on high-order cumulant" in "Foreign Electronic Measurement Technology" (2022, 41(07): 126-132) extract five kinds of high-order cumulant features of the signal and use a decision tree classifier for modulation signal recognition, which uses a single high-order cumulant feature, and the performance is not ideal under low signal-to-noise ratio without sample expansion, and it is difficult to realize the intra-class recognition of MFSK signals.

[0006] The search results of the existing literature show that most of the existing modulation signal recognition methods based on neural networks need to manually adjust the network hyperparameters, have great randomness and are severely dependent on expert experience, even though some technologies use optimization algorithms to optimize the LSTM neural network hyperparameters, the types of hyperparameters found are still insufficient, and some hyperparameters still need to be determined manually; In addition, in view of the problem that the traditional instantaneous or cumulant features do not utilize the time sequence information between samples, which is not conducive to the training of the LSTM neural network, a data preprocessing method is proposed, which expands the feature samples into a data set suitable for the input of the LSTM neural network; In view of the problem that the hyperparameters of the LSTM neural network are difficult to determine, a quantum polar bear mechanism is proposed to evolve the LSTM network hyperparameters, solving the application limitation that the existing modulation signal recognition methods based on LSTM neural network rely on manual selection to obtain optimal hyperparameters. SUMMARY

[0007] The purpose of the present application is to provide a signal recognition method combining mixed feature sample expansion and evolved LSTM network.

[0008] The purpose of the present application is achieved by the following technical solutions:

[0009] A signal recognition method combining mixed feature sample expansion and evolved LSTM network, the steps are as follows:

[0010] Step one: Establish a modulated signal sample library to obtain a modulated signal dataset;

[0011] Step two: Extract instantaneous features and cumulative features from the modulated signal dataset obtained in step one to construct a mixed feature library;

[0012] Step three: Preprocess the mixed features extracted in step two;

[0013] Step four: Determine the objective function;

[0014] Train the LSTM neural network using the training set and the set hyperparameters, and use the misjudgment rate of the LSTM network on the validation set as the objective function, that is, the fitness function F = 1-A c , wherein A c is the recognition accuracy on the validation set; after training, the effect of training is verified by the validation set;

[0015] Step five: Initialize the quantum arctic wolf mechanism parameters;

[0016] Including the number of individuals N q in the population, the maximum number of iterations G max , the search space dimension M, and the upper bound U = [u1, u2, …, u M ] and the lower bound L = [l1, l2, …, l M ], the group behavior probability P, the hunting attack probability Q in the group behavior, the persecution attack probability is 1-Q, the update coefficient w1, and the mutation probability Selection probability Randomly initialize the quantum position of the arctic wolf in [0, 1] and map it to the corresponding position , wherein i ∈ {1, 2, …, N q}, the symbol “⊙” represents the multiplication of the elements in the corresponding position in the matrix or vector, and the superscript 0 represents the iteration number is 0;

[0017] Step six: Map each dimension of the position of the i-th arctic wolf in the population to the hyperparameters of the LSTM network, i ∈ {1, 2, …, N q}, train the LSTM network using the training set and the corresponding hyperparameters, and use the misjudgment rate of the LSTM network on the validation set as the fitness function of the i-th arctic wolf;

[0018] Step seven: Enter the iteration loop, let t = t + 1, and for the i-th arctic wolf in the population, calculate the value of the iteration number related variable m1(t) m1(t) = exp(t / Gmax -1), compared with the value of P, if less than P, enter the group behavior; if m1(t) is greater than or equal to P, carry out the carrion behavior;

[0019] Step eight: for the i-th arctic wolf in the t-th iteration, if the survival probability is greater than 0.5, carry out the survival behavior, and update the quantum rotation angle as wherein m2(t) = exp(1-t / G max )-t / G max , wherein: represents a random number between 0 and 1 uniformly distributed for the i-th arctic wolf in the t-th iteration; otherwise, go to step nine;

[0020] Step nine: update the quantum position of each arctic wolf according to the quantum rotation angle and the quantum gate strategy as wherein: represents the d-th dimension of the quantum position of the i-th arctic wolf in the t-th iteration, is the corresponding updated quantum rotation angle, is a random number between 0 and 1 uniformly distributed, is the mutation probability of the quantum position when the quantum rotation angle is 0;

[0021] Step ten: calculate the position of the i-th arctic wolf in the new generation as convert the elements of each dimension into the corresponding LSTM network hyperparameters, and take the misjudgment rate on the validation set after training the LSTM network as the corresponding fitness function

[0022] Step eleven: adopt the reservation strategy to reserve the updated arctic wolf: if the updated arctic wolf has better fitness, reserve the corresponding position, quantum position and fitness after updating, otherwise do not reserve; update the quantum position corresponding to the optimal individual and the position update the survival probability of each arctic wolf is

[0023] Step twelve: judge whether the iteration number t reaches the maximum iteration number G max , if t = G max , output the optimal arctic wolf position as the optimization result, otherwise return to step seven for iteration;

[0024] Step thirteen: convert the elements of each dimension into the corresponding optimal LSTM network hyperparameters, train the LSTM neural network with the optimal hyperparameters, and test the modulation signal recognition performance of the LSTM network after training is completed, and analyze the results.

[0025] Further, the digital modulation signal in step one includes MASK, MPSK, MFSK, MSK and MQAM, in order to reduce the influence of inter-symbol interference, a shaping filter is added to the MASK, MPSK, MFSK and MQAM signals at the transmitting end, the shaping filter adopts a square root raised cosine roll-off function to perform shaping processing on the digital baseband signal, through a Gaussian white noise channel, the received signal at the receiving end obtains a modulation signal data set.

[0026] Further, the step two is specifically:

[0027] After the receiver receives the signal, a Hilbert transform is performed to obtain an analytic form of the signal, that is, In the formula, s(t) is the analytic signal of the original signal y(t), j is a complex unit, and is the Hilbert transform of y(t), and In the formula, represents convolution operation; and the frequency response is

[0028] The sampling frequency f s The original signal y(t) is sampled to obtain a discrete sequence y(n) with a total number of The analytic form of the original signal y(t) is The instantaneous amplitude is A(n), and then The instantaneous phase is θ(n), and then

[0029] Since the principal value interval of the arctangent function is (-π / 2, π / 2), θ(n) may produce a sudden change of ±π, and the phase is obtained by adjusting it to have a value in [0, 2π).

[0030] The relationship between the actual instantaneous phase ε(n) and is In the formula, mod represents the remainder operation; therefore, There is phase folding; since the unfolded instantaneous phase φ(n) satisfies φ(n) = 2πf c T s n + ε(n) + θ, in the formula, f c is the carrier frequency, T s is the sampling period, and θ is the initial phase; from the above formula, it can be seen that the unfolded instantaneous phase is a linear phase component caused by the carrier frequency and a nonlinear component caused by ε(n) and θ; a correction sequence {c(n)} is added to the sequence to achieve unfolding, which is defined as At this time, the unfolded instantaneous phase estimation value is The estimated value of the de-wrapping instantaneous phase nonlinear component is

[0031] The instantaneous frequency sequence is obtained by differentiating the de-wrapping instantaneous phase sequence, i.e. In the formula, f s is the sampling frequency;

[0032] On the basis of obtaining the instantaneous amplitude, frequency and phase of the signal, further extract the multiple characteristic quantities of the instantaneous information of the digital modulation signal, and obtain six characteristic parameters, including the maximum value of the spectral density of the zero-centered normalized instantaneous amplitude The standard deviation σ of the absolute value of the zero-centered normalized instantaneous amplitude aa The standard deviation σ of the zero-centered non-weak signal segment instantaneous phase nonlinear component dp The standard deviation σ of the absolute value of the zero-centered non-weak signal segment instantaneous phase nonlinear component ap The standard deviation σ of the absolute value of the zero-centered normalized non-weak signal segment instantaneous frequency af The variance of the normalized instantaneous frequency

[0033] After the receiver obtains the received baseband signal through down-conversion and matched filtering, according to the high-order cumulant theory, five effective cumulant characteristics Z1, Z2, Z3, Z4 and Z5 of the signal are extracted; a random process x(t) is set, and the k-order cumulant is defined as C kx (f1,f2,...,f k )=Cum[x(t),x(t+f1),...,x(t+f k-1 )], in which Cum(·) is a cumulant operation, which is represented as the calculation of high-order mixed moments; the p-order mixed moment of the random process is M pq =E{x(t) p-q x * (t) q}, in which x*(t) represents the complex conjugate of the random process, and q is the number of conjugate multiplications; the cumulants of the random process are as follows: 1. The second-order cumulant is represented as C 20 =M 20 , C 21 =M 21 ; 2. The fourth-order cumulant is represented as: C 41 =M 41 -3M 20 M 21 ; 3. The sixth-order cumulant is represented as: In order to reduce the influence of signal energy on the characteristic parameters, the ratio form is used to extract five better performance characteristic parameters as follows:

[0034] The signal is processed in segments, and the characteristic parameters of each segment are extracted, and finally a mixed feature library containing 11 characteristic parameters is obtained.

[0035] Furthermore, the preprocessing of the mixed features in step 3 includes normalization, sample expansion, and division into training set, validation set, and test set. The sample expansion uses a sliding rectangular window to treat three consecutive feature samples as input variables of an LSTM neural network.

[0036] Assume that the feature set composed of the extracted mixed features is M=[m1,m2,...,m J ], where J is the total number of samples, and the i-th feature sample is Where: superscript T is the transposition symbol, superscript (i) represents the i-th sample;

[0037] The feature set is normalized using the zero mean and unit variance normalization method: Calculate the mean μ of each feature in the feature set = [μ1,μ2,...,μ 11 ] T and standard deviation σ=[σ1,σ2,...,σ 11 ] T , then the normalized feature set is Where: the dot ". / " means dividing all elements in each row of the matrix by the elements in the corresponding row of the vector;

[0038] Use a rectangular window with a length of 3 and a moving step of 1 to transform the feature set Truncated to data segments, i.e. The kth sequence sample is Where: Since the sliding window length is 3, the last two samples are incomplete, so the last two incomplete samples are discarded. That is, a feature sequence set N for LSTM network recognition is obtained, where the size of each sequence sample is 11×3; then, the feature sequence set N and the corresponding label T are randomly divided into training set, validation set and test set according to a certain ratio.

[0039] Furthermore, in step 6, the LSTM network hyperparameters to be optimized include the number of hidden layer neurons N h , small batch size S b , number of training rounds I m , learning rate l r and the drop probability p d , the learning rate is mapped to the logarithm of the variable ω4=lg(l r ); Let the row vector of all hyperparameters be h=[N h,S b ,I m ,ω4,p d ]=[h1,h2,h3,h4,h5], let its upper bound be and the lower bound is Through the upper and lower bounds, a linear mapping is performed to the upper bound U and lower bound L of the search space of the quantum Arctic wolf mechanism, that is, in Will

[0040] Mapped vector As the position of the i-th Arctic wolf in the t-th iteration of the quantum Arctic wolf mechanism In the fitness function, the inverse transformation is performed to obtain the corresponding LSTM network hyperparameters, namely Where: i∈{1,2,...,N q}, N q is the number of individuals in the population, d∈{1,2,...,M} is the dth hyperparameter, and the independent variable dimension is M=5; then the actual hyperparameters used for training are: the number of hidden layer neurons Small batch size Number of training rounds In the formula: round represents rounding off decimals; learning rate Drop probability

[0041] The LSTM network is trained using the hyperparameters corresponding to the ith Arctic wolf in the tth iteration. The accuracy on the validation set is Then its fitness function is Then calculate the corresponding survival probability Right now Assume the number of iterations t=0.

[0042] Furthermore, in step 7, for the i-th Arctic wolf in the population, if m1(t)=exp(t / G max -1)<P, then group behavior is initiated, which includes two strategies: hunting attack and persecution attack. The hunting attack strategy is performed with probability Q and the quantum rotation angle is updated as Where: r i t is a normally distributed random number with mean 0 and variance 1 for the tth iteration of the i-th Arctic wolf, n a Is in [2,N q / 2] represents the number of Arctic wolves to be updated. The quantum position of the kth Arctic wolf that attacks the group in the tth iteration is n randomly selected from the population in each iteration. a Made up of different Arctic wolves, is the quantum position of the optimal arctic wolf, c(t) = cos (πt / 2G max ) is a step length coefficient gradually changing from 1 to 0 with iteration; the quantum rotation angle of the persecution attack strategy update is and the update step length is defined as wherein: is the i-th arctic wolf in the t-th iteration satisfying a uniformly distributed random number between (0, 2), and w1 are update coefficients, v t (m) is the average value of the quantum position corresponding to the dimension of all arctic wolves in the t-th iteration population otherwise m1(t) >= P, the quantum rotation angle of the carrion behavior update is wherein: is the quantum position of a randomly selected arctic wolf in the t-th iteration of carrion behavior, and represent the i-th arctic wolf in the t-th iteration satisfying a uniformly distributed random number between [-1, 1], p r is a uniformly distributed random number between [0, 1], is a set selection probability, is the quantum position of the optimal arctic wolf in the t-th iteration population, represents the quantum position of a randomly selected arctic wolf in the t-th iteration population.

[0043] The present application has the following beneficial effects:

[0044] (1) The present application first extracts 11 kinds of instantaneous features and mixed features of cumulative quantity features, and uses sample expansion as modulation signal recognition feature library, which greatly improves the recognition accuracy compared with other methods based on single instantaneous feature or single cumulative quantity feature.

[0045] (2) The modulation signal recognition method of quantum arctic wolf mechanism evolution LSTM neural network designed in the present application designs a quantum arctic wolf mechanism, which can efficiently optimize the hyperparameters of LSTM neural network, and can effectively improve the modulation recognition accuracy compared with LSTM neural network with artificial selection of hyperparameters.

[0046] (3) The simulation experiment proves the effectiveness of the modulation signal recognition method of quantum arctic wolf mechanism evolution LSTM neural network under Gaussian noise, which breaks through the application limitations of traditional LSTM neural network relying on artificial selection of hyperparameters, such as dependence on expert experience and randomness. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1The LSTM neural network unit structure used in the application is shown in the schematic diagram.

[0048] Figure 2 The modulation signal recognition method flow chart of the mixed feature sample expansion combined with the quantum arctos wolf mechanism evolution LSTM neural network designed by the application is shown in the schematic diagram.

[0049] Figure 3 The eleven mixed feature sample expansion methods proposed by the application are compared with the single instantaneous feature or cumulative quantity feature recognition accuracy in the comparative literature through the designed quantum arctos wolf mechanism evolution LSTM neural network.

[0050] Figure 4 The signal recognition method of the mixed feature sample expansion combined with the quantum arctos wolf mechanism evolution LSTM neural network designed by the application is compared with the modulation signal recognition method of the LSTM neural network with artificial selection of super parameters in the comparative diagram. DETAILED DESCRIPTION

[0051] The application will be further described below in combination with the drawings.

[0052] According to Figure 1 and Figure 2 , the signal recognition method of the mixed feature sample expansion combined with the evolution LSTM network is as follows:

[0053] Step one, establish a modulation signal sample library to obtain the modulation signal data set under Gaussian noise.

[0054] The modulation signals to be identified are 2ASK, 4ASK, 2PSK, 4PSK, 2FSK, 4FSK, MSK, 8QAM and 16QAM, a total of 9 kinds of modulation signals, and are not limited to these 9 kinds of modulation modes. The symbol rate f d = 38400 baud, the sampling frequency f s = 3.072 MHz, and the number of sampling points per symbol is 80. The carrier frequency f c = 408 kHz, and the carrier frequencies of 2FSK are 306 kHz and 510 kHz, with a frequency interval of 204 kHz; the carrier frequencies of 4FSK are 255 kHz, 357 kHz, 459 kHz and 561 kHz, with a frequency interval of 102 kHz. A shaping filter is added to the MASK, MPSK, MFSK and MQAM signals at the transmitting end, and the shaping filter adopts a square root raised cosine roll-off function to shape the digital baseband signal, and the expression is In the formula, t is the sampling time, tau is the raised cosine roll-off coefficient, T sis a symbol period. The roll-off factor is set to τ = 0.4 and the duration is 4 symbol periods. The signals are transmitted through a Gaussian white noise channel with a signal-to-noise ratio ranging from -10 dB to 10 dB at an interval of 2 dB. After receiving the signals, each modulated signal is divided into a plurality of data segments with a length of 40 symbol periods .

[0055] Step two, extracting instantaneous features and cumulative features from the modulated signal data set to construct a hybrid feature library.

[0056] After receiving the signals, the receiver performs a Hilbert transform to obtain an analytic form of the signals, i.e. where s(t) is an analytic signal of the original signal y(t), j is a complex unit, and is a Hilbert transform of y(t), and where represents a convolution operation. The frequency response is

[0057] The original signal y(t) is sampled at a sampling frequency f s to obtain a discrete sequence y(n) with a total number of N points, and the analytic form is The instantaneous amplitude is A(n), and then The instantaneous phase is θ(n), and then

[0058] Since the principal value interval of the arctangent function is (-π / 2, π / 2), θ(n) can produce a sudden change of ±π. The phase θ(n) is adjusted to obtain a phase in the range [0, 2π) where

[0059] The relationship between the actual instantaneous phase ε(n) and is where mod represents a modulo operation. Therefore, there is phase folding. Since the unfolded instantaneous phase φ(n) satisfies φ(n) = 2πf c T s n + ε(n) + θ, where f c is a carrier frequency, T s is a sampling period, and θ is an initial phase. From the above formula, the unfolded instantaneous phase is a linear phase component caused by the carrier frequency and a nonlinear component caused by ε(n) and θ. A correction sequence {c(n)} is added to the sequence to achieve unfolding, which is defined as At this time, the unfolded instantaneous phase estimate is In the case of complete synchronization of the carrier and the symbol, the estimate of the nonlinear component of the unfolded instantaneous phase is

[0060] The instantaneous frequency sequence can be obtained by differentiating the deconvolved instantaneous phase sequence, that is, Where, f s is the sampling frequency.

[0061] On the basis of the instantaneous amplitude, frequency and phase of the signal, multiple characteristic quantities of the instantaneous information of the digital modulation signal are further extracted to obtain six characteristic parameters, including the maximum value of the spectral density of the zero-centered normalized instantaneous amplitude The standard deviation σ of the absolute value of the zero-centered normalized instantaneous amplitude aa , the standard deviation σ of the instantaneous phase nonlinear component of the zero-centered non-weak signal segment dp , the standard deviation σ of the absolute value of the instantaneous phase nonlinear component of the zero-centered non-weak signal segment ap , the standard deviation of the absolute value of the instantaneous frequency of the zero-centered normalized non-weak signal segment σ af , the variance of the normalized instantaneous frequency

[0062] Characteristic parameter 1. Maximum value of the spectral density of the zero-centered normalized instantaneous amplitude Where a cn (i) is the zero-centered normalized instantaneous amplitude, and a cn (i) = a n (i)-1, Where a(i) is the instantaneous amplitude of the signal, is the length of each signal segment. It reflects the changing characteristics of the instantaneous amplitude of the signal.

[0063] Characteristic parameter 2. Standard deviation σ of the absolute value of the zero-centered normalized instantaneous amplitude aa , σ aa Reflects the absolute amplitude information of the signal.

[0064] Characteristic parameter 3. Standard deviation σ of the instantaneous phase nonlinear component of the zero-centered non-weak signal segment dp , Where a t It is the set amplitude threshold, usually set to 1. is the zero-centered instantaneous phase nonlinear component, and Where: is the instantaneous phase of the signal. dp It reflects the changing characteristics of the signal's instantaneous phase.

[0065] Characteristic parameter 4. Standard deviation σ of the absolute value of the instantaneous phase nonlinear component of the zero-centered non-weak signal segment ap , σ apThe change characteristic of the instantaneous absolute phase of the signal is embodied.

[0066] Feature parameter 5. Standard deviation σ of the absolute value of the zero-centered normalized instantaneous frequency of the non-weak signal segment af , In the formula, f cn (i) is the zero-centered normalized instantaneous frequency, f cn (i) = f n (i)-1, is the normalized instantaneous frequency, and f(i) is the instantaneous frequency. σ af The change characteristic of the instantaneous absolute frequency is embodied.

[0067] Feature parameter 6. Variance of the normalized instantaneous frequency The change characteristic of the instantaneous frequency is embodied.

[0068] After the received baseband signal is obtained by down-conversion and matched filtering at the receiver, five effective cumulant features Z1, Z2, Z3, Z4 and Z5 of the signal are extracted according to the high-order cumulant theory. Let a random process x(t), the k-order cumulant of which can be defined as C kx (f1,f2,...,f k ) = Cum[x(t),x(t+f1),...,x(t+f k-1 )], in which Cum(·) is a cumulant operation, which can be represented as the calculation of high-order mixed moments. The p-order mixed moment of the random process is M pq =E{x(t) p-q x * (t) q}, in which x*(t) represents the complex conjugate of the random process, and q is the number of conjugate multiplications. The cumulants of the random process are as follows: 1. The second-order cumulant can be represented as C 20 =M 20 , C 21 =M 21 . 2. The fourth-order cumulant can be represented as: C 41 =M 41 -3M 20 M 21 . The sixth-order cumulant can be represented as: In order to reduce the influence of signal energy on the feature parameters, the five better performance feature parameters are extracted in the form of a ratio as follows:

[0069] The signal is segmented for processing, the feature parameters of each segment of the signal are extracted, and finally a mixed feature library containing 11 feature parameters is obtained.

[0070] Step 3: Preprocessing of mixed features.

[0071] The preprocessing of mixed features includes normalization, sample expansion, and division into three parts: training set, validation set, and test set. Sample expansion uses a sliding rectangular window to take three consecutive feature samples as input variables of an LSTM neural network.

[0072] Assume that the feature set composed of the extracted mixed features is M=[m1,m2,...,m J ], where J is the total number of samples, and the i-th feature sample is Where: superscript T is the transposition symbol, and superscript (i) represents the i-th sample.

[0073] The feature set is normalized using the zero mean and unit variance normalization method: Calculate the mean μ of each feature in the feature set = [μ1,μ2,...,μ 11 ] T and standard deviation σ=[σ1,σ2,...,σ 11 ] T , then the normalized feature set is Where: The dot “. / ” divides all elements in each row of the matrix by the elements in the corresponding row of the vector.

[0074] Use a rectangular window with a length of 3 and a moving step of 1 to transform the feature set Truncated to data segments, i.e. The kth sequence sample is Where: Since the sliding window length is 3, the last two samples are incomplete, so the last two incomplete samples are discarded. At this point, the feature sequence set N for LSTM network recognition is obtained, where the size of each sequence sample is 11 × 3. Subsequently, the feature sequence set N and the corresponding label T are randomly divided into a training set, a validation set, and a test set according to a certain ratio.

[0075] Step 4: Determine the objective function.

[0076] The LSTM neural network is trained using the training set and the set hyperparameters. The misjudgment rate of the LSTM network on the validation set is used as the objective function. After the training is completed, the training effect is verified by the validation set to obtain the recognition accuracy A on the validation set. c , the misjudgment rate of the LSTM network on the validation set is used as the objective function, that is, the fitness function F = 1-A c .

[0077] The LSTM network works as follows:

[0078] Figure 1 is a structure diagram of a unit of LSTM neural network, which contains three gates, and the gating mechanism mainly includes three parts of forget gate, input gate and output gate. The forget gate mainly controls what historical information is remembered or forgotten, and the information with output value close to 1 is retained and the information with output value close to 0 is forgotten. The output of the forget gate of the k-th LSTM unit is f k = sigmoid(W f ⊙x k +W f ⊙h k-1 +b f ), wherein sigmoid(x) = 1 / (1+e -x ) is a sigmoid function, limiting the output to (0, 1), W f is a weight matrix of the forget gate, x k is the input of the current LSTM unit, “⊙” represents multiplication of elements at corresponding positions of matrices or vectors, h k-1 is the output of the previous LSTM unit, and b f is a threshold matrix of the forget gate. The output of the input gate is i k = sigmoid(W i ⊙x k +W i ⊙h k-1 +b i ), wherein W i is a weight matrix of the input gate, and b i is a threshold matrix of the input gate. The input gate determines which information should be added to the memory unit. The output of the output gate is o k = sigmoid(W o ⊙x k +W o ⊙h k-1 +b o ), wherein W o is a weight matrix of the output gate, and b o is a threshold matrix of the output gate. The temporary memory state of the current LSTM unit is , wherein tanh(x) = (e x -e -x ) / (e x +e -x ) is a hyperbolic tangent function, limiting the output to -1 and 1, W c is a weight matrix of the temporary memory state, and b c is a threshold matrix. The hidden state of the current LSTM unit is updated to Finally, the output of the LSTM unit is h k =o k ⊙tanh(c k ).

[0079] In the classification task, an output layer needs to be connected after the LSTM hidden layer. Let the output of the current sample in the tth iteration be where k e {1, 2,..., K}, K is the total number of categories required by the network output. The activation function is the softmax function, so the probability of the kth category of the actual output can be obtained as According to the label y = [y1, y2,..., y K ] corresponding to the sample, the loss function for training the LSTM network is the cross entropy: where: represents the actual output of the current sample in the tth iteration, W t-1 is the network weight and threshold parameter matrix of the current iteration number, indicating that the loss function is related to the weight matrix and the expected output.

[0080] The Adam optimizer is used to update the weights of the LSTM neural network. Let the current gradient be where: represents the partial derivative of the loss function with respect to the weight parameter W. The biased first moment estimate can be updated as u t = β1·u t-1 +(1-β1)·g t , where the exponential decay rate is defined as β1, which mainly controls the update weight of the first moment estimate and the gradient. The value is generally 0.9. The biased second moment estimate can be updated as v t = β2·v t-1 +(1-β2)·(g t ⊙g t ), where: β2 is the exponential decay rate, mainly used to control the update weight of the second moment estimate and the gradient square. The default value is 0.999. The first moment estimate is corrected for bias as where: is the tth power of β1. The second moment estimate is corrected for bias as where: is the tth power of β2. Therefore, the weight parameter update in the tth iteration is where: η is the learning rate, μ is a minimum constant to ensure that the denominator is not zero, and the default value is 10 -8 .

[0081] Step five, initialize the quantum polar bear mechanism parameters.

[0082] Including the number of individuals N q in the population, the maximum number of iterations G max , the search space dimension M, and the upper bound U = [u1, u2,..., u M ] and the lower bound L = [l1, l2,..., l M, group behavior probability P, hunting attack probability Q in group behavior, persecution attack probability 1-Q in group behavior, update coefficient w1, mutation probability Selection probability Randomly initialize the quantum position of the arctic wolf between [0, 1] And map to the corresponding position Where i∈{1,2,...,N q}, the symbol "⊙" represents the multiplication of elements in the corresponding position of the matrix or vector, and the superscript 0 represents the iteration number 0.

[0083] Step six, map each dimension of the position of the i-th arctic wolf in the population to the hyperparameters of the LSTM network, i∈{1,2,...,N q}, train the LSTM network with the training set and the corresponding hyperparameters, and use the misclassification rate of the LSTM network on the validation set as the fitness function of the i-th arctic wolf.

[0084] Let the LSTM network hyperparameters to be optimized include the number of hidden layer neurons N h , the batch size S b , the number of training rounds I m , the learning rate l r , and the dropout probability p d . Take the logarithm of the learning rate to map it to the variable ω4=lg(l r ). Let the row vector composed of all hyperparameters be h=[N h ,S b ,I m ,ω4,p d ]=[h1,h2,h3,h4,h5], and let its upper bound be and the lower bound be Linearly map the upper bound and the lower bound to the upper bound U and the lower bound L of the search space of the quantum arctic wolf mechanism, that is, Where Map the mapped vector to the position of the i-th arctic wolf in the t-th iteration of the quantum arctic wolf mechanism , and then inverse transform it in the fitness function to obtain the corresponding LSTM network hyperparameters, that is, Where: i∈{1,2,...,N q}, N q is the number of individuals in the population, and d∈{1,2,...,M} is the d-th hyperparameter, with a dimension of M=5. The actual hyperparameters used for training are: the number of hidden layer neurons , the batch size , and the number of training rounds Where: round represents rounding to the nearest integer. The learning rate is Discarding probability

[0085] The LSTM network is trained with the corresponding hyperparameters of the ith arctic wolf in the tth iteration, and the accuracy on the validation set is The fitness function is Then the corresponding survival probability is calculated That is Let the iteration number t = 0.

[0086] Step seven, enter the iteration loop, let t = t + 1, for the ith arctic wolf in the population, calculate the value of the variable m1(t) related to the iteration number m1(t) = exp(t / G max -1), compared with the value of P, if less than P, enter the group behavior, with probability Q, the hunting attack strategy update quantum rotation angle is Where: r i t Is the ith arctic wolf tth iteration of the normal distribution random number with mean 0 and variance 1, n a Is a random integer between [2, N q / 2] represents the number of arctic wolves for updating, Represents the kth arctic wolf for group attack in the tth iteration, is composed of n a Different arctic wolves randomly selected from the population in each iteration, Is the quantum position of the optimal arctic wolf, c(t) = cos(πt / 2G max ) is the step size coefficient gradually changing from 1 to 0 with iteration; with probability 1-Q, the persecution attack strategy update quantum rotation angle is And define the update step Where: Is the ith arctic wolf in the tth iteration, which satisfies the uniform distribution between (0, 2), And are the tth iteration, which satisfy the uniform distribution between (0, 2π) and (0, π) respectively, w1 is the update coefficient, v t (m) is the average value of the quantum position of all arctic wolves in the tth iteration population Otherwise, m1(t) is greater than or equal to P, and the carrion behavior update quantum rotation angle is Where: Is the quantum position of a randomly selected arctic wolf in the tth iteration of carrion behavior, And Respectively represent the ith arctic wolf in the tth iteration, which satisfies the uniform distribution between [-1, 1], p r Is a random number uniformly distributed between [0, 1], is the set selection probability, is the quantum position of the optimal arctic wolf in the tth iteration population, represents the quantum position of a randomly selected arctic wolf in the tth iteration population.

[0087] Step eight, for the ith arctic wolf in the tth iteration, if the survival probability then the survival behavior is performed, and the quantum rotation angle is updated as where m2(t) = exp(1-t / G max -t / G max , in which: indicates a random number uniformly distributed between (0, 1) of the ith arctic wolf in the tth iteration; otherwise, go to step nine.

[0088] Step nine, update the quantum position of each arctic wolf according to the quantum rotation angle and the quantum gate strategy as where: represents the quantum position of the ith arctic wolf in the tth iteration, is the corresponding updated quantum rotation angle, is a random number uniformly distributed between [0, 1], is the mutation probability of the quantum position when the quantum rotation angle is 0.

[0089] Step ten, calculate the position of the ith arctic wolf in the new generation as convert the elements of each dimension into the corresponding LSTM network hyperparameters, and use the misjudgment rate on the validation set after training the LSTM network as the corresponding fitness function

[0090] Step eleven, adopt the reservation strategy to reserve the updated arctic wolf: if the updated arctic wolf has a better fitness, then reserve the corresponding position, quantum position, and fitness after updating, otherwise do not reserve. Update the quantum position of the optimal individual and the position Update the survival probability of each arctic wolf is

[0091] Step twelve, determine whether the iteration number t reaches the maximum iteration number G max , if t = G max , then output the optimal arctic wolf position as the optimization result, otherwise return to step seven to continue iteration.

[0092] Step thirteen, convert The elements of each dimension are converted into the corresponding optimal LSTM network hyperparameters, and the LSTM neural network with the optimal hyperparameters is trained. After the training is completed, the modulated signal recognition performance of the LSTM network is tested using the test set, and the results are analyzed.

[0093] exist Figure 3 In the present invention, the hybrid feature sample expansion method proposed in the present invention and the LSTM neural network evolution method through the designed quantum Arctic wolf mechanism are recorded as MF-QAW-LSTM, the three instantaneous features described in the "Modulation recognition method based on instantaneous feature parameters" published by Liu Sheng et al. in "Communication Technology" (2022, 55(05): 559-564) and the modulation signal recognition method combined with the three-layer fully connected neural network are recorded as IF-TFCNN, and the five high-order cumulative quantity features described in the "Digital modulation recognition based on high-order cumulative quantity" published by Zhang Wenna et al. in "Foreign Electronic Measurement Technology" (2022, 41(07): 126-132) and the modulation signal recognition method combined with the decision tree are recorded as CF-DT.

[0094] exist Figure 4 In the present invention, the hybrid feature sample expansion method proposed and the LSTM neural network evolution method through the designed quantum Arctic wolf mechanism are recorded as MF-QAW-LSTM, and the LSTM neural network modulation signal recognition method with artificially selected hyperparameters is recorded as MS-LSTM.

[0095] The simulation experiment parameters of the modulation signal recognition method based on hybrid feature sample expansion and combined with the quantum Arctic wolf mechanism evolution LSTM neural network are set as follows:

[0096] The modulation signals to be identified are 9 modulation signals, including 2ASK, 4ASK, 2PSK, 4PSK, 2FSK, 4FSK, MSK, 8QAM and 16QAM, and are not limited to these 9 modulation methods. d =38400baud, sampling frequency f s =3.072MHz, the number of sampling points per symbol is 80. Carrier frequency f c = 408kHz. The carrier frequencies for 2FSK are 306kHz and 510kHz, with a frequency spacing of 204kHz. The carrier frequencies for 4FSK are 255kHz, 357kHz, 459kHz, and 561kHz, with a frequency spacing of 102kHz. The modulated signal is filtered using a root-raised cosine shaping filter with a roll-off factor of 0.4 and a duration of four symbol periods. The signal-to-noise ratio range is -10dB to 10dB, with a 2dB interval. Each modulated signal is divided into multiple data segments of 40 symbols in length, totaling 3200 points.

[0097] Six instantaneous features and five cumulative features were extracted. Each signal at each signal-to-noise ratio contained 502 feature samples to form the original feature set. These samples were normalized to zero mean and unit variance, and a sliding rectangular window of length 3 was used to obtain 500 feature sequence samples for each signal at each signal-to-noise ratio. The training, validation, and test sets were randomly divided in a 3:1:1 ratio, resulting in a total of 29,700 samples in the training set, and 9,900 samples each in the validation and test sets.

[0098] The LSTM network hyperparameter ranges for the quantum Arctic Wolf mechanism optimization are set as follows: the number of hidden layer neurons, the mini-batch size, the number of training rounds, the logarithm of the learning rate, and the upper bound of the dropout probability are set as The corresponding lower bound is

[0099] The parameters of the quantum Arctic wolf mechanism are set as follows: group behavior probability P = 0.5, hunting attack probability Q = 0.7 in group behavior, persecution attack probability 1-Q = 0.3 in group behavior, dimension M = 5, search space upper bound U = [1,1,1,1,1], lower bound L = [-1,-1,-1,-1,-1], persecution attack update coefficient u1 = 0.74, mutation probability when quantum rotation angle is 0 Selection probability The number of Arctic wolves in the population N q =20, maximum number of iterations G max =20.

[0100] from Figure 3 It can be seen that compared with the comparative methods IF-TFCNN and CF-DT, the MF-QAW-LSTM method proposed in the present invention extracts richer signal features to form 11 mixed feature libraries, and expands samples through a sliding rectangular window. The expanded samples containing time series information are used to train the LSTM neural network, and the quantum Arctic wolf mechanism is proposed to optimize the five key hyperparameters of the LSTM neural network. Therefore, the recognition accuracy is greatly improved, which expands the application limitations of the traditional modulation signal recognition method based on the LSTM neural network under low signal-to-noise ratio.

[0101] from Figure 4 It can be seen that the MF-QAW-LSTM method proposed in the present invention is able to automatically optimize the LSTM neural network hyperparameters compared with the LSTM neural network method MS-LSTM in which hyperparameters are manually selected, avoiding the problem of requiring expert experience and difficulty in obtaining the optimal hyperparameters when manually determining the hyperparameters. The recognition accuracy is greatly improved under low signal-to-noise ratios, expanding the application limitations of traditional modulation signal recognition methods based on LSTM neural networks under low signal-to-noise ratios.

[0102] The above merely provides the preferred embodiments of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the principles and technical scope of the present application shall fall into the scope of the present application.

Claims

1. A signal recognition method combining hybrid feature sample expansion with an evolved LSTM network, characterized by: Here are the steps: Step 1: Establish a modulation signal sample library to obtain a modulation signal data set; Step 2: Extract instantaneous features and cumulative features from the modulated signal data set obtained in step 1 to construct a hybrid feature library; Step 3: Preprocess the mixed features extracted in step 2; Step 4: Determine the objective function; The LSTM neural network is trained using the training set and the set hyperparameters, and the error rate of the LSTM network on the validation set is used as the objective function, that is, the fitness function F = 1-A c , where A c is the recognition accuracy on the validation set. After the training is completed, the validation set is used to verify the training effect. Step 5: Initialize the parameters of the quantum Arctic wolf mechanism; Including the number of individuals in the population N q , maximum number of iterations G max , the search space dimension M and the upper bound U=[u1,u2,...,u M ] and the lower bound L=[l1,l2,…,l M ], group behavior probability P, hunting attack probability Q in group behavior, persecution attack probability 1-Q in group behavior, update coefficient w1, mutation probability Selection probability Randomly initialize the quantum position of the Arctic wolf between [0,1] And mapped to the corresponding position where i∈{1,2,…,N q }, the symbol "⊙" represents the multiplication of elements at corresponding positions in the matrix or vector, and the superscript 0 represents the number of iterations is 0; Step 6: Map the elements of each dimension of the position of the i-th Arctic wolf in the population to the hyperparameters of the LSTM network, i∈{1,2,…,N q }, use the training set and corresponding hyperparameters to train the LSTM network, and use the LSTM network's misclassification rate on the validation set as the fitness function of the i-th Arctic wolf; Step 7: Enter the iteration loop, let t = t + 1, for the i-th Arctic wolf in the population, calculate the value of the variable m1(t) related to the number of iterations m1(t) = exp(t / G max -1), compared with the value of P, if it is less than P, it will enter the group behavior; if m1(t) is greater than or equal to P, it will perform scavenging behavior; Step 8: For the i-th Arctic wolf in the t-th iteration, if the survival probability Then perform survival behavior and update the quantum rotation angle to Where m2(t) = exp(1-t / G max )-t / G max , where: Indicates the random number uniformly distributed between (0,1) for the i-th Arctic wolf in the t-th iteration; otherwise, proceed to step 9; Step 9: Update the quantum position of each Arctic wolf according to the quantum rotation angle and quantum gate strategy: Where: represents the dth dimension of the quantum position of the tth iteration of the i-th Arctic wolf, is the corresponding updated quantum rotation angle, is a random number uniformly distributed between [0,1]. is the probability of mutation of quantum position when the quantum rotation angle is 0; Step 10: Calculate the position of the new generation i-th Arctic wolf as Convert the elements of each dimension into the corresponding LSTM network hyperparameters. After training the LSTM network, use the error rate on the validation set as the corresponding fitness function. Step 11: Use the retention strategy to retain the updated Arctic wolf: if the updated Arctic wolf has better fitness, retain the corresponding updated position, quantum position and fitness; otherwise, do not retain; update the quantum position corresponding to the optimal individual and location Update the survival probability of each Arctic wolf for Step 12: Determine whether the number of iterations t reaches the maximum number of iterations G max , if t=G max , then output the optimal Arctic wolf position As the optimization result, otherwise return to step 7 to continue iteration; Step 13: The elements of each dimension are converted into the corresponding optimal LSTM network hyperparameters, and the LSTM neural network with the optimal hyperparameters is trained. After the training is completed, the modulated signal recognition performance of the LSTM network is tested using the test set, and the results are analyzed.

2. The signal recognition method of a hybrid feature sample expansion combined with an evolved LSTM network according to claim 1, characterized in that: The digital modulation signals in step 1 include MASK, MPSK, MFSK, MSK and MQAM. To reduce the impact of inter-symbol interference, a shaping filter is added to the MASK, MPSK, MFSK and MQAM signals at the transmitting end. The shaping filter uses a square root raised cosine roll-off function to shape the digital baseband signal. The signal is received at the receiving end through a Gaussian white noise channel to obtain a modulated signal data set.

3. The signal recognition method of a hybrid feature sample expansion combined with an evolved LSTM network according to claim 1, characterized in that: The step 2 is specifically as follows: After the receiver receives the signal, it performs Hilbert transform on it to obtain its analytical form, that is, Where s(t) is the analytical signal of the original signal y(t), j is a complex unit, and is the Hilbert transform of y(t), we have Where, represents the convolution operation; Its frequency response is With sampling frequency f s Sampling the original signal y(t), the total number of points is The discrete sequence y(n) of The instantaneous amplitude is A(n), then The instantaneous phase is θ(n), then Since the main value interval of the inverse tangent function is (-π / 2,π / 2), θ(n) may produce a sudden change of ±π at this time, and the phase is adjusted to obtain a value in [0,2π) have The actual instantaneous phase ε(n) is related to The relationship is Where mod represents the remainder operation; therefore, There is phase wrapping; since the dewrapping instantaneous phase φ(n) satisfies φ(n)=2πf c T s n+ε(n)+θ, where f c is the carrier frequency, T s Sampling is the period, θ is the initial phase; From the above formula, we know that the de-wrapping instantaneous phase is composed of the linear phase component caused by the carrier frequency and the nonlinear component caused by ε(n) and θ; it is necessary to Add a correction sequence {c(n)} to achieve deconvolution, defined as At this time, the deconvolution instantaneous phase estimate is When the carrier and code elements are completely synchronized, the estimated value of the de-warping instantaneous phase nonlinear component is The instantaneous frequency sequence is obtained by differentiating the deconvolved instantaneous phase sequence, that is, Where, f s is the sampling frequency; On the basis of the instantaneous amplitude, frequency and phase of the signal, multiple characteristic quantities of the instantaneous information of the digital modulation signal are further extracted to obtain six characteristic parameters, including the maximum value of the spectral density of the zero-centered normalized instantaneous amplitude The standard deviation σ of the absolute value of the zero-centered normalized instantaneous amplitude aa , the standard deviation σ of the instantaneous phase nonlinear component of the zero-centered non-weak signal segment dp , the standard deviation σ of the absolute value of the instantaneous phase nonlinear component of the zero-centered non-weak signal segment ap , the standard deviation of the absolute value of the instantaneous frequency of the zero-centered normalized non-weak signal segment σ af , the variance of the normalized instantaneous frequency After the receiver obtains the received baseband signal through down-conversion and matched filtering, the five effective cumulant features Z1, Z2, Z3, Z4 and Z5 of the signal are extracted according to the high-order cumulant theory. Suppose a random process x(t), whose k-order cumulant is defined as C kx (f1,f2,…,f k )=Cum[x(t),x(t+f1),…,x(t+f k-1 )], where: Cum(·) is the cumulative operation, which is expressed as the calculation of high-order mixing moments; the p-order mixing moment of the random process is M pq =E{x(t) p-q x * (t) q }, where x*(t) represents the complex conjugate of the random process, and q is the number of conjugate multiplications; the cumulants of each order of the random process are as follows:

1. The second-order cumulant is expressed as: C 20 =M 20 , C 21 =M 21 ; 2. The fourth-order cumulant is expressed as: C 41 =M 41 -3M 20 M 21 ;3. The sixth-order cumulant is expressed as: In order to reduce the influence of signal energy on characteristic parameters, five characteristic parameters with better performance are extracted in the form of ratios as follows: The signal is processed in segments, and the characteristic parameters of each segment are extracted, and finally a mixed feature library containing 11 characteristic parameters is obtained.

4. The signal recognition method of a hybrid feature sample expansion combined with an evolved LSTM network according to claim 1, characterized in that: The preprocessing of the mixed features in step 3 includes normalization, sample expansion, and division into training set, validation set, and test set. The sample expansion uses a sliding rectangular window to treat three consecutive feature samples as input variables of an LSTM neural network. Assume that the feature set composed of the extracted mixed features is M=[m1,m2,…,m J ], where J is the total number of samples, and the i-th feature sample is Where: superscript T is the transposition symbol, superscript (i) represents the i-th sample; The feature set is normalized using the zero mean and unit variance normalization method: the mean μ of each feature in the feature set is calculated as [μ1, μ2, …, μ 11 ] T and standard deviation σ=[σ1,σ2,…,σ 11 ] T , then the normalized feature set is Where: ". / " means dividing all elements in each row of the matrix by the elements in the corresponding row of the vector; Use a rectangular window with a length of 3 and a moving step of 1 to transform the feature set Truncated to data segments, i.e. The kth sequence sample is Where: Since the sliding window length is 3, the last two samples are incomplete, so the last two incomplete samples are discarded. That is, a feature sequence set N for LSTM network recognition is obtained, where the size of each sequence sample is 11×3; then, the feature sequence set N and the corresponding label T are randomly divided into training set, validation set and test set according to a certain ratio.

5. The signal recognition method of a hybrid feature sample expansion combined with an evolved LSTM network according to claim 1, characterized in that: In step 6, the LSTM network hyperparameters to be optimized include the number of hidden layer neurons N h , small batch size S b , number of training rounds I m , learning rate l r and the drop probability p d , the learning rate is mapped to the logarithm of the variable ω4=lg(l r ); Let the row vector of all hyperparameters be h=[N h ,S b ,I m ,ω4,p d ]=[h1,h2,h3,h4,h5], let its upper bound be and the lower bound is Through the upper and lower bounds, a linear mapping is performed to the upper bound U and lower bound L of the search space of the quantum Arctic wolf mechanism, that is, in After mapping, the vector As the position of the i-th Arctic wolf in the t-th iteration of the quantum Arctic wolf mechanism In the fitness function, the inverse transformation is performed to obtain the corresponding LSTM network hyperparameters, namely Where: i∈{1,2,…,N q }, N q is the number of individuals in the population, d∈{1,2,…,M} is the dth hyperparameter, and the independent variable dimension is M=5; then the actual hyperparameters used for training are: the number of neurons in the hidden layer Small batch size Number of training rounds In the formula: round represents rounding off decimals; learning rate Drop probability The LSTM network is trained using the hyperparameters corresponding to the ith Arctic wolf in the tth iteration. The accuracy on the validation set is Then its fitness function is Then calculate the corresponding survival probability Right now Assume the number of iterations t=0.

6. The signal recognition method of a hybrid feature sample expansion combined with an evolved LSTM network according to claim 1, characterized in that: In step 7, for the i-th Arctic wolf in the population, if m1(t)=exp(t / G max -1)<P, then group behavior is initiated, which includes two strategies: hunting attack and persecution attack. The hunting attack strategy is performed with probability Q and the quantum rotation angle is updated as Where: r i t is a normally distributed random number with mean 0 and variance 1 for the tth iteration of the i-th Arctic wolf, n a Is in [2,N q / 2] represents the number of Arctic wolves to be updated. The quantum position of the kth Arctic wolf that attacks the group in the tth iteration is n randomly selected from the population in each iteration. a Made up of different Arctic wolves, is the quantum position of the optimal Arctic wolf, c(t)=cos(πt / 2G max ) is the step size coefficient that gradually changes from 1 to 0 as the iteration proceeds; the persecution attack strategy updates the quantum rotation angle with probability 1-Q And define the update step size Where: The i-th Arctic wolf in the t-th iteration satisfies the uniformly distributed random number between (0,2). They are random numbers uniformly distributed between (0,2π) and (0,π) at the tth iteration, w1 is the update coefficient, v t (m) is the average value of the corresponding dimension of all Arctic wolf quantum positions in the t-th iteration population Otherwise m1(t)≥P performs scavenging behavior and updates the quantum rotation angle to Where: is the quantum position of a randomly selected Arctic wolf in the t-th iteration of scavenging behavior, and They represent the random numbers uniformly distributed between [-1,1] for the i-th Arctic wolf in the t-th iteration, p r is a random number uniformly distributed between [0,1]. is the set selection probability, is the quantum position of the optimal Arctic wolf in the t-th iteration population, Represents the quantum position of a randomly selected Arctic wolf from the population in the t-th iteration.

Images