A method for generating a frequency-agile radar signal

Abstract

This invention relates to a method for generating frequency-agile radar signals, belonging to the field of radar frequency-hopping signal technology. The invention includes S1, quantizing an initial chaotic sequence into an integer chaotic sequence; and S2, determining the optimal control parameter u of the integer chaotic sequence using a simulated annealing algorithm. b S3, Generate the target chaotic sequence; S4, Construct the first frequency-agile radar sequence based on the target chaotic sequence; S5, Establish the data training set Ψ; S6, Train the CNN network model using the data training set Ψ, and use the CNN network to identify the given chaotic sequence and find the optimal threshold λ. k S6, perform binary quantization on the given chaotic sequence to determine the second frequency-agile radar sequence; S7, construct the target frequency-agile radar sequence. This invention utilizes two optimization methods—simulated annealing and convolutional neural networks—to enhance the randomness of the target frequency-agile radar sequence, enabling it to solve the problems of traditional radar systems, such as the ease of signal detection and interception, and susceptibility to interference.

Description

Technical Field

[0001] This invention belongs to the field of radar frequency hopping signal technology, specifically relating to a method for generating frequency-agile radar signals. Background Technology

[0002] With the continuous development of radar jamming technology, the electromagnetic warfare environment in which radar operates is becoming increasingly complex, which places higher demands on the radar's high-precision measurement capabilities and sufficient anti-jamming capabilities. Traditional radar systems suffer from problems such as simple signal modulation methods, single signal generation methods, and fixed signal processing methods. This results in traditional radar systems having disadvantages such as easy signal detection and interception, greater susceptibility to jamming, and greater impact from jamming.

[0003] Currently, radar waveform modulation signals are developing rapidly and have gradually become a research hotspot. As a crucial application of frequency domain countermeasures in waveform modulation technology, improving the operational capabilities of frequency-agile radar in complex electronic environments has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to address the problems of traditional radar systems, such as the ease with which signals can be detected and intercepted, and their susceptibility to interference and the significant impact of interference, by providing a method for generating frequency-agile radar signals, which can improve the working capability of frequency-agile radar in complex electronic environments.

[0005] A method for generating frequency-agile radar signals includes the following steps:

[0006] S1, quantize the initial chaotic sequence of length N into an integer chaotic sequence;

[0007] S2, using simulated annealing algorithm to determine the optimal control parameters u for the integer chaotic sequence. b And based on the optimal control parameter u b Generate a target chaotic sequence of length N;

[0008] S3, construct the first frequency-agile radar sequence based on the target chaotic sequence;

[0009] S4. Based on the given chaotic sequence length M and the set initial value interval ε of the chaotic sequence, establish a data training set Ψ; this data training set Ψ is used to find the optimal threshold λ for the given chaotic sequence when performing binary quantization processing. k ;

[0010] S5, using the training set Ψ, train the CNN network model to obtain the CNN network; use this CNN network to identify a given chaotic sequence of length N and find the optimal threshold λ. k ;

[0011] S6, based on the optimal threshold λ k The given chaotic sequence is subjected to binary quantization to obtain a binary sequence; the second frequency-agile radar sequence is obtained based on the binary sequence.

[0012] S7, construct a target frequency agile radar sequence based on the first frequency agile radar sequence and the second frequency agile radar sequence.

[0013] Furthermore, step S1 includes the following steps:

[0014] S11, Generate the initial chaotic sequence based on the given initial value of the chaotic sequence and the length N of the target chaotic sequence;

[0015] S12, quantize the initial chaotic sequence into the integer chaotic sequence.

[0016] Furthermore, step S2 specifically includes the following steps:

[0017] S21, Construct the autocorrelation function with the minimum main lobe-side lobe ratio for this integer chaotic sequence;

[0018] S22, invert the autocorrelation function and determine the maximum main lobe-side lobe ratio function of the integer chaotic sequence;

[0019] S23, the simulated annealing algorithm is used to iteratively optimize the maximum main lobe-to-side lobe ratio function to obtain the optimal control parameter u for the integer chaotic sequence. b ;

[0020] S24, based on the optimal control parameter u b Generate a target chaotic sequence of length N.

[0021] Furthermore, step S2 is followed by S0.

[0022] S0: Using the target chaotic sequence generated in step S2 as the initial value, generate the next chaotic sequence of length N. Repeat steps S1 to S2 on the next chaotic sequence to obtain the next target chaotic sequence.

[0023] Furthermore, the data training set Ψ is established through the following steps:

[0024] S41, Based on the given chaotic sequence length M and the set initial value interval ε of the chaotic sequence, generate... A basic chaotic sequence of length M;

[0025] S42, for each basic chaotic sequence, use x thresholds (λ) 01 ,λ 02 ,...,λ 0i ...,λ 0xBinary quantization is performed to obtain x basic binary sequences of length M. By converting each P binary number into a decimal number, all generated basic binary sequences are converted into basic decimal sequences, resulting in x basic frequency-agile radar sequences of length L. P < M, i ∈ (1, x);

[0026] S43, autocorrelation calculations are performed on the x basic frequency-agile radar sequences generated in each basic chaotic sequence to obtain the basic frequency-agile radar sequence with the best randomness in the basic chaotic sequence and the preferred threshold corresponding to the best basic frequency-agile radar sequence.

[0027] S44, construct a data training set Ψ by combining all basic chaotic sequences with the preferred threshold corresponding to the basic chaotic sequence.

[0028] Furthermore, constructing the data training set Ψ also includes the following steps:

[0029] Step 1: Repeat steps S42 to S43 to determine the range [a, b] of all preferred threshold values;

[0030] Step 2: Divide the interval [a, b] into T preferred thresholds and obtain a set of preferred thresholds.

[0031] Step 3: Using the methods described in steps S42 and S43, select the optimal threshold from the T optimal thresholds for each basic chaotic sequence; and combine all basic chaotic sequences with the optimal thresholds corresponding to the basic chaotic sequence to form the data training set Ψ.

[0032] Furthermore, when using a CNN network to identify a given bastard sequence, the initialization method is chosen to be the Glorot initialization method of Keras with uniform distribution characteristics, the optimizer is chosen to be the Adam optimizer, the network learning rate is 0.001, and the network loss function is the cross-entropy loss function.

[0033] Furthermore, when using a CNN network to identify a given sequence of jerks, K-fold cross-validation is used to improve data utilization.

[0034] Furthermore, the length D of the target frequency-agile radar sequence is

[0035] D = L + Q × N

[0036] Where Q is the number of target chaotic sequences.

[0037] Furthermore, M = P × N.

[0038] The beneficial effects of this invention are as follows:

[0039] Frequency-hopping radar signals are radar signals whose carrier frequency changes continuously over time. Frequency-hopping signals can counteract active interference. Therefore, the better the randomness of a frequency-agile radar signal, the stronger its anti-jamming capability.

[0040] Simulated annealing is an optimization algorithm that simulates the annealing process of solid materials. It is a stochastic optimization algorithm based on the Monte Carlo iterative solution strategy. Simulated annealing has the advantages of fast optimization speed, strong robustness, and low susceptibility to getting trapped in local optima. It can be combined with integer quantization.

[0041] CNNs, or Convolutional Neural Networks, are a type of deep feedforward neural network that incorporates convolutional computations. They possess powerful feature parameter extraction capabilities and are one of the representative algorithms of deep learning. This invention creatively combines CNNs with binary quantization methods to determine the optimal threshold λ for binary quantization of the target chaotic sequence. k It can map the chaotic sequence of a target into a frequency-agile radar sequence with good random performance, and has good real-time performance and randomness, so as to reduce the probability of being intercepted by jammers.

[0042] In the frequency agile radar signal generation method provided by the present invention, the frequency agile radar sequence is divided into two parts. One part uses a simulated annealing algorithm to enhance the randomness of the part, and the other part uses a convolutional neural network method to enhance the randomness of the part. Then the two parts are spliced ​​together to form the target frequency agile radar sequence.

[0043] In other words, this invention enhances the randomness of target frequency-agile radar sequences by utilizing two optimization methods: simulated annealing algorithm and convolutional neural network method. This enables the invention to solve the problems of traditional radar systems, such as the ease with which signals can be detected and intercepted, and the susceptibility to interference and the significant impact of interference.

[0044] Therefore, the frequency agile radar signal generation method provided by this invention can improve the working capability of frequency agile radar in complex electronic environments; and the radar frequency agile sequence designed by this invention can improve anti-interference capability while generating a normal target detection capability. Attached Figure Description

[0045] Figure 1 This is a schematic diagram illustrating the generation of the first frequency-agile radar sequence in this embodiment.

[0046] Figure 2 This is a schematic diagram illustrating the generation of the second frequency-agile radar sequence in this embodiment.

[0047] Figure 3 This is a schematic diagram of the CNN training process in this embodiment. Detailed Implementation

[0048] The present invention will be further described in detail below with reference to experimental examples and specific embodiments. However, this should not be construed as limiting the scope of the above-mentioned subject matter of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0049] To address the issues of traditional radar systems having relatively easy signals to be detected and intercepted, and being susceptible to interference and significant interference effects, this embodiment provides a method for generating frequency-agile radar signals, which can improve the working capability of frequency-agile radar in complex electronic environments.

[0050] like Figures 1 to 3 As shown, this embodiment includes the following steps:

[0051] S1 quantizes the initial chaotic sequence of length N into an integer chaotic sequence.

[0052] Specifically, this embodiment first generates an initial chaotic sequence, and then quantizes this initial chaotic sequence into an integer chaotic sequence. In this embodiment, the initial chaotic sequence is generated based on a given initial value of the chaotic sequence and the length N of the target chaotic sequence to be generated. The chaotic sequence expression is:

[0053]

[0054] S1 in this embodiment 1 specifically includes the following steps.

[0055] S11, Generate an initial chaotic sequence based on the given initial value of the chaotic sequence and the length N of the target chaotic sequence;

[0056] S12, quantize the initial chaotic sequence into an integer chaotic sequence.

[0057] According to the expression for chaotic sequences, the control parameter u of the chaotic sequence is... b This will affect the chaotic performance of the chaotic sequence, and thus can be addressed by changing the control parameter u. b To achieve better random performance for chaotic sequences.

[0058] S2, using simulated annealing algorithm to determine the optimal control parameters u for the integer chaotic sequence. b And based on the optimal control parameter u b Generate a target chaotic sequence of length N.

[0059] To use simulated annealing to control the parameters u of a chaotic sequence bTo find the optimal value, we first need to construct the objective function. Simulated annealing can find the global minimum of the objective function within its domain. To find the global maximum of the objective function within its domain using simulated annealing, we need to invert the objective function.

[0060] To optimize the random performance of the frequency-agile radar sequence, this embodiment requires constructing the maximum main lobe-to-side lobe ratio function of the integer chaotic sequence in step S1, and then using a simulated annealing algorithm to adjust the control parameter u. b Optimization. It should be noted that after designing the objective function, those skilled in the art should define the initial value and upper and lower limits of the control parameter ub, and specify the range of variation of the control parameter ub.

[0061] Specifically, S2 in this embodiment includes the following steps:

[0062] S21, Construct the autocorrelation function with the minimum main lobe-side lobe ratio for this integer chaotic sequence;

[0063] S22, invert the autocorrelation function and determine the maximum main lobe-side lobe ratio function of the integer chaotic sequence;

[0064] S23, the simulated annealing algorithm is used to iteratively optimize the maximum main lobe-to-side lobe ratio function to obtain the optimal control parameter u for the integer chaotic sequence. b ;

[0065] S24, based on the optimal control parameter u b Generate a target chaotic sequence of length N.

[0066] S3. After generating the target chaotic sequence, this embodiment constructs the first frequency-agile radar sequence based on the target chaotic sequence.

[0067] It should be noted that the length of the initial chaotic sequence generated in this embodiment is N, the length of the integer chaotic sequence is N, the length of the target chaotic sequence generated through step S24 is N, and the length of the first frequency-agile radar sequence directly generated through step S24 is N.

[0068] When implementing this embodiment, if a first frequency-agile radar sequence with a length of 2N, 3N, 4N... is to be generated, the target chaotic sequence generated in step S24 can be used as the initial value and steps S1 and S2 can be repeated to obtain a new target chaotic sequence. The newly obtained target chaotic sequence and the previously generated target chaotic sequence are then used together to construct the first frequency-agile radar sequence.

[0069] Furthermore, this embodiment also includes S0 after step S2.

[0070] S0: Using the target chaotic sequence generated in step S2 as the initial value, generate the next chaotic sequence of length N. Repeat steps S1 to S2 on the next chaotic sequence to obtain the next target chaotic sequence.

[0071] It should be noted that those skilled in the art should perform the cyclic process in S0 as needed. That is, if a first frequency-agile radar sequence of length N is required, the cyclic process does not need to be repeated; if a first frequency-agile radar sequence of length 2N is required, step S0 can be repeated once, and the newly obtained target chaotic sequence can be concatenated with the previously obtained target chaotic sequence to form the first frequency-agile radar sequence; if a first frequency-agile radar sequence of length 3N is required, step S0 can be repeated twice, and the latest obtained target chaotic sequence can be concatenated with the two previously obtained target chaotic sequences to form the first frequency-agile radar sequence, and so on.

[0072] S4. Based on the given chaotic sequence length M and the set initial value interval ε of the chaotic sequence, establish a data training set Ψ; this data training set Ψ is used to find the optimal threshold λ for the given chaotic sequence when performing binary quantization processing. k .

[0073] Specifically, in this embodiment, the data training set Ψ is established through the following steps:

[0074] S41, Based on the given chaotic sequence length M and the set initial value interval ε of the chaotic sequence, generate... There are M basic chaotic sequences. Since the initial values ​​of all chaotic sequences are in the range (-0.5, 0.5), the interval length of the chaotic sequence is 1.

[0075] S42, for each basic chaotic sequence, use x thresholds (λ) 01 ,λ 02 ,...,λ 0i ...,λ 0x Binary quantization is performed to obtain x basic binary sequences of length M; by converting each P binary number into a decimal number, all the generated basic binary sequences are converted into basic decimal sequences, resulting in x basic frequency-agile radar sequences of length L. P < M, i ∈ (1, x); It should be noted that L should be an integer, and the range of values ​​for the frequency-agile radar sequence is [0, 2]. P -1].

[0076] S43, autocorrelation calculations are performed on x basic frequency-agile radar sequences generated from each basic chaotic sequence to obtain the basic frequency-agile radar sequence with the best randomness in the basic chaotic sequence and the optimal threshold corresponding to the optimal basic frequency-agile radar sequence.

[0077] S44, construct a data training set Ψ by combining all basic chaotic sequences with the preferred threshold corresponding to the basic chaotic sequence.

[0078] To more accurately obtain the required threshold and map the target chaotic sequence to a frequency-agile radar sequence with good random performance, this embodiment further includes the following steps in constructing the data training set Ψ.

[0079] Step 1: Repeat steps S42 to S43 to determine the range [a, b] of all preferred threshold values;

[0080] Step 2: Divide the interval [a, b] into equal parts and obtain a set of preferred thresholds consisting of T preferred thresholds;

[0081] Step 3: Using the methods described in steps S42 and S43, select the optimal threshold from the T optimal thresholds for each basic chaotic sequence; and combine all basic chaotic sequences with the optimal thresholds corresponding to the basic chaotic sequence to form the data training set Ψ.

[0082] Specifically, each basic chaotic sequence is binary quantized using T optimal thresholds to obtain T binary sequences of length N. All generated binary sequences are then converted to decimal sequences by converting each P binary number to a decimal number, resulting in T frequency-agile radar sequences of length L. Autocorrelation is then performed on the T frequency-agile radar sequences generated from each basic chaotic sequence to obtain the frequency-agile radar sequence with the best randomness in that basic chaotic sequence and the optimal threshold corresponding to that optimal frequency-agile radar sequence.

[0083] S5, train a CNN network on the training set Ψ with a given chaotic sequence of length M to determine the optimal threshold λ for the given chaotic sequence. k .

[0084] In this embodiment, when using a CNN network to identify a given sequence of jerks, the training flowchart is as follows: Figure 3 As shown in the figure. This embodiment uses the Glorot initialization method from Keras, which has uniform distribution characteristics, and the Adam optimizer. The network learning rate is 0.001, and the cross-entropy loss function is used. K-fold cross-validation is used during training to improve data utilization and reduce the possibility of overfitting.

[0085] S6, based on the optimal threshold λ k A given chaotic sequence is subjected to binary quantization to obtain a binary sequence; the second frequency-agile radar sequence is obtained from the binary sequence.

[0086] A mapping relationship can be established between chaotic sequences and frequency-agile radar sequences to transform fractional sequences into integer sequences, thereby generating frequency-hopping radar signals. In choosing the mapping relationship, binary quantization is easier to implement than integer quantization and can enhance the randomness of the generated frequency-hopping sequence, thus improving the anti-jamming capability of the frequency-agile radar signal. However, in the application of binary quantization, a threshold is required. Using different thresholds for the same chaotic sequence can lead to frequency-agile radar sequences with varying degrees of randomness. Therefore, different threshold choices will affect the randomness of the frequency-agile radar sequence. Consequently, when using a target chaotic sequence to generate a frequency-agile radar sequence, different threshold choices can easily lead to the frequency-agile radar sequence generation expression being intercepted by jammers, resulting in a significant reduction in anti-jamming capability.

[0087] CNNs, or Convolutional Neural Networks, are a type of feedforward neural network that incorporates convolutional computations and has a deep structure. They possess powerful feature parameter extraction capabilities and are one of the representative algorithms of deep learning. This embodiment creatively combines CNNs with binary quantization methods to determine the optimal threshold λ for binary quantization of the target chaotic sequence. k It can map the chaotic sequence of a target into a frequency-agile radar sequence with good random performance, and has good real-time performance and randomness, so as to reduce the probability of being intercepted by jammers.

[0088] Specifically, when generating the training data set Ψ, the length of the chaotic sequence used is 256. Since the initial value of the chaotic sequence is in the range (-0.5, 0.5), if the interval ε of the chaotic sequence is set to 0.0001, 10,000 basic chaotic sequences will be generated. Multiple thresholds are then set to convert each basic chaotic sequence into a binary sequence, resulting in a binary sequence length of 256. In this embodiment, every 8 binary numbers can be converted into decimal numbers, thus generating 32 frequency-agile radar sequences, with a value range of [0, 255].

[0089] To select optimal thresholds with good randomness, a coarse screening of thresholds is necessary. Through repeated experiments, it was found that the optimal thresholds are concentrated in the range [0.21, 0.3]. To obtain more precise thresholds, the interval [0.21, 0.3] was subdivided into ten equal parts, determining ten thresholds: {0.21 0.22 0.23 0.24 0.25 0.26 0.27 0.28 0.29 0.30}. Then, optimal thresholds were selected for 10,000 basic chaotic sequences. These ten thresholds were divided into ten labels, and the threshold corresponding to each basic chaotic sequence was transformed into a label, resulting in a matrix of size 10000 × 257.

[0090] First, each basic chaotic sequence and its corresponding optimal threshold are organized into labels for the CNN network. Each basic chaotic sequence corresponds to one of the ten labels: 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9.

[0091] Where b n256 λ represents the 256th number in the nth Bernoulli sequence. n The threshold representing the optimal selection threshold for the nth Bernoulli sequence of length 256. A CNN network is used to train the training set Ψ, and the training process is as follows: Figure 3 As shown.

[0092] To facilitate training, the selection of neurons and network layers was based on the elastic pants method proposed by Google scientist Vincent Vanhoucke. The optimized neural network structure is shown in Table 1:

[0093] In this embodiment, the initialization method used is the Glorot initialization of Keras, which has uniform distribution characteristics. The optimizer is the Adam optimizer, the network learning rate is 0.001, and the network loss function is the cross-entropy loss function. During training, K-fold cross-validation is used to improve data utilization and reduce the possibility of overfitting.

[0094] The CNN network achieved a validation accuracy of 96.8%, indicating that the CNN network was trained effectively. Through training, the CNN network was able to select the optimal threshold λ online. k CNN networks possess powerful pattern recognition capabilities. By transforming the threshold optimization problem into a pattern recognition problem, they can accurately and quickly determine the optimal threshold λ. k This enables the rapid generation of frequency-agile radar sequences.

[0095] S7. The first frequency-agile radar sequence and the second frequency-agile radar sequence are spliced ​​together to obtain the target frequency-agile radar sequence.

[0096] In this embodiment, the first frequency-agile radar sequence can be spliced ​​in front of the second frequency-agile radar sequence, or the first frequency-agile radar sequence can be spliced ​​in behind the second frequency-agile radar sequence.

[0097] In this embodiment, the length D of the target frequency-agile radar sequence is

[0098] D = L + Q × N

[0099] Where Q is the number of target chaotic sequences.

[0100] Preferably, M can be made equal to P × N. That is, the length M of the given chaotic sequence can be equal to the product of the number Q of the target chaotic sequence and the fixed bit truncation number P used in the binary-to-decimal conversion.

[0101] Frequency-hopping radar signals are radar signals whose carrier frequency changes continuously over time. Frequency-hopping signals can counteract active interference. Therefore, the better the randomness of a frequency-agile radar signal, the stronger its anti-jamming capability.

[0102] Simulated annealing is an optimization algorithm that simulates the annealing process of solid materials. It is a stochastic optimization algorithm based on the Monte Carlo iterative solution strategy. Simulated annealing has the advantages of fast optimization speed, strong robustness, and low susceptibility to getting trapped in local optima. It can be combined with integer quantization.

[0103] CNNs, or Convolutional Neural Networks, are a type of feedforward neural network that incorporates convolutional computations and has a deep structure. They possess powerful feature parameter extraction capabilities and are one of the representative algorithms of deep learning. This embodiment creatively combines CNNs with binary quantization methods to determine the optimal threshold λ for binary quantization of the target chaotic sequence. k It can map the chaotic sequence of a target into a frequency-agile radar sequence with good random performance, and has good real-time performance and randomness, so as to reduce the probability of being intercepted by jammers.

[0104] In the frequency agile radar signal generation method provided in this embodiment, the frequency agile radar sequence is divided into two parts. One part uses a simulated annealing algorithm to enhance the randomness of the part, and the other part uses a convolutional neural network method to enhance the randomness of the part. Then, the two parts are spliced ​​together to form the target frequency agile radar sequence.

[0105] In other words, this embodiment enhances the randomness of the target frequency-agile radar sequence by using two optimization methods: simulated annealing algorithm and convolutional neural network method. This enables this embodiment to solve the problems of traditional radar systems, such as the difficulty of signal detection and interception, and the susceptibility to interference and the significant impact of interference.

[0106] Therefore, the frequency agile radar signal generation method provided in this embodiment can improve the working capability of frequency agile radar in complex electronic environments; and the frequency agile radar sequence designed in this embodiment can improve anti-interference capability while generating a normal target detection capability.

[0107] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating frequency-agile radar signals, characterized in that: Includes the following steps, S1, quantize the initial chaotic sequence of length N into an integer chaotic sequence; S2, Use simulated annealing algorithm to determine the optimal control parameters for the integer chaotic sequence. And based on the optimal control parameters Generate a target chaotic sequence of length N; S3, construct the first frequency-agile radar sequence based on the target chaotic sequence; S4, based on the given chaotic sequence length M and the set initial value interval of the chaotic sequence. Establish a data training set The training set of data Used to determine the optimal threshold for a given chaotic sequence during binary quantization. ; S5, using this data set for training The CNN network model is trained to obtain the CNN network; the CNN network is then used to identify a given chaotic sequence of length N and to determine the optimal threshold. ; S6, based on the optimal threshold The given chaotic sequence is subjected to binary quantization to obtain a binary sequence; the second frequency-agile radar sequence is obtained based on the binary sequence. S7, construct a target frequency agile radar sequence based on the first frequency agile radar sequence and the second frequency agile radar sequence.

2. The method for generating frequency-agile radar signals as described in claim 1, characterized in that: Step S1 includes the following steps: S11, Generate the initial chaotic sequence based on the given initial value of the chaotic sequence and the length N of the target chaotic sequence; S12, quantize the initial chaotic sequence into the integer chaotic sequence.

3. The method for generating frequency-agile radar signals as described in claim 2, characterized in that: Step S2 details Includes the following steps, S21, Construct the autocorrelation function with the minimum main lobe-side lobe ratio for this integer chaotic sequence; S22, invert the autocorrelation function and determine the maximum main lobe-side lobe ratio function of the integer chaotic sequence; S23. The simulated annealing algorithm is used to iteratively optimize the maximum main lobe-to-side lobe ratio function to obtain the optimal control parameters for the integer chaotic sequence. ; S24, based on the optimal control parameters Generate a target chaotic sequence of length N.

4. The method for generating frequency-agile radar signals as described in claim 2, characterized in that: Step S2 is followed by S0, S0: Using the target chaotic sequence generated in step S2 as the initial value, generate the next chaotic sequence of length N. Repeat steps S1 to S2 on the next chaotic sequence to obtain the next target chaotic sequence.

5. The method for generating frequency-agile radar signals as described in any one of claims 1-4, characterized in that: The training set Established through the following steps S41, based on the given chaotic sequence length M and the set initial value interval of the chaotic sequence... ,generate A basic chaotic sequence of length M; S42, for each basic chaotic sequence, utilize Threshold Perform binary quantization to obtain Given a set of M basic binary sequences; convert all generated basic binary sequences into basic decimal sequences by converting each P binary number into a decimal number, thus obtaining... A basic frequency-agile radar sequence of length L, wherein... , , ; S43 will be generated from each basic chaotic sequence Autocorrelation calculations are performed on each of the basic frequency-agile radar sequences to obtain the basic frequency-agile radar sequence with the best randomness in the basic chaotic sequence and the optimal threshold corresponding to the optimal basic frequency-agile radar sequence. S44, construct a data training set by combining all basic chaotic sequences and the corresponding optimal thresholds for each basic chaotic sequence. .

6. The method for generating frequency-agile radar signals as described in claim 5, characterized in that: Construct the data training set It also includes the following steps, Step 1: Repeat steps S42 to S43 to determine the range of all preferred threshold values. ; Step two, divide the interval The thresholds are divided and a set of preferred thresholds consisting of T preferred thresholds is obtained. Step 3: Using the methods described in steps S42 and S43, select the optimal threshold from the T preferred thresholds for each basic chaotic sequence; and combine all basic chaotic sequences with the optimal thresholds corresponding to each basic chaotic sequence to form the data training set. .

7. The method for generating frequency-agile radar signals as described in claim 1, characterized in that: When using a CNN network to identify a given sequence of jerks, the initialization method chosen is the Glorot initialization of Keras with uniform distribution characteristics, the optimizer is the Adam optimizer, the network learning rate is 0.001, and the network loss function is the cross-entropy loss function.

8. The method for generating frequency-agile radar signals as described in claim 1, characterized in that: When using a CNN network to identify a given sequence of jerks, K-fold cross-validation is used to improve data utilization.

9. The method for generating frequency-agile radar signals as described in claim 5, characterized in that: The length D of the target frequency-agile radar sequence is Where Q is the number of target chaotic sequences.

10. The method for generating frequency-agile radar signals as described in claim 9, characterized in that: 。

Images