Multifunctional radar signal sequence prediction method based on similarity analysis

Through similarity analysis method, the multifunctional radar signal sequence is reconstructed and weighted to predict multifunctional radar signal sequences, and the problem of relying on prior knowledge for multifunctional radar signal sequence prediction in the prior art is solved, and high accuracy prediction is achieved in non-ideal scenarios.

CN119064885AActive Publication Date: 2024-12-03UNIV OF ELECTRONICS SCI & TECH OF CHINA +2
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411321756.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-23
Publication Date
2024-12-03
Estimated Expiration
2044-09-23

AI Technical Summary

Technical Problem

The existing multifunctional radar signal sequence prediction methods rely on a large amount of prior knowledge and pulse style information, making it difficult to effectively predict multifunctional radar signal sequences in non-ideal third-party signal reception scenarios.

Method used

The radar signal sequence parameters are reconstructed through a trend filtering algorithm, and the dynamic time regularization algorithm calculates the signal sequence similarity, and a weighted prediction prototype is constructed using the historical moment signal sequence with the highest similarity to realize signal sequence prediction.

Benefits of technology

It does not rely on pulse style prior information, improves signal sequence prediction accuracy, and is suitable for multifunctional radar signal sequence prediction in non-ideal third-party signal reception scenarios, which is significantly better than existing methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119064885A_ABST
    Figure CN119064885A_ABST
Patent Text Reader

Abstract

The invention discloses a multifunctional radar signal sequence prediction method based on similarity analysis, and the method comprises the steps: firstly carrying out the preprocessing of a radar signal sequence, reconstructing radar signal sequence parameters through a trend filtering algorithm, and carrying out the simplified representation of the signal sequence; calculating the similarity between the current and historical moment signal sequences through a dynamic time warping algorithm, constructing a prediction prototype by using a plurality of historical moment signal sequences with the highest similarity, and weighting the prediction prototype to obtain a radar signal sequence prediction result. Compared with an existing method, the method does not depend on pulse pattern prior information, the signal sequence prediction accuracy can be improved, the multifunctional radar signal sequence in a non-ideal third-party signal receiving scene can be effectively predicted, and the method has important significance in the field of radar signal analysis and processing. The method is suitable for predicting the multifunctional radar signal sequence in a non-ideal detection scene.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of multifunctional radars, and in particular relates to a multifunctional radar signal sequence prediction method based on similarity analysis. Background Art

[0002] Multifunction radar refers to a radar system that can perform multiple tasks and support multiple functions. It has flexible and changeable working modes and changeable waveform characteristics, so it has been widely used. The variability of multifunction radars poses a huge challenge to third-party signal reception and analysis. At present, in the multifunction radar behavior recognition technology, the receiving system mainly focuses on the radar's current transmission signal, but this method is not applicable to multifunction radars with flexible waveform changes. Only when the receiving system can effectively predict the multifunction radar signal at future moments can it better grasp the radar's behavioral intentions and formulate the optimal response strategy.

[0003] At present, there are still few studies on the prediction of multi-function radar signal sequences at home and abroad. The literature "Modelling, Learning and Prediction of Complex Radar Emitter Behavior, 2019 18th IEEEInternational Conference On Machine Learning And Applications (ICMLA), IEEE, 2019, pp.305-310" uses an extended multi-function radar signal hierarchical structure model and proposes a multi-function radar signal sequence prediction method based on long short-term memory network (LSTM), and trains prediction networks for different levels of the signal. The literature "Prediction of Multi-Function Radar Signal Sequence Using Encoder-Decoder Structure, 2022 7th International Conference on Signal and ImageProcessing (ICSIP), IEEE, 2022, pp.152-156" proposes a multi-function radar signal sequence prediction method based on encoder-decoder structure, which achieves good prediction results. However, the existing methods rely heavily on prior knowledge and require a large number of signal sequence samples as training data. Therefore, there is a lack of methods independent of prior information in the field of multi-function radar signal sequence prediction, which still needs further research. Summary of the invention

[0004] The present invention provides a multi-function radar signal sequence prediction method based on similarity analysis, which can effectively predict the multi-function radar signal sequence in a non-ideal third-party signal reception scenario.

[0005] The technical solution adopted by the present invention is: a multifunctional radar signal sequence prediction method based on similarity analysis, and the specific steps are as follows:

[0006] S1, radar signal sequence preprocessing;

[0007] The multi-function radar signal sequence intercepted by the receiver is preprocessed, the radar signal sequence parameters are reconstructed through the trend filtering algorithm, and the signal sequence is divided into sub-segments and simplified.

[0008] S2, radar signal sequence similarity calculation;

[0009] Based on step S1, the similarity between the current and historical signal sequences is calculated using the dynamic time warping (DTW) algorithm.

[0010] S3, signal sequence prediction result output;

[0011] Based on the signal sequence similarity calculated in step S2, a prediction prototype is constructed using the post-sequence corresponding to the historical moment signal sequence with the highest similarity, and the prediction prototype is weighted to obtain the radar signal sequence prediction result.

[0012] Furthermore, the step S1 is specifically as follows:

[0013] S11, preprocessing the multi-function radar signal sequence intercepted by the receiving party;

[0014] The multi-function radar signal sequence intercepted by the receiver is recorded as {p, a}. The sequence segment switching point is found according to the pulse amplitude information of the signal sequence. The segment switching point of the radar signal sequence is T seg The expression is as follows:

[0015] T seg ={t||a t -a t-1 |>ε s}

[0016] Where p represents the signal parameters to be predicted, including pulse repetition interval and pulse width, a represents pulse amplitude; a t represents the amplitude of the tth pulse of the signal sequence, ε s Indicates the pulse amplitude change threshold.

[0017] Then the sequence is segmented according to the switching points of the signal sequence segments, and the i-th sequence segment is recorded as p i .

[0018] S12, based on the signal sequence segmentation result, using a trend filtering algorithm, from the signal sequence segment p i Reconstruct the signal sequence parameters in and obtain the reconstruction result of the i-th signal sequence parameters

[0019] Among them, the signal sequence fragment p i There are various errors in it, including: missed pulses and false pulses.

[0020] The noise in the signal sequence parameters is modeled as a Gaussian mixture distribution, and the prior function of the signal sequence parameters is modeled as a Gibbs distribution. The optimization objective function expression is as follows:

[0021]

[0022] in, represents the parameters that need to be estimated, represents the true value of the signal parameter to be estimated, express The nth element of i (n) represents the signal sequence segment observation value p i The nth element of , Π, Σ represent the set of weights and variances of each Gaussian distribution in the Gaussian mixture distribution, and N represents the signal sequence The length of M represents the number of Gaussian distributions. represents the posterior function. represents a single Gaussian distribution, π m represents the weight of the mth Gaussian distribution, η m Represents the variance of the mth Gaussian distribution. φ λ,c (·) represents the sparsity inducing function, and c is a small positive value used to avoid φ λ,c (·) is calculated to be 0. F is a matrix, Representation vector The lth element of .

[0023] In the optimization objective function, the specific expression of the sparse induction function is as follows:

[0024]

[0025] Here, λ represents the proportionality coefficient.

[0026] In the optimization objective function, for different signal sequence parameter modulation types, different matrices F need to be set to extract the sparsity of the signal sequence in different domains.

[0027] S13, based on the signal sequence parameter reconstruction result, dividing the signal sequence into sub-segments, that is, performing different processing on signal sequence segments of different modulation types;

[0028] S14. Based on the signal sequence sub-segment division result, perform different simplified representations on the signal sequence sub-segments for different signal sequence parameter modulation types.

[0029] Furthermore, the step S2 is specifically as follows:

[0030] First, construct the DTW distance matrix W, and the initialization expression is as follows:

[0031]

[0032] Among them, r and s represent the indices of the matrix, and W(r,s) represents the cumulative distance from (0,0) to (r,s).

[0033] The expression for iterative calculation of matrix W elements is as follows:

[0034] W(1,1)=D(p i,1 ,p j,1 )

[0035]

[0036] 1≤r≤R,1≤s≤S

[0037] Among them, D(p i,r ,p j,s ) represents the signal sequence sub-fragment p i,r With p j,s For different signal sequence parameter modulation types, different distance calculation methods are used.

[0038] According to the distance matrix W, the signal sequence fragment p i With p j The DTW distance expression between them is as follows:

[0039]

[0040] Then, the DTW distance is converted into the similarity between [0,1], denoted as S(p i ,p j ). Signal sequence fragment p i With p j The similarity definition expression between them is as follows:

[0041]

[0042] Where σ represents the scaling factor.

[0043] Finally, let the sliding window length be W, take the latest W signal sequence fragments as the current signal sequence, denoted as P cur ; Take W signal sequence fragments in the remaining signal sequence in the form of sliding windows each time, where the i-th sliding window is recorded as Calculate the similarity between the current and historical signal sequence sliding windows. The expression is as follows:

[0044]

[0045] Among them, P cur (w) They represent the w-th sub-segment of the i-th sliding window of the current signal sequence and the historical signal sequence respectively.

[0046] Furthermore, the step S3 is specifically as follows:

[0047] First, according to the similarity between the current and historical signal sequence sliding windows obtained in step S2, find the J sliding windows with the highest similarity, which are numbered as v j ,j=1,2,...,J, and its corresponding post-order sequence is recorded as The subsequent sequences of the J sliding windows with the highest similarity are used as signal sequence prediction prototypes.

[0048] Then, according to the similarity between the J historical moment signal sequence sliding windows and the current moment signal sequence, the weight ω of the J prototypes in weighting is calculated. j , the expression is as follows:

[0049]

[0050] Finally, according to the weight ω j , weight each prototype and get the signal sequence prediction result p pred The expression is as follows:

[0051]

[0052] When the prediction prototypes have different lengths, the prototypes are of equal length, and the signal sequence prediction result expression is as follows:

[0053]

[0054] Among them, T p represents the length of the sequence taken, express Front T p elements.

[0055] Beneficial effects of the present invention: The method of the present invention first preprocesses the radar signal sequence, reconstructs the radar signal sequence parameters through a trend filtering algorithm, simplifies the signal sequence, and then calculates the similarity between the current and historical signal sequences through a dynamic time warping algorithm. Finally, a prediction prototype is constructed using several historical signal sequences with the highest similarity, and the prediction prototype is weighted to obtain the radar signal sequence prediction result. Compared with the existing methods, the method of the present invention does not rely on pulse pattern prior information, and can improve the accuracy of signal sequence prediction. It can effectively predict multi-function radar signal sequences in non-ideal third-party signal reception scenarios, which is of great significance in the field of radar signal analysis and processing, and is suitable for the prediction of multi-function radar signal sequences in non-ideal detection scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The present invention is a flowchart of a method for predicting a multifunctional radar signal sequence based on similarity analysis.

[0057] Figure 2 Schematic diagram of radar signal sequence prediction prototype construction and output in an embodiment of the present invention.

[0058] Figure 3 Schematic diagram of the prediction accuracy of radar signal sequence at different error levels in an embodiment of the present invention.

[0059] Figure 4 Schematic diagram of the prediction accuracy of radar signal sequence under different prediction step sizes in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The method of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0061] like Figure 1 As shown, a flow chart of a multifunctional radar signal sequence prediction method based on similarity analysis of the present invention, the specific steps are as follows:

[0062] S1, radar signal sequence preprocessing;

[0063] The multi-function radar signal sequence intercepted by the receiver is preprocessed, the radar signal sequence parameters are reconstructed through the trend filtering algorithm, and the signal sequence is divided into sub-segments and simplified.

[0064] S2, radar signal sequence similarity calculation;

[0065] Based on step S1, the similarity between the current and historical signal sequences is calculated using the dynamic time warping (DTW) algorithm.

[0066] S3, signal sequence prediction result output;

[0067] Based on the signal sequence similarity calculated in step S2, a prediction prototype is constructed using the post-sequence corresponding to the historical moment signal sequence with the highest similarity, and the prediction prototype is weighted to obtain the radar signal sequence prediction result.

[0068] In this embodiment, the step S1 is specifically as follows:

[0069] S11, preprocessing the multi-function radar signal sequence intercepted by the receiving party;

[0070] The multi-function radar signal sequence intercepted by the receiver is recorded as {p, a}. The sequence segment switching point is found according to the pulse amplitude information of the signal sequence. The segment switching point of the radar signal sequence is T seg The expression is as follows:

[0071] T seg ={t||a t -a t-1 |>ε s}

[0072] Where p represents the signal parameters to be predicted, including pulse repetition interval and pulse width, a represents pulse amplitude; a t represents the amplitude of the tth pulse of the signal sequence, ε s Indicates the pulse amplitude change threshold.

[0073] Then the sequence is segmented according to the switching points of the signal sequence segments, and the i-th sequence segment is recorded as p i .

[0074] S12, based on the signal sequence segmentation result, using a trend filtering algorithm, from the signal sequence segment p i Reconstruct the signal sequence parameters in and obtain the reconstruction result of the i-th signal sequence parameters

[0075] Among them, the signal sequence fragment p i There are various errors in it, including: missed pulses and false pulses.

[0076] The noise in the signal sequence parameters is modeled as a Gaussian mixture distribution, and the prior function of the signal sequence parameters is modeled as a Gibbs distribution. The optimization objective function expression is as follows:

[0077]

[0078] in, represents the parameters that need to be estimated, represents the true value of the signal parameter to be estimated, express The nth element of i(n) represents the signal sequence segment observation value p i The nth element of , Π, Σ represent the set of weights and variances of each Gaussian distribution in the Gaussian mixture distribution, and N represents the signal sequence The length of M represents the number of Gaussian distributions. represents the posterior function. represents a single Gaussian distribution, π m represents the weight of the mth Gaussian distribution, η m Represents the variance of the mth Gaussian distribution. φ λ,c (·) represents the sparsity inducing function, and c is a small positive value used to avoid φ λ,c (·) is calculated to be 0. F is a matrix, Representation vector The lth element of .

[0079] In the optimization objective function, the specific expression of the sparse induction function is as follows:

[0080]

[0081] Here, λ represents the proportionality coefficient.

[0082] In the optimization objective function, for different signal sequence parameter modulation types, different matrices F need to be set to extract the sparsity of the signal sequence in different domains, as follows:

[0083] In this embodiment, for constant and group-variable modulation sequence parameters, in order to extract the sparsity of the signal sequence parameters in the first-order difference domain, a first-order difference matrix is ​​used as the matrix F, and the definition expression is as follows:

[0084]

[0085] In this example, for the slip modulation sequence parameters, in order to extract the sparsity of the signal sequence parameters in the second-order difference domain, the second-order difference matrix is ​​used as the matrix F, and the definition expression is as follows:

[0086]

[0087] In this embodiment, for the sinusoidal modulation sequence parameters, in order to extract the sparsity of the signal sequence parameters in the frequency domain, the Fourier matrix is ​​used as the matrix F, and the definition expression is as follows:

[0088]

[0089] in, N represents the signal sequence The length of K represents the number of FFT points.

[0090] S13, based on the signal sequence parameter reconstruction result, dividing the signal sequence into sub-segments, that is, performing different processing on signal sequence segments of different modulation types;

[0091] In this embodiment, for constant, group-variable and sliding modulation sequence parameters, the trend switching point in the signal sequence segment is found, and the signal sequence segment is further divided into sub-segments. The j-th sub-segment of the i-th sequence segment is recorded as p i,j ; The signal sequence fragment p i,j Internal trend switching point T trend The expression is as follows:

[0092]

[0093] Among them, D (k+1) represents the k+1 order difference matrix, Representation vector The u-th element of t Represents a certain threshold value. Different order difference matrices are used for different parameter modulation types. For constant and group-variable modulation sequence parameters, let k = 0, that is, use the first-order difference matrix; for sliding modulation sequence parameters, let k = 1, that is, use the second-order difference matrix.

[0094] In this embodiment, the sine modulation sequence parameters are not further segmented, that is, the sequence segment p i Contains only one sub-segment p i,1 .

[0095] S14. Based on the signal sequence sub-segment division result, according to different signal sequence parameter modulation types, perform different simplified representations on the signal sequence sub-segments;

[0096] In this embodiment, for constant, group-variable and sliding modulation sequence parameters, the signal sequence sub-segment is converted into a triple consisting of an initial value, a length and a slope, and the expression is as follows:

[0097] r i,j = ij ,l ij ,k ij >

[0098] Among them, b ij Indicates the signal sequence subfragment p i,j The initial value of l ij Indicates the signal sequence subfragment p i,j The length of k ij Indicates the signal sequence subfragment p i,j The slope of the trend.

[0099] ​In this embodiment, for the sinusoidal modulation sequence parameters, Fourier transform is used to convert the signal sequence sub-segment into a frequency domain representation, and the expression is as follows:

[0100]

[0101] In this embodiment, step S2 is specifically as follows:

[0102] First, construct the DTW distance matrix W, and the initialization expression is as follows:

[0103]

[0104] Among them, r and s represent the indices of the matrix, and W(r,s) represents the cumulative distance from (0,0) to (r,s).

[0105] The expression for iterative calculation of matrix W elements is as follows:

[0106] W(1,1)=D(p i,1 ,p j,1 )

[0107]

[0108] 1≤r≤R,1≤s≤S

[0109] Among them, D(p i,r ,p j,s ) represents the signal sequence sub-fragment p i,r With p j,s For different signal sequence parameter modulation types, different distance calculation methods are used.

[0110] In this embodiment, for constant, group-variable and sliding modulation sequence parameters, the sequence sub-segment p i,r With p j,s The distance expression between them is as follows:

[0111]

[0112] Among them, b 1 , l 1 , k 1 Respectively represent p i,r With p j,s The initial value, length and trend slope of the shorter one. Similarly, b 2 , l 2 , k 2 The longer one.

[0113] In this embodiment, for the sinusoidal modulation sequence parameters, the sequence sub-segment p i,r With p j,s The distance expression between them is as follows:

[0114]

[0115] According to the distance matrix W, the signal sequence fragment p i With p j The DTW distance expression between them is as follows:

[0116]

[0117] Then, the DTW distance is converted into the similarity between [0,1], denoted as S(p i ,p j ). Signal sequence fragment p i With p j The similarity definition expression between them is as follows:

[0118]

[0119] Where σ represents the scaling factor.

[0120] Finally, let the sliding window length be W, take the latest W signal sequence fragments as the current signal sequence, denoted as P cur ; Take W signal sequence fragments in the remaining signal sequence in the form of sliding windows each time, where the i-th sliding window is recorded as Calculate the similarity between the current and historical signal sequence sliding windows. The expression is as follows:

[0121]

[0122] Among them, P cur (w) They represent the w-th sub-segment of the i-th sliding window of the current signal sequence and the historical signal sequence respectively.

[0123] like Figure 2 As shown, in this embodiment, step S3 is specifically as follows:

[0124] First, according to the similarity between the current and historical signal sequence sliding windows obtained in step S2, find the J sliding windows with the highest similarity, which are numbered as v j ,j=1,2,...,J, and its corresponding post-order sequence is recorded as The subsequent sequences of the J sliding windows with the highest similarity are used as signal sequence prediction prototypes.

[0125] Then, according to the similarity between the J historical moment signal sequence sliding windows and the current moment signal sequence, the weight ω of the J prototypes in weighting is calculated. j , the expression is as follows:

[0126]

[0127] Finally, according to the weight ω j , weight each prototype and get the signal sequence prediction result p pred The expression is as follows:

[0128]

[0129] When the prediction prototypes have different lengths, the prototypes are of equal length, and the signal sequence prediction result expression is as follows:

[0130]

[0131] Among them, T p represents the length of the sequence taken, express Front T p elements.

[0132] This embodiment further performs simulation verification on the method of the present invention.

[0133] Simulation 1: Simulate the prediction of radar signal PRI sequence to establish a multi-function radar signal PRI sequence data set in a multi-target detection scenario, including a variety of PRI modulation types. Define the error level as the degree of false pulses and missed pulses in the signal sequence. When the error level is α, the false pulse ratio and the missed pulse ratio are both α; define the signal sequence prediction accuracy as the ratio of the number of predictions with a mean absolute percentage error (MAPE) less than 10% to the total number of predictions. Set the error level range to 0 to 0.15, with a step of 0.015. Input the generated signal sequence data set into the method of the present invention and the method based on the encoder-decoder structure (denoted as GRU-TD), and the obtained radar signal sequence prediction accuracy is as follows: Figure 3 As shown in the figure, it can be seen that the method of the present invention is significantly better than the existing method at different error levels.

[0134] Simulation 2: Establish a multi-function radar signal PRI sequence data set in a multi-target detection scenario, with a fixed error level of 0.15. Input the generated signal sequence data set into the method of the present invention and the method based on the encoder-decoder structure, and calculate the signal sequence prediction accuracy under different prediction step sizes, such as Figure 4 As shown in the figure, it can be seen that the accuracy of the measurement method of the present invention is maintained above 0.7 under different prediction step sizes, which is significantly better than the existing method.

[0135] The above simulation results show that the method of the present invention can significantly improve the accuracy of signal sequence prediction compared with the existing methods, thereby verifying that the method of the present invention is practical and effective in multi-function radar signal sequence prediction.

[0136] In summary, compared with the existing methods, the method of the present invention does not rely on prior information of pulse patterns, and can improve the accuracy of signal sequence prediction. It can effectively predict the multi-function radar signal sequence in non-ideal third-party signal reception scenarios, which is of great significance in the field of radar signal analysis and processing, and is suitable for the prediction of multi-function radar signal sequences in non-ideal detection scenarios.

[0137] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A multifunctional radar signal sequence prediction method based on similarity analysis, the specific steps are as follows: S1, radar signal sequence preprocessing; Preprocess the multi-function radar signal sequence intercepted by the receiver, reconstruct the radar signal sequence parameters through the trend filtering algorithm, and divide the signal sequence into sub-segments and simplify the representation; S2, radar signal sequence similarity calculation; Based on step S1, the similarity between the current and historical signal sequences is calculated using a dynamic time warping (DTW) algorithm; S3, signal sequence prediction result output; Based on the signal sequence similarity calculated in step S2, a prediction prototype is constructed using the post-sequence corresponding to the historical moment signal sequence with the highest similarity, and the prediction prototype is weighted to obtain the radar signal sequence prediction result.

2. The method for predicting a multifunctional radar signal sequence based on similarity analysis according to claim 1, characterized in that: The step S1 is specifically as follows: S11, preprocessing the multi-function radar signal sequence intercepted by the receiving party; The multi-function radar signal sequence intercepted by the receiver is recorded as {p, a}. The sequence segment switching point is found according to the pulse amplitude information of the signal sequence. The segment switching point of the radar signal sequence is T seg The expression is as follows: T seg ={t||a t -a t-1 |>ε s } Where p represents the signal parameters to be predicted, including pulse repetition interval and pulse width, a represents pulse amplitude; a t represents the amplitude of the tth pulse of the signal sequence, ε s Indicates the pulse amplitude change threshold; Then the sequence is segmented according to the switching points of the signal sequence segments, and the i-th sequence segment is recorded as p i ; S12, based on the signal sequence segmentation result, using a trend filtering algorithm, from the signal sequence segment p i Reconstruct the signal sequence parameters in and obtain the reconstruction result of the i-th signal sequence parameters Among them, the signal sequence fragment p i There are various errors in it, including: missed pulses, false pulses; The noise in the signal sequence parameters is modeled as a Gaussian mixture distribution, and the prior function of the signal sequence parameters is modeled as a Gibbs distribution. The optimization objective function expression is as follows: in, represents the parameters that need to be estimated, represents the true value of the signal parameter to be estimated, express The nth element of i (n) represents the observed value p of the signal sequence segment i The nth element of , Π, Σ represent the set of weights and variances of each Gaussian distribution in the Gaussian mixture distribution, and N represents the signal sequence The length of M represents the number of Gaussian distributions. represents the posterior function; represents a single Gaussian distribution, π m represents the weight of the mth Gaussian distribution, η m represents the variance of the mth Gaussian distribution; φ λ,c (·) represents the sparsity inducing function, and c is a small positive value used to avoid φ λ,c (·) The calculation result is 0; F is a matrix, Representation vector The lth element of In the optimization objective function, the specific expression of the sparse induction function is as follows: Among them, λ represents the proportionality coefficient; In the optimization objective function, for different signal sequence parameter modulation types, different matrices F need to be set to extract the sparsity of the signal sequence in different domains; S13, based on the signal sequence parameter reconstruction result, dividing the signal sequence into sub-segments, that is, performing different processing on signal sequence segments of different modulation types; S14. Based on the signal sequence sub-segment division result, perform different simplified representations on the signal sequence sub-segments for different signal sequence parameter modulation types.

3. The method for predicting a multifunctional radar signal sequence based on similarity analysis according to claim 1, characterized in that: The step S2 is specifically as follows: First, construct the DTW distance matrix W, and the initialization expression is as follows: Among them, r and s represent the index of the matrix, and W(r,s) represents the cumulative distance from (0,0) to (r,s); The expression for iterative calculation of matrix W elements is as follows: W(1,1)=D(p i,1 ,p j,1 ) 1≤r≤R,1≤s≤S Among them, D(p i,r ,p j,s ) represents the signal sequence sub-fragment p i,r With p j,s The distance between them; for different signal sequence parameter modulation types, different distance calculation methods are used; According to the distance matrix W, the signal sequence fragment p i With p j The DTW distance expression between them is as follows: Then, the DTW distance is converted into the similarity between [0,1], denoted as S(p i ,p j ); signal sequence fragment p i With p j The similarity definition expression between them is as follows: Among them, σ represents the scaling factor; Finally, let the sliding window length be W, take the latest W signal sequence fragments as the current signal sequence, denoted as P cur ; Take W signal sequence fragments in the remaining signal sequence in the form of sliding windows each time, where the i-th sliding window is recorded as Calculate the similarity between the current and historical signal sequence sliding windows. The expression is as follows: Among them, P cur (w) They represent the w-th sub-segment of the i-th sliding window of the current signal sequence and the historical signal sequence respectively.

4. The method for predicting a multifunctional radar signal sequence based on similarity analysis according to claim 1, characterized in that: The step S3 is specifically as follows: First, according to the similarity between the current and historical signal sequence sliding windows obtained in step S2, find the J sliding windows with the highest similarity, which are numbered as v j ,j=1,2,...,J, and its corresponding post-order sequence is recorded as The subsequent sequences of the J sliding windows with the highest similarity are used as the signal sequence prediction prototypes; Then, according to the similarity between the J historical moment signal sequence sliding windows and the current moment signal sequence, the weight ω of the J prototypes in weighting is calculated. j , the expression is as follows: Finally, according to the weight ω j , weight each prototype and get the signal sequence prediction result p pred The expression is as follows: When the prediction prototypes have different lengths, the prototypes are of equal length, and the signal sequence prediction result expression is as follows: Among them, T p represents the length of the sequence taken, express Front T p elements.

Citation Information

Patent Citations

  • Radar target detection method and system based on convolutional neural network sequence classification

    CN111880158A

  • Hydrological time sequence similarity search method based on FAD and DTW

    CN114911846A

  • DTW-DCRNN-based chemical industry park noise prediction method

    CN115659609A

  • Multifunctional radar working mode identification method based on time sequence segmentation and clustering

    CN117949897A