An intelligent method for extracting timing parameters of radar composite jammers based on frequency agility

Abstract

The present invention discloses a method for intelligently extracting the timing parameters of radar composite interference based on frequency agility, which relates to the field of radar technology. The method utilizes an interference separation network that integrates convolution enhancement and a Transformer attention mechanism to separate and process radar echo signals. The interference separation network enhances local feature extraction capabilities through convolution operations and introduces a cross-scale attention mechanism to capture the global correlation of interference signals, thereby achieving precise separation of multi-component composite ISRJ signals to obtain multiple separated ISRJ component signals. Furthermore, through dynamic threshold detection and a robustness measurement algorithm based on pulse detection and clustering optimization, accurate measurement of interference parameters such as the slice width, forwarding width, and forwarding times of the ISRJ component signals is achieved based on the time domain characteristics of the ISRJ component signals.

Description

Technical Field

[0001] The present invention belongs to the field of radar technology, and in particular relates to a method for intelligently extracting timing parameters of radar composite interference based on frequency agility. Background Art

[0002] In today's complex and ever-changing electromagnetic environment, radar systems are facing severe challenges from new jamming threats centered around digital radio frequency memory (DRFM) technology. Interrupted-sampling repeater jamming (ISRJ) intercepts radar signals and performs intermittent sampling, modulation, and forwarding operations, generating a large number of false targets in the time and frequency domains. This significantly impairs the radar's ability to measure parameters of real targets. In complex ISRJ scenarios, multiple jamming signals couple with each other in the space-time-frequency domain, and their parameter components overlap, making it difficult for traditional parameter estimation methods to effectively separate and accurately extract them. Interference perception, as a key step in the anti-jamming process, has a significant impact on subsequent anti-jamming waveform design and strategy selection. Therefore, overcoming the parameter estimation bottleneck for complex ISRJ has become a core prerequisite for anti-jamming waveform design. Existing ISRJ parameter estimation methods achieve good parameter estimation results in single jamming scenarios, but effective and in-depth research on complex ISRJ parameter extraction remains lacking. The few existing methods mostly involve complex time-frequency domain computations and are not ideal when operating with low interference-to-noise ratios and a limited number of training samples. Summary of the Invention

[0003] To address the above-mentioned problems in the prior art, the present invention provides a method for intelligently extracting timing parameters of radar composite interference based on frequency agility. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0004] The present invention provides a method for intelligently extracting timing parameters of radar composite interference based on frequency agility, comprising:

[0005] Step 1: Preprocessing a radar echo signal to obtain a preprocessed echo signal, wherein the radar echo signal is a frequency agile radar echo signal including a composite ISRJ;

[0006] Step 2: Input the preprocessed echo signal into the trained composite ISRJ separation network to obtain multiple separated ISRJ component signals, wherein the composite ISRJ separation network is an interference separation network that integrates convolution enhancement and Transformer attention mechanism;

[0007] Step 3: Perform dynamic threshold calculation on each ISRJ component signal, and perform binarization processing on each ISRJ component signal according to the dynamic threshold calculation result to obtain the binarized ISRJ signal corresponding to each ISRJ component signal;

[0008] Step 4: Use the clustering optimization algorithm to perform clustering optimization processing on each binary ISRJ signal, and obtain the estimated value of the interference parameter of each ISRJ component signal based on the clustering optimization result.

[0009] Compared with the prior art, the present invention has the following beneficial effects:

[0010] The present invention's frequency-agile, intelligent extraction method for radar composite interference time series parameters utilizes an interference separation network that integrates convolutional enhancement and a Transformer attention mechanism to separate and process radar echo signals. This interference separation network leverages convolution operations to enhance local feature extraction capabilities and introduces a cross-scale attention mechanism to capture the global correlation of interference signals, thereby achieving precise separation of multi-component composite ISRJ signals into multiple separated ISRJ component signals. Furthermore, through dynamic threshold detection and a robustness measurement algorithm based on pulse detection and clustering optimization, accurate measurement of interference parameters such as slice width, forward width, and forward count of ISRJ component signals is achieved based on the time domain characteristics of the ISRJ component signals.

[0011] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following preferred embodiments are specifically cited and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Schematic diagram of a method for intelligently extracting timing parameters of radar composite interference based on frequency agility provided by an embodiment of the present invention;

[0013] Figure 2 This is a flowchart illustrating an intelligent method for extracting timing parameters of radar composite interference based on frequency agility, provided by an embodiment of the present invention;

[0014] Figure 3 This is an example diagram of a time domain model of a composite ISRJ signal provided by an embodiment of the present invention;

[0015] Figure 4 This is an example diagram of a time-frequency domain model of a composite ISRJ signal provided by an embodiment of the present invention;

[0016] Figure 5 Schematic diagram of the structure of a composite ISRJ separation network provided by an embodiment of the present invention;

[0017] Figure 6 is a schematic diagram of the structure of an attention block provided by an embodiment of the present invention;

[0018] Figure 7 is a schematic structural diagram of a one-dimensional convolutional layer provided by an embodiment of the present invention;

[0019] Figure 8 This is a comparison chart of the parameter estimation accuracy of the method proposed in the present invention under different JNR conditions under dual interference conditions;

[0020] Figure 9 This is a comparison chart of the parameter estimation accuracy of the method proposed in the present invention under three interference conditions and different JNR conditions;

[0021] Figure 10 This is a comparison chart of the parameter estimation accuracy of the method proposed in the present invention under the conditions of double interference and different training set sizes;

[0022] Figure 11 This is a comparison chart of the parameter estimation accuracy of the method proposed in the present invention under three interference conditions and different training set sizes. DETAILED DESCRIPTION

[0023] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following is a detailed description of an intelligent extraction method of radar composite interference timing parameters based on frequency agility proposed in accordance with the present invention, in combination with the accompanying drawings and specific implementation methods.

[0024] The aforementioned and other technical contents, features, and effects of the present invention are clearly presented in the following detailed description of the specific embodiments in conjunction with the accompanying drawings. Through the description of the specific embodiments, a deeper and more specific understanding of the technical means and effects adopted by the present invention to achieve the intended purpose can be obtained. However, the accompanying drawings are provided for reference and illustration purposes only and are not intended to limit the technical solutions of the present invention.

[0025] The embodiment of the present invention provides a method for intelligently extracting timing parameters of radar composite interference based on frequency agility. Figure 1 , Figure 1 Schematic diagram of an intelligent extraction method of radar composite interference timing parameters based on frequency agility provided by an embodiment of the present invention, such as Figure 1 As shown, the method for intelligently extracting timing parameters of radar composite interference based on frequency agility of this embodiment includes the following steps:

[0026] Step 1: Preprocessing the radar echo signal to obtain a preprocessed echo signal, where the radar echo signal is a frequency agile radar echo signal including a composite ISRJ;

[0027] Step 2: Input the preprocessed echo signal into the trained composite ISRJ separation network to obtain multiple separated ISRJ component signals. The composite ISRJ separation network is an interference separation network that integrates convolutional enhancement and Transformer attention mechanism.

[0028] Step 3: Perform dynamic threshold calculation on each ISRJ component signal, and perform binarization processing on each ISRJ component signal according to the dynamic threshold calculation result to obtain the binarized ISRJ signal corresponding to each ISRJ component signal;

[0029] Step 4: Use the clustering optimization algorithm to perform clustering optimization processing on each binary ISRJ signal, and obtain the estimated value of the interference parameter of each ISRJ component signal based on the clustering optimization result.

[0030] In this embodiment, the interference parameters include slice width, forwarding width, and forwarding times.

[0031] See Figure 2 , Figure 2 FIG. 1 is a flow chart illustrating an intelligent method for extracting timing parameters of radar composite interference based on frequency agility provided by an embodiment of the present invention. Figure 2 As shown in the figure, the intelligent extraction method of radar composite interference time series parameters based on frequency agility of the present invention uses an interference separation network that integrates convolution enhancement and Transformer attention mechanism, namely a composite ISRJ separation network, to separate and process radar echo signals, namely composite ISRJ signals. The interference separation network uses convolution operation to enhance the local feature extraction capability and introduces a cross-scale attention mechanism to capture the global correlation of interference signals, thereby achieving accurate separation of multi-component composite ISRJ signals to obtain multiple separated ISRJ component signals. On this basis, through dynamic threshold detection and using a robustness measurement algorithm based on pulse detection and clustering optimization, the slice width of the ISRJ component signal is achieved according to the time domain characteristics of the ISRJ component signal. , forwarding width Accurate measurement of interference parameters such as number of forwarding times.

[0032] Furthermore, the method for intelligently extracting timing parameters of radar composite interference based on frequency agility of this embodiment is described in detail.

[0033] First, the frequency-agile radar echo including the composite ISRJ When the jamming system deploys multiple jammers, the received radar echo signal contains ISRJ component signals with different jamming parameters, which randomly overlap in the time domain and frequency domain. At this time, the received radar echo signal can be expressed as:

[0034] ;

[0035] in, is the received radar echo signal, is the amplitude of the radar echo signal, is the target echo signal, Indicates the ISRJ component signals, Indicates time, and Represent the target signal and The delay of the ISRJ component signal, is the total number of ISRJ component signals included, The mean is 0 and the variance is Gaussian noise.

[0036] See Figure 3 , Figure 3 This is an example diagram of a time domain model of a composite ISRJ signal provided by an embodiment of the present invention, such as Figure 3 As shown in the figure, it contains a noisy target signal and three ISRJ component signals (ISRJ1, ISRJ2, and ISRJ3). Each ISRJ component signal is characterized by different interference parameters: slice width, forward width, and forward count. The composite ISRJ signal composed of these overlapping components is the received radar echo signal.

[0037] See Figure 4 , Figure 4 This is an example diagram of a time-frequency domain model of a composite ISRJ signal provided by an embodiment of the present invention, such as Figure 4 As shown in the figure, it can be seen that each ISRJ component signal is highly overlapped in the time domain and time-frequency domain. At the same time, the influence of noise makes it difficult to directly measure interference parameters through envelope detection, time-frequency filtering, etc.

[0038] In an optional embodiment, step 1 includes: performing maximum normalization processing on the radar echo signal, compressing the radar echo signal to a dynamic range of [0, 1], and obtaining a preprocessed echo signal.

[0039] In this embodiment, the radar echo signal can be converted into an input format suitable for the composite ISRJ separation network by performing maximum value normalization processing on the radar echo signal.

[0040] It should be noted that, in order to simplify processing and improve the accuracy of interference signal separation, this embodiment treats the frequency-agile radar echo signal and random noise in the received signal as one signal component, and treats each ISRJ component signal as a separate signal component. Therefore, the received time domain signal can be simplified as:

[0041] ;

[0042] in, Represents the superposition component of target echo and noise, that is, target signal, Indicates the ISRJ component signals, is the number of sampling points contained in a pulse repetition interval, is the total number of ISRJ component signals included.

[0043] In this embodiment, the composite ISRJ separation network includes: an encoder, a mask network and a decoder, wherein the encoder is used to extract features from the preprocessed echo signal to obtain a coding feature sequence; the mask network is used to implement nonlinear mapping of the coding feature sequence to multiple groups of time domain masks to obtain multiple mask vectors; the decoder is used to perform a deconvolution operation on the Kronecker product of the multiple mask vectors and the coding feature sequence to obtain multiple ISRJ component signals.

[0044] For a radar echo signal containing a composite ISRJ, it can be estimated based on the deep learning model Independent ISRJ component signals , T It is a time series. The interference separation network proposed in this embodiment continues the paradigm architecture of encoder-mask network-decoder. In view of the limitations of traditional time-domain convolutional networks in modeling long-range temporal dependencies, this embodiment innovatively constructs a mask network based on the hierarchical Transformer self-attention mechanism. The mask network achieves global context perception through a multi-head self-attention mechanism, and combines the gated convolution module to enhance the local feature extraction capability, forming a hybrid architecture with the advantages of both global and local modeling, which effectively solves the problem of long-term temporal correlation attenuation caused by the limitation of local receptive field of traditional convolutional networks.

[0045] See Figure 5 , Figure 5 FIG. 1 is a schematic diagram of a composite ISRJ separation network provided by an embodiment of the present invention. Figure 5 As shown in FIG, in this embodiment, the encoder includes a cascade of the first one-dimensional convolution layer (1-D Conv) and the ReLU layer. The encoder is used to process the input sequence and map it to a higher-dimensional space to provide important feature information for subsequent processing. Assume that the size of the encoder convolution kernel is , the step size is , the number of filters is , input sequence ,in b is the batch size, T is a time series. After passing through the encoder, the encoded feature sequence can be expressed as:

[0046] ;

[0047] in, , is the time series after the encoder, , Represents the ReLU (Rectified Linear Unit, linear rectification) activation function, Represents the input sequence Perform a one-dimensional convolution operation. To simplify the subsequent derivation, the batch size is omitted. b .

[0048] In this embodiment, the mask network includes a normalization layer (LayerNorm), a position encoding layer, a first point-wise convolutional layer (1-1 Conv), multiple attention blocks, a ReLU layer, a second point-wise convolutional layer (1-1 Conv), and a Sigmoid layer. In the multiple attention blocks, the output of the previous attention block serves as the input of the next attention block. The encoded feature sequence is passed through the normalization layer and the position encoding layer to obtain a feature sequence that retains global temporal information. The feature sequence that retains global temporal information is then passed to the multiple attention blocks via the first point-wise convolutional layer. The multiple attention blocks utilize a self-attention mechanism to capture long-range dependencies in the input feature sequence. The feature sequence output by the last attention block is dimensionally expanded through a ReLU layer and a second point-wise convolutional layer. The dimensionally expanded feature sequence is then mapped through a Sigmoid layer to obtain multiple mask vectors.

[0049] The role of the mask network is to realize the nonlinear mapping of the encoder output to multiple sets of time domain masks. The encoding feature sequence output by the encoder is ( N The feature dimension is consistent with the number of filters, L The time series after the encoder is first normalized and positionally encoded to preserve global temporal information, and then passed to the attention block for sequential processing through point-wise convolution. The attention block combines convolution operations with the attention mechanism to achieve a hybrid process of local feature extraction and global relationship modeling. The attention block uses a self-attention mechanism to capture long-range dependencies. Finally, the attention block is connected through residuals for better training. The output of the current attention block serves as the input of the next attention block, and the process is repeated. Finally, the output of the attention block passes through the ReLU activation function and the point-wise convolution layer, which changes the dimension of the sequence from Expand to ,for C ISRJ component signal estimation CThe sigmoid activation function is then used to map the size of the mask vector to a value between 0 and 1. These values between 0 and 1 represent the confidence that each dimension of the mask vector belongs to a different ISRJ component signal.

[0050] See Figure 6 , Figure 6 is a structural diagram of an attention block provided by an embodiment of the present invention, such as Figure 6 As shown, the attention block of this embodiment adopts a hybrid architecture that integrates one-dimensional convolution (1-DConv) and attention mechanism, aiming to collaboratively extract local features and model global dependencies. The one-dimensional convolution layer adopts Figure 7 The following is a schematic diagram of the structure of a one-dimensional convolutional layer. This one-dimensional convolutional layer enhances feature expression and gradient flow while maintaining computational efficiency. The attention design of this embodiment is based on the Transform architecture and combines the gated attention unit (GAU) for long sequence modeling to enhance the processing capability of complex ISRJ signals. The following is the core processing flow of this module:

[0051] like Figure 6 As shown in the figure, for the input sequence of the attention block, two one-dimensional convolutional layers (1-DConv) are first used to perform preliminary feature extraction. The first convolution focuses on capturing local features (such as short-time spectral features). The convolution output is then distributed and adjusted (similar to layer normalization) through scaling and offset operations to ensure feature stability. Rotational Position Encoding (RoPE) is then introduced to embed absolute or relative position information into the feature vector to address the lack of position sensitivity in the attention mechanism. The query matrix is then generated based on the standardized features. Q With the key matrix K , and obtain the attention matrix through improved attention calculation , , is the query matrix, is the bond matrix, is a learnable bias term, The rectified linear unit (RLU) represents squared activation. This design improves the ability of attention weights to focus on salient features through nonlinear enhancement and squaring operations. Finally, the attention output is processed by a second one-dimensional convolutional layer (1-DConv), achieving a deep fusion of local features and global context. Simultaneously, a residual network is used to generate a discriminative high-dimensional representation in the final output layer, providing an optimized feature space for interference separation tasks.

[0052] Please continue to see Figure 5 In this embodiment, the decoder includes a second one-dimensional convolutional layer (1-D Conv), which is a transposed convolutional layer of the first one-dimensional convolutional layer.

[0053] In the decoding stage, in order to restore the separated feature sequence to the time domain signal, the transposed convolution layer (also called deconvolution layer) is used as the core component of the decoder. The decoder is designed symmetrically with the encoder and uses the same stride and kernel size as the encoder to ensure that the features can be effectively reverse mapped back to the original input space. Each mask vector is multiplied element by element to obtain a separated feature sequence, which is finally decoded into a time domain signal by the decoder, that is, multiple ISRJ component signals are obtained.

[0054] It should be noted that the data samples used in the training process of the composite ISRJ separation network of this embodiment can be obtained in the following manner: collecting echo signals, which are echo signals containing composite ISRJ, that is, including interference signals and target signals; independently performing maximum normalization processing on each echo signal, and uniformly compressing the echo signal and the interference signal to the dynamic range of [0,1]; stacking the interference signal and the target signal along the channel dimension to construct a multi-channel signal matrix of the target signal containing the interference signal, and the multi-channel signal matrix serves as the label of the echo signal; using dynamic zero-value filling technology to align sequences of different lengths, converting them into PyTorch tensors and adjusting the dimensional order to form an input format suitable for the deep learning model.

[0055] In an optional embodiment, step 3 includes:

[0056] Step 3.1: Based on the preset window length, calculate the dynamic threshold of each ISRJ component signal. The calculation formula of the dynamic threshold is:

[0057] ;

[0058] ;

[0059] Where, is the dynamic threshold, For time, is the sensitivity coefficient, is the local statistic of the ISRJ component signal within the window length, represents the median operation, is the window length, Represents the ISRJ component signal within the window length.

[0060] It is understandable that, since the interference energies of different ISRJ component signals are different, this embodiment designs a dynamic threshold detection method based on interference energy, which tracks amplitude changes through local statistics and avoids the prior dependence of a fixed threshold.

[0061] Step 3.2: Compare each ISRJ component signal with the corresponding dynamic threshold, and obtain the binary ISRJ signal corresponding to each ISRJ component signal according to the comparison result. The binary ISRJ signal is calculated according to the following formula:

[0062] ;

[0063] Where, Represents the binary ISRJ signal, where Indicates that the signal is in forwarding state. Indicates that the signal is in the sliced state.

[0064] In an optional embodiment, step 4 includes:

[0065] Step 4.1: Obtain a slice width set and a forwarding width set corresponding to each ISRJ component signal according to each binarized ISRJ signal.

[0066] In this embodiment, by traversing the binary ISRJ signal and recording the width (number of sampling points) of each forwarding (interval of consecutive 1s) and slice (interval of consecutive 0s), the slice width set and forwarding width set corresponding to the ISRJ component signal can be obtained.

[0067] Step 4.2: Filter the slice width set and forwarding width set using a greedy clustering algorithm to obtain a dominant slice width set and a dominant forwarding width set corresponding to each ISRJ component signal.

[0068] In this embodiment, in order to suppress the influence of noise and outliers, the similarity metric is defined as , are two different data values in the set to be filtered, where is the similarity threshold, which can be , if the absolute difference between two data points is less than or equal to the threshold , they are considered similar, otherwise they are not similar. The slice width set and forwarding width set are screened by greedy clustering algorithm to obtain the dominant slice width set and dominant forwarding width set.

[0069] Step 4.3: Density clustering optimization is performed on the dominant slice width set and the dominant forwarding width set respectively to obtain an estimated value of the slice width and an estimated value of the forwarding width corresponding to each ISRJ component signal.

[0070] In this embodiment, the dominant slice width set Find a value in , so that is the center and the radius is The number of samples in the neighborhood of The value of is used as an estimate of the slice width and can be expressed as:

[0071] ;

[0072] Where, represents an estimate of the slice width, represents the set of dominant slice widths, Represents the similarity threshold, here we take , The dominant slice width set The two different data values in .

[0073] In this embodiment, the dominant forwarding width set Perform density clustering, remove clusters with sample numbers less than the threshold (such as single-sample clusters), and take the average of the remaining samples to obtain an estimated forwarding width. , which can be expressed as:

[0074] ;

[0075] Where, represents the estimated value of the forward width, Represents the dominant forwarding width set The number of data in The dominant forward width set The data after removing outliers.

[0076] Step 4.4: Based on the estimated value of the slice width and the estimated value of the forwarding width, calculate and obtain the estimated value of the number of forwarding times corresponding to each ISRJ component signal.

[0077] In this embodiment, based on the characteristics of intermittent sampling forwarding interference, the estimated value of the number of forwarding times is calculated according to the following formula:

[0078] ;

[0079] Where, Indicates an estimate of the number of retweets.

[0080] The single ISRJ component signal obtained through the composite ISRJ separation network can effectively suppress the influence of noise, but there will be fragmentation of time domain slices. Therefore, in this embodiment, based on the characteristics of the separated ISRJ component signal, a robustness measurement algorithm based on pulse detection and clustering optimization is proposed. This method integrates adaptive dynamic threshold detection, multi-level pulse segmentation and clustering optimization algorithm, and can achieve accurate solution of interference parameters based on the characteristics of the separated ISRJ component signal.

[0081] In order to verify the effectiveness and stability of the intelligent extraction method of composite interference timing parameters proposed in the present invention, a series of complex interference scenario simulation experiments under different interference-to-noise ratio conditions were set up for verification.

[0082] (1) Dataset

[0083] To ensure a close approximation to actual conditions, the interference parameters for each ISRJ component in the composite ISRJ signal were randomly generated, with the JNR controlled to randomly jump between 0 and 20 dB. Matlab was used to generate dual- and triple-interference datasets, with 5,000 entries used for training, 1,000 for validation, and 1,000 for testing. The specific parameter configurations are shown in Table 1.

[0084] Table 1

[0085]

[0086] (2) Training settings

[0087] We use the SI-SDR training loss for optimization, a loss function widely used to evaluate source separation tasks. We train the model using the Adam optimizer for 100 epochs, a learning rate of 1e-3, a batch size of 8, and gradient clipping to limit the norm of the training gradient to 5.

[0088] (3) Evaluation indicators

[0089] The measurement index is measured by the ISRJ interference parameter measurement accuracy, that is, the slice width measurement accuracy , Forward Width Accuracy , and the accuracy of forwarding times The slice width and forwarding width errors are 0.15 respectively. and 0.25 In addition, in order to measure the overall measurement effect, we define , represents the proportion of samples in which all interference parameters are accurately measured.

[0090] (4) Comparative test

[0091] Experiment ① completes the separation and measurement of compound ISRJ signals in two and three ISRJ scenarios. The simulation experimental test results of the proposed method are compared with those in the literature [1] (L. Liu, Q. Lv, J. Liu, Y. Wu, and Y. Quan, “Adaptive Parameter Estimation for Compound Interrupted Sampling and Repeating Jamming,” IEEE Signal Process. Lett., vol. 31, pp. 496–500, 2024) and the literature [2] (Q. Lv, L. Liu, Y. Wu, M. Peng, and Y. Quan, “A Framework for Compound Interrupted Sampling Repeater Jamming Parameter Measurement,” IEEE Trans. Instrum. Meas., vol. 74, pp. 1–4, 2025).

[0092] Experiment ② analyzes the performance advantages of the proposed method in measuring interference parameters under different JNR conditions, including double interference and triple interference scenarios.

[0093] Experiment ③ analyzes the performance advantages of the method proposed in this invention under different training set ratios.

[0094] ①Interference parameter extraction experiment of composite ISRJ signal

[0095] The proposed method is compared with the ISRJ separation network DCPE-ISRJ in [1] and the ISRJ separation network CSNet-ISRJ in [2]. All three methods are tested in two scenarios: dual interference and triple interference. Table 2 shows the test results.

[0096] Table 2

[0097]

[0098] The results show that compared with DCPE-ISRJ, the proposed method achieves 215.8% of the accuracy in dual interference scenarios. Performance improvement, achieving 1183% in three interference scenarios This is mainly because DCPE-ISRJ can only process interference signals with the same JNR, while the proposed method can process interference signals with random JNR. Compared with CSNet-ISRJ, it achieves 11.3% improvement in dual interference scenarios. Performance improvement, achieving 7.9% in three-interference scenario Performance improvements.

[0099] ②Performance comparison under different JNRs

[0100] To comprehensively evaluate the performance of the proposed method, Monte Carlo simulation experiments were conducted under different interference-to-noise ratio (JNR) conditions. In the experimental scenario, the signal-to-noise ratio (SNR) was fixed at 0 dB, and the JNR value range was set to 5 dB, 10 dB, 15 dB, and 20 dB. Tests were conducted in dual-interference and multi-interference scenarios, and compared with the method proposed in reference [2].

[0101] See Figure 8 and Figure 9 , Figure 8 This is a comparison chart of the parameter estimation accuracy of the method proposed in this invention under dual interference conditions and different JNR conditions. Figure 9 This is a comparison chart of the parameter estimation accuracy of the method proposed in the present invention under three interference conditions and different JNR conditions. Figure 8 and Figure 9 As shown in the figures, in both double-interference and triple-interference scenarios, the proposed method shows significant performance improvement under low JNR conditions, the parameter measurement accuracy reaches more than 90%, and the system stability is excellent.

[0102] ③The impact of different training set sizes

[0103] In order to evaluate the effect of training set size on model performance, the model was trained by using sample sizes gradually increasing from 1000 to 5000, and its corresponding performance indicators were tested. In addition, the obtained results were compared with the method proposed in the literature [2]. The specific comparison results are as follows: Figure 10 and Figure 11 As shown, Figure 10 This is a comparison chart of the parameter estimation accuracy of the method proposed in the present invention under the conditions of double interference and different training set sizes; Figure 11 The figure shows a comparison of the parameter estimation accuracy of the proposed method under three interference conditions and different training set sizes. As can be seen from the figure, in the face of double interference and triple interference scenarios, as the size of the training dataset increases, the performance of the proposed method and the method in reference [2] shows a steady improvement trend. However, it is worth noting that when the training sample set is small, the proposed method shows higher accuracy than the method in reference [2].

[0104] The proposed method only requires a training set of 2,000 samples to achieve performance comparable to that achieved by using a training set of 5,000 samples in [2]. This demonstrates that the proposed method is not only more effective on small datasets, but can also significantly reduce the number of training samples required to achieve ideal model performance.

[0105] It should be noted that, in this document, relational terms such as first and second are used solely to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. Furthermore, the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not explicitly listed. Without further limitation, an element defined by the phrase "comprising a..." does not preclude the presence of additional identical elements in the article or device comprising the element. Terms such as "connected" or "connected" are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. References to orientations or positional relationships, such as "upper," "lower," "left," and "right," are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate description and simplify the description of the present invention. They do not indicate or imply that the device or element referred to must have, be constructed, or operate in a specific orientation, and are therefore not to be construed as limiting the present invention.

[0106] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0107] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for intelligently extracting timing parameters of radar composite interference based on frequency agility, characterized in that: include: Step 1: Preprocessing a radar echo signal to obtain a preprocessed echo signal, wherein the radar echo signal is a frequency agile radar echo signal including a composite ISRJ; Step 2: Input the preprocessed echo signal into the trained composite ISRJ separation network to obtain multiple separated ISRJ component signals, wherein the composite ISRJ separation network is an interference separation network that integrates convolution enhancement and Transformer attention mechanism; Step 3: performing dynamic threshold calculation on each ISRJ component signal, and performing binarization processing on each ISRJ component signal according to the dynamic threshold calculation result to obtain a binarized ISRJ signal corresponding to each ISRJ component signal; Step 3 includes: Step 3.1: Calculate the dynamic threshold of each ISRJ component signal based on the preset window length. The calculation formula of the dynamic threshold is: ； ； Where, is the dynamic threshold, For time, is the sensitivity coefficient, is the local statistic of the ISRJ component signal within the window length, represents the median operation, is the window length, Represents the ISRJ component signal within the window length; Step 3.2: Compare each ISRJ component signal with the corresponding dynamic threshold, and obtain a binary ISRJ signal corresponding to each ISRJ component signal according to the comparison result. The binary ISRJ signal is calculated according to the following formula: ； Where, Represents the binary ISRJ signal; Step 4: Use the clustering optimization algorithm to perform clustering optimization processing on each binary ISRJ signal, and obtain the estimated value of the interference parameter of each ISRJ component signal based on the clustering optimization result.

2. The method for intelligently extracting timing parameters of radar composite interference based on frequency agility according to claim 1, characterized in that: The step 1 comprises: Performing maximum normalization processing on the radar echo signal, compressing the radar echo signal to a [0, 1] dynamic range, and obtaining the preprocessed echo signal.

3. The method for intelligently extracting timing parameters of radar composite interference based on frequency agility according to claim 1, characterized in that: The composite ISRJ separation network includes: an encoder, a mask network and a decoder, wherein: The encoder is used to extract features from the preprocessed echo signal to obtain a coding feature sequence; The mask network is used to implement nonlinear mapping of the coding feature sequence to multiple groups of time domain masks to obtain multiple mask vectors; The decoder is used to perform a deconvolution operation on the Kronecker product of the multiple mask vectors and the encoding feature sequence to obtain the multiple ISRJ component signals.

4. The method for intelligently extracting timing parameters of radar composite interference based on frequency agility according to claim 3, characterized in that: The encoder includes a cascade of a first one-dimensional convolutional layer and a ReLU layer.

5. The method for intelligently extracting timing parameters of radar composite interference based on frequency agility according to claim 3, characterized in that: The mask network includes a normalization layer, a position encoding layer, a first point-wise convolution layer, multiple attention blocks, a ReLU layer, a second point-wise convolution layer and a Sigmoid layer, wherein, Among the multiple attention blocks, the output of the previous attention block serves as the input of the next attention block; The encoded feature sequence is passed through the normalization layer and the position encoding layer to obtain a feature sequence retaining global temporal information, and the feature sequence retaining global temporal information is passed to the multiple attention blocks through the first point convolution layer; The multiple attention blocks use the self-attention mechanism to capture the long-range dependencies in the input feature sequence. The feature sequence output by the last attention block is dimensionalized through the ReLU layer and the second point-to-point convolution layer. The feature sequence after dimensional expansion is mapped through the Sigmoid layer to obtain the multiple mask vectors.

6. The method for intelligently extracting timing parameters of radar composite interference based on frequency agility according to claim 4, characterized in that: The decoder includes a second one-dimensional convolutional layer, which is a transposed convolutional layer of the first one-dimensional convolutional layer.

7. The method for intelligently extracting timing parameters of radar composite interference based on frequency agility according to claim 1, characterized in that: The interference parameters include slice width, forwarding width and forwarding times.

8. The method for intelligently extracting timing parameters of radar composite interference based on frequency agility according to claim 7, characterized in that: The step 4 comprises: Step 4.1: Obtain a slice width set and a forwarding width set corresponding to each ISRJ component signal according to each binarized ISRJ signal; Step 4.2: Screening the slice width set and the forwarding width set by a greedy clustering algorithm to obtain a dominant slice width set and a dominant forwarding width set corresponding to each ISRJ component signal; Step 4.3: performing density clustering optimization on the dominant slice width set and the dominant forwarding width set respectively to obtain an estimated value of the slice width and an estimated value of the forwarding width corresponding to each ISRJ component signal; Step 4.4: Calculate an estimated value of the number of forwarding times corresponding to each ISRJ component signal based on the estimated value of the slice width and the estimated value of the forwarding width.

9. The method for intelligently extracting timing parameters of radar composite interference based on frequency agility according to claim 8, characterized in that: The estimated value of the slice width is calculated according to the following formula: ； Where, represents an estimate of the slice width, represents the set of dominant slice widths, represents the similarity threshold, The dominant slice width set The two different data values in ; The estimated value of the forward width is calculated according to the following formula: ； Where, represents the estimated value of the forward width, Represents the dominant forwarding width set The number of data in The dominant forward width set The data after removing outliers; The estimated value of the number of forwarding times is calculated according to the following formula: ； Where, Indicates an estimate of the number of retweets.

Images