Radar active jamming identification method based on time-frequency analysis and deep network

Abstract

The invention discloses a radar active interference identification method based on time-frequency analysis and a deep network, and belongs to the field of radar active interference identification, and the method comprises the steps: constructing radar active interference signals of different interference types, and obtaining interference time-frequency signals corresponding to all kinds of radar active interference signals; matching filtering responses corresponding to various radar active interference signals are obtained; synthesizing the various interference time-frequency signals and the various matched filtering responses based on interference types to obtain various interference synthesis signals; training the ResNet network based on various interference synthesis signals to obtain an interference identification classification model; and obtaining a to-be-identified signal, and obtaining an interference type of the to-be-identified signal by using the interference identification classification model to complete radar active interference identification. According to the invention, the problem that the neural network cannot accurately identify the radar interference type and cannot identify the mixing effect of various interferences because the background noise is too large to submerge the identified interference signal is solved.

Description

Technical Field

[0001] The present invention belongs to the field of radar active interference identification, and in particular relates to a radar active interference identification method based on time-frequency analysis and deep network. Background Art

[0002] Active jammer detection is a crucial step in radar's ability to counter electromagnetic interference in complex electromagnetic environments. In recent years, a large number of new jammer patterns have emerged, significantly restricting radar detection performance. To ensure radar can fully utilize its capabilities in complex electromagnetic environments, it is essential to improve radar's ability to detect active jammers.

[0003] To ensure reliable radar target detection and tracking, numerous anti-interference technologies have emerged. These technologies primarily involve two steps: identification and suppression. Identifying interference is a prerequisite for suppression. Traditional radar active jammer identification methods rely on manual selection of feature parameters, lacking universal and persuasive single or multi-dimensional features. Any deviation in the selected features or classifier design can lead to reduced recognition rates. Summary of the Invention

[0004] To address the aforementioned shortcomings of the existing technology, the present invention provides a radar active jammer identification method based on time-frequency analysis and deep neural networks. This method addresses the problem of excessive background noise drowning out the jammer signal, which can lead to the neural network's inability to accurately identify the type of radar jammer and the inability to distinguish between mixed effects of multiple jammers.

[0005] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a radar active interference identification method based on time-frequency analysis and deep network, comprising: Construct radar active jamming signals of different jamming types, and perform short-time Fourier transform on various radar active jamming signals to obtain the jamming time-frequency signals corresponding to various radar active jamming signals; Use matched filtering technology to perform pulse compression processing on various radar active jamming signals to obtain the matched filtering responses corresponding to various radar active jamming signals; Various interference time-frequency signals and various matched filter responses are synthesized based on the interference type to obtain various interference synthesis signals; Based on various types of interference synthesis signals, the ResNet network is trained to obtain an interference recognition and classification model; Obtain the signal to be identified, use the interference identification classification model to obtain the interference type of the signal to be identified, and complete the radar active interference identification.

[0006] The beneficial effects of the present invention are as follows: due to the suppression effect of matched filtering itself on background noise, this method can achieve good recognition effect even when the signal to be recognized is relatively small; at the same time, an interference allocation mechanism and an identification feedback mechanism are introduced when training the ResNet network, and different types of interference signals are synthesized based on the interference allocation mechanism, so that the obtained interference recognition and classification model has the ability to recognize multiple interference types in scenarios.

[0007] Furthermore, when training the ResNet network, an interference allocation mechanism is introduced into the input layer of the ResNet network, and a recognition feedback mechanism is introduced into the output layer.

[0008] The beneficial effect of the above further scheme is: introducing interference allocation mechanism and recognition feedback mechanism when training the ResNet network, synthesizing different types of interference signals based on the interference allocation mechanism, so that the obtained interference recognition and classification model has the ability to recognize multiple interference types in scenarios.

[0009] Furthermore, the interference coordination mechanism is specifically as follows: The input layer is expanded into several input heads, each of which is connected to an interference allocation unit; the interference allocation unit is connected to the convolutional stem of the ResNet network; the number of input heads is the same as the number of interference types.

[0010] The beneficial effect of the above further solution is: the input layer is expanded into multiple input heads, which correspond one-to-one to the interference types, and can cooperate with the interference allocation mechanism to synthesize different types of interference signals.

[0011] Furthermore, the interference allocation unit is provided with an interference allocation vector:

[0012] in, allocate vectors for interference; For the A binary signal indicating whether the interference synthesis signal participates in the allocation, 1 means participation, 0 means no participation; is the total amount of interference types.

[0013] The beneficial effect of the above further solution is: setting the interference allocation vector to control the opening and closing of the input head and control the participation of the interference synthesis signal.

[0014] Furthermore, the interference allocation unit is configured to synthesize the interference composite signals participating in the allocation based on the interference allocation vector to obtain an integrated signal.

[0015] The beneficial effects of the above further solution are: setting interference allocation vectors and synthesizing integrated signals to realize multiple interference scenarios.

[0016] Furthermore, the identification feedback mechanism is used to coordinate participation control feedback and coordinate vector updates.

[0017] The beneficial effects of the above further scheme are: the two feedback mechanisms are both based on the output probability of the ResNet network, which can be used as a phased verification mechanism during the ResNet network training process to ensure the recognition accuracy of the interference recognition classification model.

[0018] Furthermore, the deployment participation control feedback is specifically as follows: Based on the probability of each interference type predicted by the ResNet network, the interference allocation vector is matched to determine whether both conditions 1 and 2 are met. If so, the allocation participation amount is increased by 1; otherwise, the allocation participation amount remains unchanged. Condition 1: The type probability of each interference composite signal participating in the deployment is greater than the set interference probability; The second condition is that the difference between the maximum probability of each interference type not participating in the allocation and the minimum probability of the interference type participating in the allocation is greater than or equal to the distance threshold.

[0019] The beneficial effects of the above further scheme are: two conditions, one for ensuring the success of recognition, and the other for ensuring the significance of recognition. When the model can clearly identify the interference type, the amount of interference type is increased. In this way, the difficulty increases step by step, which can reduce the difficulty of training in subsequent levels.

[0020] Furthermore, the deployment vector is updated as follows: Determine whether the allocation participation amount increases. If so, randomly select one from the binary signals that are 0 in the current allocation vector and set it to 1. Otherwise, determine whether condition one is met but condition two is not met. If so, set the binary signal corresponding to the interference type with the smallest current predicted probability participating in the allocation to 0, and set the binary signal corresponding to the interference type with the largest current predicted probability not participating in the allocation to 1. Otherwise, the allocation vector remains unchanged.

[0021] The beneficial effect of the above further scheme is: when condition one is met but condition two is not met, it means that the recognition is successful but not significant. At this time, the interference type is replaced to find the distinguishing features of the non-participating type with the highest probability and the participating type with the lowest probability, thereby achieving a significant recognition effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Flow chart of the method of the present invention.

[0023] Figure 2 In the embodiment of the present invention, the original signal is included Time-frequency diagram after the nine signals are synthesized.

[0024] Figure 3 In the embodiment of the present invention, the original signal is included Time-frequency diagram after the nine signals are synthesized.

[0025] Figure 4 The ResNet network structure introduces the interference coordination mechanism and the identification feedback mechanism in the embodiment of the present invention. DETAILED DESCRIPTION

[0026] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0027] like Figure 1 As shown, in one embodiment of the present invention, a radar active interference identification method based on time-frequency analysis and deep network includes: Construct radar active jamming signals of different jamming types, and perform short-time Fourier transform on various radar active jamming signals to obtain the jamming time-frequency signals corresponding to various radar active jamming signals; Use matched filtering technology to perform pulse compression processing on various radar active jamming signals to obtain the matched filtering responses corresponding to various radar active jamming signals; Various interference time-frequency signals and various matched filter responses are synthesized based on the interference type to obtain various interference synthesis signals; Based on various types of interference synthesis signals, the ResNet network is trained to obtain an interference recognition and classification model; Obtain the signal to be identified, use the interference identification classification model to obtain the interference type of the signal to be identified, and complete the radar active interference identification.

[0028] In this example, eight common radar active jammers were first modeled. Short-time Fourier transforms (SFTs) were used as a time-frequency conversion tool to obtain time-frequency plots of the echo signal and its matched filter, transforming the radar signal from the time domain to the time-frequency domain. Experiments were conducted using eight typical linear frequency modulation (LFM) radar active jammers: intermittent sampling and forwarding jammers, spectrum dispersion jammers, slicing jammers, comb spectrum jammers, noise amplitude modulation jammers, noise frequency modulation jammers, noise product jammers, and noise convolution jammers.

[0029] Figure 2 Including the original signal The time-frequency of the nine signals after synthesis (the signal in the signal-to-noise ratio is the interference signal used for identification), Figure 3 Including the original signal Time-frequency diagram after the nine signals are synthesized.

[0030] When the signal-to-noise ratio is high, the time-frequency distribution characteristics of various types of interference and the original signal can be clearly analyzed. When background noise increases and the system signal-to-noise ratio decreases, the time-frequency diagram of the interference signal being identified is submerged in the noise, unable to provide effective details and features for network recognition. In this case, a large number of interference features can still be found in the matched filter response of the interference signal. Therefore, this method can effectively identify and classify interference types when the environmental noise gradually increases.

[0031] In this embodiment, the two mechanisms introduced are only used for training. After the training is completed, the interference recognition and classification model obtained is the ResNet network with the mechanisms removed.

[0032] like Figure 4 As shown in Figure 1, when training the ResNet network, an interference adjustment mechanism is introduced into the input layer of the ResNet network, and a recognition feedback mechanism is introduced into the output layer.

[0033] The interference coordination mechanism is specifically as follows: The input layer is expanded into several input heads, each of which is connected to an interference allocation unit; the interference allocation unit is connected to the convolutional stem of the ResNet network; the number of input heads is the same as the number of interference types.

[0034] The interference allocation unit is provided with an interference allocation vector:

[0035] in, allocate vectors for interference; For the A binary signal indicating whether the interference synthesis signal participates in the allocation, 1 means participation, 0 means no participation; is the total amount of interference types.

[0036] In this embodiment, the binary signal is equivalent to the switch signal of the input head.

[0037] The interference allocation unit is used to synthesize the interference synthesis signals participating in the allocation based on the interference allocation vector to obtain an integrated signal.

[0038] The identification feedback mechanism is used to coordinate participation control feedback and coordinate vector updates.

[0039] In this embodiment, the allocation participation amount is initially 1 and then gradually increases, but the allocation vector is still updated.

[0040] The deployment participation control feedback is specifically: Based on the probability of each interference type predicted by the ResNet network, the interference allocation vector is matched to determine whether both conditions 1 and 2 are met. If so, the allocation participation amount is increased by 1; otherwise, the allocation participation amount remains unchanged. Condition 1: The type probability of each interference composite signal participating in the deployment is greater than the set interference probability; The second condition is that the difference between the maximum probability of each interference type not participating in the allocation and the minimum probability of the interference type participating in the allocation is greater than or equal to the distance threshold.

[0041] In this embodiment, the interference probability decreases as the number of participating types increases, and the distance threshold also changes accordingly.

[0042] Coordinate vector updates, specifically: Determine whether the allocation participation amount increases. If so, randomly select one from the binary signals that are 0 in the current allocation vector and set it to 1. Otherwise, determine whether condition one is met but condition two is not met. If so, set the binary signal corresponding to the interference type with the smallest current predicted probability participating in the allocation to 0, and set the binary signal corresponding to the interference type with the largest current predicted probability not participating in the allocation to 1. Otherwise, the allocation vector remains unchanged.

[0043] In this embodiment, when condition one is met but condition two is not met, it means that the features of the type with the highest probability of participating in the allocation and the type with the lowest probability of not participating in the allocation are similar. In this case, the network needs to learn the features of the type with the lowest probability of not participating in the allocation to distinguish between the two.

Claims

1. A radar active jammer identification method based on time-frequency analysis and deep network, characterized in that: include: Construct radar active jamming signals of different jamming types, and perform short-time Fourier transform on various radar active jamming signals to obtain the jamming time-frequency signals corresponding to various radar active jamming signals; Use matched filtering technology to perform pulse compression processing on various radar active jamming signals to obtain the matched filtering responses corresponding to various radar active jamming signals; Various interference time-frequency signals and various matched filter responses are synthesized based on the interference type to obtain various interference synthesis signals; Based on various types of interference synthesis signals, the ResNet network is trained to obtain an interference recognition and classification model; Obtain the signal to be identified, use the interference identification classification model to obtain the interference type of the signal to be identified, and complete the radar active interference identification.

2. The radar active interference identification method based on time-frequency analysis and deep network according to claim 1 is characterized in that: When training the ResNet network, an interference allocation mechanism is introduced in the input layer of the ResNet network, and a recognition feedback mechanism is introduced in the output layer.

3. The radar active interference identification method based on time-frequency analysis and deep network according to claim 1 is characterized in that: The interference coordination mechanism is specifically as follows: The input layer is expanded into several input heads, each of which is connected to an interference allocation unit; the interference allocation unit is connected to the convolutional stem of the ResNet network; the number of input heads is the same as the number of interference types.

4. The radar active interference identification method based on time-frequency analysis and deep network according to claim 3 is characterized in that: The interference allocation unit is provided with an interference allocation vector: in, allocate vectors for interference; For the A binary signal indicating whether the interference synthesis signal participates in the allocation, 1 means participation, 0 means no participation; is the total amount of interference types.

5. The radar active interference identification method based on time-frequency analysis and deep network according to claim 4 is characterized in that: The interference allocation unit is used to synthesize the interference synthesis signals participating in the allocation based on the interference allocation vector to obtain an integrated signal.

6. The radar active interference identification method based on time-frequency analysis and deep network according to claim 1 is characterized in that: The identification feedback mechanism is used to coordinate participation control feedback and coordinate vector updates.

7. The radar active interference identification method based on time-frequency analysis and deep network according to claim 6 is characterized in that: The deployment participation control feedback is specifically: Based on the probability of each interference type predicted by the ResNet network, the interference allocation vector is matched to determine whether both conditions 1 and 2 are met. If so, the allocation participation amount is increased by 1; otherwise, the allocation participation amount remains unchanged. Condition 1: The type probability of each interference composite signal participating in the deployment is greater than the set interference probability; The second condition is that the difference between the maximum probability of each interference type not participating in the allocation and the minimum probability of the interference type participating in the allocation is greater than or equal to the distance threshold.

8. The radar active interference identification method based on time-frequency analysis and deep network according to claim 7 is characterized in that: Coordinate vector updates, specifically: Determine whether the allocation participation amount increases. If so, randomly select one from the binary signals that are 0 in the current allocation vector and set it to 1. Otherwise, determine whether condition one is met but condition two is not met. If so, set the binary signal corresponding to the interference type with the smallest current predicted probability participating in the allocation to 0, and set the binary signal corresponding to the interference type with the largest current predicted probability not participating in the allocation to 1. Otherwise, the allocation vector remains unchanged.

Images