Radar target and jamming signal separation method based on lightweight neural network

Abstract

The present application provides a radar target and interference signal separation method based on a lightweight neural network, which solves the problem that the existing technology cannot extract reliable pure target signals and directly obtain target information by post-processing, and has low feasibility and poor generalization. The method comprises: preprocessing the time domain echo signal to obtain a pretreated signal; inputting the pretreated signal into a mixed attention mechanism network trained to obtain a separated target echo signal; wherein the mixed attention mechanism network comprises an encoder module, a separation module, an inter-stage connection module and a decoder module; and obtaining specific information of the estimated target according to the separated target echo signal. The method realizes direct processing of time domain signals, has good separation accuracy and calculation efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar signal processing, and in particular to a radar target and interference signal separation method based on a light neural network. BACKGROUND

[0002] With the increasing diversification of modern air operations, the gradual application of unmanned aerial vehicles, unmanned ships and unmanned boats, and the development of various electronic detection means, radar has a significant advantage in terms of long detection distance, not being affected by fog, clouds and rain, having a certain penetration ability, and working all day long. It plays an increasingly irreplaceable role in applications such as electronic reconnaissance, target observation, and anti-unmanned vehicle. Due to the emergence and development of digital radio frequency memory (DRFM), jamming machines can quickly intercept radar transmitted signals, generate flexible and variable jamming signals according to extracted radar technical parameters such as carrier frequency, repetition frequency, and direction of arrival, so that the detection environment of modern radars is becoming increasingly complex. A repeater jammer can make an active deception jamming system copy the collected radar transmitted signals at a sub-second processing speed to produce a high jamming-to-signal ratio false target in the radar echo to deceive the radar. When facing such deceptive jamming as intermittent sampling repeater jamming (ISRJ), traditional time-frequency analysis methods have obvious limitations and difficulties in suppressing interference and extracting radar echo signals. The resolution of short-time Fourier transform (STFT) and wavelet transform in the time-frequency domain usually has a certain trade-off between frequency resolution and time resolution, i.e., it is impossible to achieve the best in both at the same time. This resolution limitation may not effectively distinguish between jamming and target echo signals. In addition, most existing radar anti-jamming methods are mainly designed to suppress interference in the echo signal by designing a transmitted waveform with a smaller interference duty cycle in the echo, while methods that directly separate interference cannot extract reliable pure target signals and perform post-processing to directly obtain target information. Moreover, the methods have low feasibility and poor generalization. In the face of complex electromagnetic interference environments with numerous interference patterns and combinations, there is no effective solution to the problem of separating interference and extracting pure target signals. SUMMARY

[0003] The present application provides a radar target and interference signal separation method based on a light neural network, which solves the problem that most methods that directly separate interference in the prior art cannot extract reliable pure target signals and perform post-processing to directly obtain target information, and have low feasibility and poor generalization. The method directly processes time-domain signals, has good separation accuracy and computational efficiency.

[0004] The present application provides a radar target and interference signal separation method based on a light neural network, which solves the problem that most methods that directly separate interference in the prior art cannot extract reliable pure target signals and perform post-processing to directly obtain target information, and have low feasibility and poor generalization. The method directly processes time-domain signals, has good separation accuracy and computational efficiency.

[0005] preprocessing the time-domain echo signal to obtain a preprocessed signal and full features corresponding to the preprocessed signal;

[0006] inputting the preprocessed signal into a mixed attention mechanism network trained to obtain a separated target echo signal; wherein the mixed attention mechanism network comprises an encoder module, a separation module, an inter-level connection module and a decoder module; the encoder module is configured to encode the full features and extract multi-scale features of the preprocessed signal; the separation module is configured to fuse the multi-scale features to obtain fused features, and calculate global attention weights by using a global attention module to obtain a global self-attention graph; the inter-level connection module is configured to recover the precision of the multi-scale features to obtain recovered multi-scale features; and the decoder module is configured to modulate the recovered multi-scale features by using the global self-attention graph to obtain the separated target echo signal;

[0007] obtaining specific information of the target according to the separated target echo signal.

[0008] In a possible implementation, the preprocessing of the time-domain echo signal to obtain the preprocessed signal comprises:

[0009] loading the time-domain echo signal and initializing variables;

[0010] reading a data file in a specified path, loading intermittent sampling and forwarding interference data and target echo data in complex form; wherein the target echo data comprises noise data;

[0011] performing data shape transformation on the intermittent sampling and forwarding interference data and the target echo data respectively to obtain reshaped interference data and reshaped target echo data;

[0012] extracting real and imaginary parts of the reshaped interference data and real and imaginary parts of the reshaped target echo data respectively;

[0013] alternately sorting the real and imaginary parts of the reshaped interference data by using a cyclic interpolation sorting method to obtain a continuous interference data long vector, and alternately sorting the real and imaginary parts of the reshaped target echo data to obtain a continuous target echo data long vector;

[0014] adding the continuous interference data long vector and the continuous target echo data long vector to obtain the preprocessed signal.

[0015] In a possible implementation, the training of the mixed attention mechanism network comprises dynamically adjusting a learning rate of the mixed attention mechanism network according to an evaluation separation index.

[0016] In a possible implementation, the formula for calculating the separation index is represented as:

[0017]

[0018] wherein, represents an estimated target echo signal; s represents a pure source signal; S target represents a target signal part extracted from the estimated signal; e noise represents a noise part, i.e., a difference between the estimated signal and the target signal; s target represents a signal enhancement signal-to-noise ratio, used to evaluate the quality of the target signal.

[0019] In a possible implementation, the encoder module comprises a first deep separable convolution layer, a second deep separable convolution layer, a third deep separable convolution layer, a first average pooling layer, a second average pooling layer, and a third average pooling layer connected in sequence.

[0020] The first deep separable convolution layer is configured to perform feature extraction on the full feature to obtain 1 / 2 scale features.

[0021] The second deep separable convolution layer is configured to perform feature extraction on the 1 / 2 scale features to obtain 1 / 4 scale features.

[0022] The third deep separable convolution layer is configured to perform feature extraction on the 1 / 4 scale features to obtain 1 / 8 scale features.

[0023] The first average pooling layer is configured to perform an average pooling operation on the 1 / 4 scale features to obtain first 1 / 8 scale features.

[0024] The second average pooling layer is configured to perform an average pooling operation on the 1 / 4 scale features to obtain second 1 / 8 scale features.

[0025] The third average pooling layer is configured to perform an average pooling operation on the full feature to obtain third 1 / 8 scale features.

[0026] In a possible implementation, the inter-stage connection module comprises a first upsampling module, a second upsampling module, and a third upsampling module connected in sequence.

[0027] The first upsampling module is configured to perform a two-fold upsampling on a point multiplication result of the fusion feature and the 1 / 8 scale feature according to the global self-attention map, and perform a point multiplication on a sampling result and the 1 / 4 scale feature to obtain first restored multi-scale features.

[0028] The second up-sampling module is configured to perform four times up-sampling on the first recovered multi-scale feature according to the global self-attention map, and the up-sampling result is multiplied by a 1 / 4 scale feature to obtain a second recovered multi-scale feature.

[0029] The third up-sampling module is configured to perform eight times up-sampling on the second recovered multi-scale feature according to the global self-attention map, and the up-sampling result is multiplied by a full feature of the preprocessed signal to obtain a third recovered multi-scale feature; wherein the 1 / 4 scale feature, the 1 / 4 scale feature and the 1 / 8 scale feature are the multi-scale features.

[0030] In a possible implementation, the decoder module comprises a first cascaded local attention layer, a second cascaded local attention layer and a third cascaded local attention layer connected in sequence.

[0031] The first cascaded local attention layer is configured to extract a local attention feature according to the first recovered multi-scale feature and the second recovered multi-scale feature to obtain a first local attention feature.

[0032] The second cascaded local attention layer is configured to extract a local attention feature according to the first local attention feature and the third recovered multi-scale feature to obtain a second local attention feature.

[0033] The third cascaded local attention layer is configured to extract a local attention feature according to the second local attention feature and the full feature of the preprocessed signal to obtain a third local attention feature.

[0034] In a possible implementation, the decoder module is configured to modulate the recovered multi-scale feature pair by using the global self-attention map to obtain a separated target echo signal, comprising:

[0035] Mapping the third local attention feature into the same spatial dimension as the time-domain echo signal to obtain a separated target echo signal.

[0036] In a possible implementation, the specific information of the target is estimated according to the separated target echo signal, comprising:

[0037] Determining the signal type of the time-domain echo signal, and performing pulse compression processing on the separated target echo signal according to the corresponding pulse compression method to obtain a pulse compression signal;

[0038] Performing coherent processing on the pulse compression signal to obtain a coherent processing signal; wherein the coherent processing signal comprises specific information of the target.

[0039] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0040] The application adopts direct processing on time domain signals, without traditional time-frequency domain conversion, so that the processing can abandon the defects such as uncoupling of phase and amplitude of signals, unsatisfactory time-frequency representation of multiple source signal separation, and long delay time for calculating a spectrum diagram, and the full convolution network has better separation accuracy and calculation efficiency for a sequence model;

[0041] For the separated signals, on the one hand, the separated interference signals can be estimated and analyzed to understand the type and specific parameters of the interference, so that a waveform with smaller duty cycle of interference signals can be constructed in subsequent radar transmission waveform design; on the other hand, the relatively clean target echo signals can be processed including pulse compression, coherent, and other processing to estimate the specific position, speed and other information of the target. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A step flow chart of a radar target and interference signal separation method based on a lightweight neural network is provided for an embodiment of the application.

[0043] Figure 2 A mixed attention mechanism network schematic diagram is provided for an embodiment of the application.

[0044] Figure 3 A structure diagram of the mixed attention mechanism network is provided for an embodiment of the application.

[0045] Figure 4a A separation target signal time-frequency diagram is provided for an embodiment of the application, with the interference type set as intermittent sampling interference.

[0046] Figure 4b A separation echo signal restoration diagram is provided for an embodiment of the application, with the interference type set as intermittent sampling interference.

[0047] Figure 4c A separation target signal pulse compression diagram is provided for an embodiment of the application, with the interference type set as intermittent sampling interference.

[0048] Figure 4d A separation target signal coherent diagram is provided for an embodiment of the application, with the interference type set as intermittent sampling interference.

[0049] Figure 5a A separation target signal time-frequency diagram is provided for an embodiment of the application, with the SNR adjusted to -10dBd.

[0050] Figure 5b A separation echo signal restoration diagram is provided for an embodiment of the application, with the SNR adjusted to -10dBd.

[0051] Figure 5cThe SNR adjustment of the separated target signal pulse pressure diagram provided by the embodiment of the present application is -10dBd;

[0052] Figure 5d The SNR adjustment of the separated target signal phase tracking diagram provided by the embodiment of the present application is -10dBd;

[0053] Figure 6a The separated target signal time-frequency diagram when the interference distance unit is the same as the target provided by the embodiment of the present application;

[0054] Figure 6b The separated echo signal restoration diagram when the interference distance unit is the same as the target provided by the embodiment of the present application;

[0055] Figure 6c The separated target signal pulse pressure diagram when the interference distance unit is the same as the target provided by the embodiment of the present application;

[0056] Figure 6d The separated target signal phase tracking diagram when the interference distance unit is the same as the target provided by the embodiment of the present application;

[0057] Figure 7a The separated target signal time-frequency diagram when the jammer is set at a distance of 3000m based on a target at a distance of 2000m provided by the embodiment of the present application;

[0058] Figure 7b The separated target signal pulse pressure diagram when the jammer is set at a distance of 3000m based on a target at a distance of 2000m provided by the embodiment of the present application;

[0059] Figure 7c The separated target signal phase tracking diagram when the jammer is set at a distance of 3000m based on a target at a distance of 2000m provided by the embodiment of the present application;

[0060] Figure 8a The original target signal time-frequency diagram when the signal type is an LFM signal provided by the embodiment of the present application;

[0061] Figure 8b The separated target signal time-frequency diagram when the signal type is an LFM signal provided by the embodiment of the present application;

[0062] Figure 8c The separated echo signal restoration diagram when the signal type is an LFM signal provided by the embodiment of the present application;

[0063] Figure 8d The separated echo signal pulse pressure diagram when the signal type is an LFM signal provided by the embodiment of the present application;

[0064] Figure 8eThe signal type provided by the embodiment of the present application is a separated target signal phase correlation graph of an LFM signal;

[0065] Figure 8f The signal type provided by the embodiment of the present application is a separated interference signal phase correlation graph of an LFM signal;

[0066] Figure 9a The signal type provided by the embodiment of the present application is a separated target signal time-frequency graph of an LFM signal with an intermittent sampling interference type;

[0067] Figure 9b The signal type provided by the embodiment of the present application is a separated echo signal time-frequency graph of an LFM signal with an intermittent sampling interference type;

[0068] Figure 9c The signal type provided by the embodiment of the present application is a separated target signal pulse compression graph of an LFM signal with an intermittent sampling interference type;

[0069] Figure 9d The signal type provided by the embodiment of the present application is a separated target signal phase correlation graph of an LFM signal with an intermittent sampling interference type;

[0070] Figure 10a The signal type provided by the embodiment of the present application is a separated target signal time-frequency graph of an LFM signal with an SNR of 0dB;

[0071] Figure 10b The signal type provided by the embodiment of the present application is a separated echo signal time-frequency graph of an LFM signal with an SNR of 0dB;

[0072] Figure 10c The signal type provided by the embodiment of the present application is a separated target signal pulse compression graph of an LFM signal with an SNR of 0dB;

[0073] Figure 10d The signal type provided by the embodiment of the present application is a separated target signal phase correlation graph of an LFM signal with an SNR of 0dB;

[0074] Figure 11a The signal type provided by the embodiment of the present application is a separated target signal time-frequency graph of an LFM signal with an interference retransmission number of 1-3 times of random original target signals;

[0075] Figure 11b The signal type provided by the embodiment of the present application is a separated target signal time-frequency graph of an LFM signal with an interference retransmission number of 1-3 times of random original target signals;

[0076] Figure 11cThe signal type provided by the embodiment of the present application is an LFM signal, and the interference retransmission times are 1-3 times of random separated echo signal recovery graphs.

[0077] Figure 11d The signal type provided by the embodiment of the present application is an LFM signal, and the interference retransmission times are 1-3 times of random separated echo signal recovery graphs.

[0078] Figure 11e The signal type provided by the embodiment of the present application is an LFM signal, and the interference retransmission times are 1-3 times of random separated echo signal recovery graphs.

[0079] Figure 11f The signal type provided by the embodiment of the present application is an LFM signal, and the interference retransmission times are 1-3 times of random separated echo signal recovery graphs.

[0080] Figure 12 The signal type provided by the embodiment of the present application is an LFM signal, and the interference retransmission times are 1-3 times of random separated echo signal recovery graphs. DETAILED DESCRIPTION

[0081] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0082] The present application provides a radar target and interference signal separation method based on a lightweight neural network, as shown in Figure 1 The method comprises the following steps S101-S103.

[0083] S101, pre-processing the time domain echo signal to obtain a pre-processed signal and full features corresponding to the pre-processed signal;

[0084] Specifically, in step S101, the time domain echo signal is pre-processed to obtain a pre-processed signal, comprising the following steps S1011-S1016.

[0085] S1011, loading and variable initialization of the time domain echo signal.

[0086] S1012, reading the data file in the specified path, loading the complex form of intermittent sampling retransmission interference data and target echo data; wherein the target echo data includes noise data.

[0087] S1013, respectively performing data shape transformation on the intermittent sampling retransmission interference data and the target echo data to obtain remodeled interference data and remodeled target echo data.

[0088] S1014, respectively extract the real part and the imaginary part of the reshaped interference data and the real part and the imaginary part of the reshaped target echo data.

[0089] S1015, using a cyclic interpolation sorting method, alternately sort the real part and the imaginary part of the reshaped interference data to obtain a continuous interference data long vector, and alternately sort the real part and the imaginary part of the reshaped target echo data to obtain a continuous target echo data long vector.

[0090] S1016, add the continuous interference data long vector and the continuous target echo data long vector to obtain a preprocessed signal.

[0091] For example, first, data loading and variable initialization are performed: by reading the.mat file stored in the specified path, intermittent sampling retransmission interference data and target echo data in complex form are loaded; wherein, the target echo data includes noise data. The data dimensions of the intermittent sampling retransmission interference data and the target echo data are determined to ensure that they conform to the predetermined shape (for example, in the experiment, 10 frames of echo data are used, each frame has 64 pulses, and each frame of data has 6400 distance sampling points <nrn>). Then data reshaping and combination are performed: the data shape is converted from the format of multiple frames to a continuous data stream, so that the originally scattered data segments are rearranged into a continuous long vector. The reshaped interference data and the reshaped target echo data are expanded so that their dimensions meet the input requirements of the model. This includes extracting the real and imaginary parts from the complex matrix of the reshaped interference data and the reshaped target echo data, respectively; data type conversion: after completing the data type conversion of the extracted real and imaginary part data, tensor processing is performed in PyTorch. The data type is converted to a format suitable for deep learning framework processing, which ensures the numerical accuracy of the data while also enabling the data to smoothly participate in the model training and verification process; then cyclic interpolation sorting is performed: in order to effectively process time series data and utilize the characteristics of complex signals, the real and imaginary part data are alternately sorted (Alternating Sorting). Specifically, for each sample, the real part data is indexed even, and the imaginary part data is indexed odd. This alternating sorting method can maintain the time series continuity of the data, while enabling the subsequent model to better learn the phase information in the data and to some extent reduce data redundancy and noise influence. Finally, data mixing is performed: the continuous interference data long vector and the continuous target echo data long vector are added to form the final mixed data, and data validity test is performed to ensure that each data segment meets the model input requirements. Data with insufficient length or incorrect dimensions are filtered, and debug information is returned. The mixed data that meets the requirements will be applied to the subsequent separation task.

[0092] S102, input the preprocessed signal into the mixed attention mechanism network trained to obtain a separated target echo signal. As shown in Figure 2 , the mixed attention mechanism network includes an encoder module, a separation module, an inter-stage connection module, and a decoder module.

[0093] The encoder module is used to encode the full feature and extract the multi-scale feature of the preprocessed signal. Specifically, as shown in Figure 3 , the encoder module includes a first depth separable convolution layer (DWConv), a second depth separable convolution layer (DWConv), a third depth separable convolution layer (DWConv), a first average pooling layer (Average Pooling), a second average pooling layer (Average Pooling), and a third average pooling layer (Average Pooling) connected in sequence.

[0094] The first deep separable convolutional layer is used for feature extraction on the extracted full feature to obtain 1 / 2 scale feature; the second deep separable convolutional layer is used for feature extraction on the 1 / 2 scale feature to obtain 1 / 4 scale feature; the third deep separable convolutional layer is used for feature extraction on the 1 / 4 scale feature to obtain 1 / 8 scale feature; the first average pooling layer is used for average pooling operation on the 1 / 4 scale feature to obtain the first 1 / 8 scale feature; the second average pooling layer is used for average pooling operation on the 1 / 2 scale feature to obtain the second 1 / 8 scale feature; and the third average pooling layer is used for average pooling on the full feature to obtain the third 1 / 8 scale feature.

[0095] The separation module is used for fusing the multi-scale features to obtain fused features, and calculating global attention weights by using a global attention module to obtain a global self-attention map; specifically, the separation module includes the global attention module.

[0096] For example, the input of the global attention module is multi-scale features [F1, F2, …, F n ] from the encoder module. The multi-scale features are fused to form a comprehensive feature representation. The global attention weights are calculated, and the attention map A GA , A GA = softmax(f(F1, F2, …, F n )) is obtained by a softmax function; wherein f(·) represents a function for calculating the similarity between features.

[0097] The global attention module (Global Attention) includes: the TasNet Layer can obtain attention signals from the multi-scale features to modulate the features of different scales in the encoder from top to bottom.

[0098] The inter-stage connection module is used for precision recovery on the multi-scale features to obtain recovered multi-scale features; specifically, as shown in Figure 3 , the inter-stage connection module includes: three first up-sampling modules (Up-sample 2x), second up-sampling modules (Up-sample 4x) and third up-sampling modules (Up-sample 8x) connected in sequence.

[0099] The first up-sampling module is used for two times up-sampling on the point multiplication result of the fused feature and the 1 / 8 scale feature according to the global self-attention map, and the sampling result is multiplied with the 1 / 4 scale feature to obtain the first recovered multi-scale feature; the second up-sampling module is used for four times up-sampling on the first recovered multi-scale feature according to the global self-attention map, and the sampling result is multiplied with the 1 / 2 scale feature to obtain the second recovered multi-scale feature;

[0100] The third up-sampling module is configured to perform eight times up-sampling on the second recovered multi-scale feature according to the global self-attention map, and the sampling result is point multiplied with the full feature of the preprocessed signal to obtain the third recovered multi-scale feature.

[0101] The decoder module is configured to modulate the recovered multi-scale feature pair by using the global self-attention map to obtain the separated target echo signal; wherein the multi-scale feature includes 1 / 2 scale feature, 1 / 4 scale feature and 1 / 8 scale feature; and the recovered multi-scale feature includes the first recovered multi-scale feature, the second recovered multi-scale feature and the third recovered multi-scale feature.

[0102] Specifically, as shown in Figure 3 The decoder module includes a first cascaded local attention layer (LA Layer), a second cascaded local attention layer (LA Layer) and a third cascaded local attention layer (LA Layer) connected in sequence.

[0103] The first cascaded local attention layer is configured to extract local attention features according to the first recovered multi-scale feature and the second recovered multi-scale feature to obtain the first local attention feature.

[0104] The second cascaded local attention layer is configured to extract local attention features according to the first local attention feature and the third recovered multi-scale feature to obtain the second local attention feature.

[0105] The third cascaded local attention layer is configured to extract local attention features according to the second local attention feature and the full feature of the preprocessed signal to obtain the third local attention feature.

[0106] Specifically, the decoder module is configured to modulate the recovered multi-scale feature pair by using the global self-attention map to obtain the separated target echo signal, including: mapping the third local attention feature into the same spatial dimension as the time domain echo signal to obtain the separated target echo signal.

[0107] As shown in Figure 3 The local attention (LA) layer is used in the decoder to restore the features to high resolution.

[0108] Specifically, in step S102, the training of the hybrid attention mechanism network includes: dynamically adjusting the learning rate of the hybrid attention mechanism network according to the evaluation separation index.

[0109] Here, the calculation formula of the evaluation separation index is:

[0110] wherein, represents the estimated target echo signal; s represents the pure source signal; S target denotes the target signal part extracted from the estimated signal; e noise denotes the noise part, i.e. the difference between the estimated signal and the target signal; s target denotes the signal enhancement signal-to-noise ratio for evaluating the quality of the target signal.

[0111] The projection form of inner product can better estimate its similarity, since the loss in the network is decreasing from large to small, so the negative value of SI-SNR is taken, and the promotion of SI-SDR can represent the improvement of separation performance.

[0112] S103, according to the separated target echo signal, the specific information of the target is estimated.

[0113] Specifically, in step S103, according to the separated target echo signal, the specific information of the target is estimated, including the following steps S1031 to S1032.

[0114] S1031, determine the signal type of the time domain echo signal, and pulse compression processing is performed on the separated target echo signal according to the corresponding pulse compression method to obtain a pulse compression signal;

[0115] S1032, the pulse compression signal is subjected to coherent processing to obtain a coherent processing signal; wherein the coherent processing signal includes the specific information of the target.

[0116] Exemplarily, the separated target echo signal is imported into Matlab, and the separated waveform data of the corresponding length is set according to the actual data, for example, a frame has N pulses, since the real part and the imaginary part are circularly interpolated by the dual module, then the separated data with a length of 2N is needed to restore a frame of echo signal, the row index of the stored data is calculated to ensure that the real part and the imaginary part are spliced in order, and the complex matrix is combined to store the signal and the interference respectively.

[0117] When it is a frequency agile signal, it cannot be directly pulse compressed by a matched filter. M sub-filters need to be constructed to complete the pulse compression processing. For the reconstructed pulse signal Each sub-pulse is subjected to segmented pulse compression processing.

[0118] Specifically, the frequency agile signal is represented as:

[0119] When the frequency agile signal is subjected to pulse compression processing, the pulse compression signal of each sub-pulse is represented as:

[0120]

[0121] Wherein, rect(·) represents a rectangular pulse signal for the ratio of time sequence and pulse; t represents the time sequence, defined as m represents the index of the current pulse number; T S denotes the width of the pulse; b(m) denotes the frequency offset (modulation parameter) associated with the mth pulse; Δf s denotes the frequency increment; u(t) denotes the complex envelope of the signal, denotes the fast time, t m denotes the slow time.

[0122] Further, intra-pulse accumulation is performed on all sub-pulses, and there is relatively small pulse compression loss. When a linear frequency modulation (LFM) signal is used, segmentation and pulse compression are not required, and a Fourier transform (FFT) is directly performed on the input signal, and then multiplied by the conjugate spectrum H of the reference chirp signal, that is, matched filtering. The coherent accumulation adopts a relatively traditional algorithm, that is, the sum of the same positions after alignment of different pulses is superimposed, and since processing is performed in the frequency domain, an FFT operation is required. It should be noted that when the separation result is good, that is, the phase information loss is small, a relatively ideal coherent processing result can be obtained.

[0123] The LFM signal is expressed as:

[0124] Where t denotes a time sequence, A s denotes the amplitude, f0 denotes the center frequency, T p denotes the pulse, and B denotes the bandwidth.

[0125] In one specific embodiment provided by the present application, for intermittent sampling and forwarding interference, the experiment first uses a variable waveform, and the first experimental group first sets the transmission waveform as an intra-pulse fixed coding sequence signal.

[0126] Main parameter settings of the first part of the first experimental group experiment ①:

[0127] Table 1 data set parameter settings

[0128]

[0129]

[0130] The interference type is set as intermittent sampling interference (sampling 1 to 3), the SNR is set as 0 dB, and the jam-to-noise ratio (JNR) and the jam-to-signal ratio (CTR) are calculated as 22.92 dB and 22.92 dB, respectively.

[0131] After the hybrid attention mechanism network provided by the present application, the obtained result is as shown in Figures 4a to 4d .

[0132] As Figures 4a to 4d It can be seen that when the SNR is fixed at 0 dB and the CTR is 22.92 dB, the network has good separation performance for targets at distances of 2000 m and 3000 m under the condition of one-to-three interference. Further increasing the noise, adjusting the SNR to-10 dB, the CTR remains unchanged at 22.92 dB, as shown in the following experiment ②, and the experimental results are shown in Figures 5a to 5d .

[0133] As shown in Figure 5a , the separated signal is not very clear due to the increase of noise, and the pulse pressure effect is not better than before, but the speed of the target signal can still be easily obtained through coherent accumulation, which is consistent with the preset parameters. Figure 4c

[0134] Considering whether the interference and target can be separated more cleanly in the same distance unit, the interference distance unit is changed to the same as the target based on experiment ②, and the following experiment ③ is performed:

[0135] As shown in Figures 6a to 6d , the effect is similar to experiment ②. Further considering whether the echo signal with larger CTR and JNR can be separated, the target normalized amplitude is 1, and the interference is 100, i.e. the CTR is 40 dB, the SNR is still-10 dB, and the JNR is 30 dB; based on the target at a distance of 2000 m, the interference machine is set at a distance of 3000 m, and the results are shown in Figures 7a to 7c .

[0136] Further, LFM is used as the radar transmitting signal, and at the same time, the randomness of the experiment is further improved, and experiment group two is performed.

[0137] (2) Main parameter settings of experiment group two experiment ①:

[0138] Table 2 dataset parameter settings

[0139] Parameter Value Parameter Value Target amplitude 1 Jammer amplitude 100 Target distance / m 2000 Jammer initial distance / m 2300 Target speed / m / s 50 Jammer speed / m / s 75

[0140] The transmitting signal type is changed to LFM signal, the interference type is set to intermittent sampling interference (one-to-one), the SNR is set to-10 dB, and the JNR and CTR are calculated to be 30 dB and 40 dB respectively.

[0141] As shown in Figures 8a to 8d , the signal separation after processing of experiment two is shown in Figures 8a to 8d It is found that LFM signal has better separation performance than frequency agile signal under similar conditions, and the network is similar to long short-term memory structure which is more sensitive to causality, and the continuity of LFM signal in time-frequency domain is more than that of frequency agile signal, and its causality is also better.​

[0142] The main parameter settings for Experiment 2, Part ② of Experiment Group 2, include random sampling time for the jammer:

[0143] Table 3 Dataset Parameter Settings

[0144]

[0145]

[0146] The transmitted signal type remains LFM signal, and the interference type is set to intermittent sampling interference ① (sample 1 to 1) and intermittent sampling interference ② (sample 1 to N, where N is a random value). The noise flag is set to 0, meaning no noise is added. The interference-to-signal ratio (CTR) is calculated to be 40dB.

[0147] like Figures 9a to 9d As shown, this is the post-processing diagram of signal separation ② in Experiment 2. Experiment ② only presents a multi-interference scenario with a high signal-to-interference ratio. To verify the network's generalization ability and usability, the velocity and initial position of the target and interference are set to random values. The specific parameters are shown in Experiment ③, as follows:

[0148] Table 4 Dataset Parameter Settings

[0149]

[0150] The transmitted signal type remains LFM signal, the interference type is set to intermittent sampling interference (1 sample to N, where N is a random value and the sampling time is random), the SNR is set to 0dB, and the dryness ratio (JNR) and interference-to-signal ratio (CTR) are calculated to be 20dB and 20dB, respectively.

[0151] The experimental results of experiment ③ are as follows Figures 10a to 10d As shown, by Figure 10d It was found that the coherent results still had tiny sidelobes, which were identified as residual interference signals. Since the raw data generated at this point was relatively random—meaning each frame fed into the network for training was essentially different—and the total number of training samples was small, it was perfectly normal for the separation effect to weaken. Figures 11a to 11f The image shown is a time-frequency graph of some of the input data.

[0152] The differences in the original generated data are clearly visible in the above figure. The starting positions of the target echo and the interference are different, while the random velocity information cannot be discerned from the figure. Furthermore, the sampling time and number of relays of the interference are also random; if they exceed a fixed range, they are truncated. The original generated target echo signal is superimposed with the interference, as shown in the figure. Figure 12 As shown.

[0153] Comparison of the above images Figure 10a It is found that the interference residual is obvious, and the reason is that the difference between the input data and the limitation of the number of samples, as described above. It should be noted that all time-frequency diagrams need to be uniformly down-converted.

[0154] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and necessary hardware. Based on such an understanding, the technical solutions of the present application can be embodied in the form of a software product or through the implementation of data migration. The computer software product can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0155] The various embodiments in the specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. The whole or part of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, mobile communication terminals, multi-processor systems, microprocessor-based systems, programmable electronic devices, network PCs, small computers, large computers, distributed computing environments including any of the above systems or devices, etc.

[0156] The above embodiments are only used to illustrate the technical solutions of the present application, and are not limited to the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some or all of the technical features. The modification or replacement does not make the essence of the corresponding technical solution deviate from the scope of the technical solutions of the present application.< / nrn>

Claims

1. A radar target and jamming signal separation method based on a lightweight neural network, characterized in that, include: The time-domain echo signal is preprocessed to obtain a preprocessed signal and the full features corresponding to the preprocessed signal; The preprocessed signal is input into the trained hybrid attention mechanism network to obtain the separated target echo signal; wherein, the hybrid attention mechanism network includes an encoder module, a separation module, an inter-level connection module, and a decoder module; the encoder module is used to encode the full features and extract the multi-scale features of the preprocessed signal; The separation module is used to fuse the multi-scale features to obtain fused features, and to calculate global attention weights using the global attention module to obtain a global self-attention map; the inter-level connection module is used to perform accuracy recovery on the multi-scale features to obtain recovered multi-scale features; the decoder module is used to modulate the recovered multi-scale feature pairs using the global self-attention map to obtain the separated target echo signal; The specific information of the target is estimated based on the separated target echo signal.

2. The radar target and jamming signal separation method based on lightweight neural network according to claim 1, characterized in that, The preprocessing of the time-domain echo signal to obtain a preprocessed signal includes: The time-domain echo signal is loaded and its variables are initialized; Read the data file in the specified path and load the intermittent sampling forwarding interference data and target echo data in complex form; wherein, the target echo data includes noise data; The intermittent sampling forwarding interference data and the target echo data are respectively subjected to data shape transformation to obtain reshaped interference data and reshaped target echo data; The real and imaginary parts of the reshaped interference data and the real and imaginary parts of the reshaped target echo data are extracted respectively. The real and imaginary parts of the reshaped interference data are alternately sorted using a cyclic interpolation sorting method to obtain a continuous long vector of interference data. Similarly, the real and imaginary parts of the reshaped target echo data are alternately sorted to obtain a continuous long vector of target echo data. The long vector of continuous interference data is added to the long vector of continuous target echo data to obtain the preprocessed signal.

3. The radar target and jamming signal separation method based on lightweight neural network according to claim 1, characterized in that, The training of the hybrid attention mechanism network includes: dynamically adjusting the learning rate of the hybrid attention mechanism network according to the evaluation separation metric.

4. The radar target and jamming signal separation method based on lightweight neural network according to claim 3, characterized in that, The calculation formula for the evaluation separation index is expressed as follows: ; in, This indicates an estimate of the target echo signal; Indicates a pure source signal; This represents the portion of the target signal extracted from the estimated signal; This represents the noise component, i.e., the difference between the estimated signal and the target signal; This indicates the signal-to-noise ratio (SNR), used to evaluate the quality of the target signal.

5. The radar target and jamming signal separation method based on lightweight neural network according to claim 1, characterized in that, The encoder module includes a first depthwise separable convolutional layer, a second depthwise separable convolutional layer, a third depthwise separable convolutional layer, a first average pooling layer, a second average pooling layer, and a third average pooling layer connected in sequence. The first depthwise separable convolutional layer is used to extract features from the full features, resulting in... Scale characteristics; The second depth-separable convolutional layer is used for the... Feature extraction is performed using scale features to obtain... Scale characteristics; The third depth-separable convolutional layer is used for the... Feature extraction is performed using scale features to obtain... Scale characteristics; The first average pooling layer is used for the... The scale features are subjected to average pooling to obtain the first Scale characteristics; The second average pooling layer is used for the... The scale features are subjected to average pooling to obtain the second... Scale characteristics; The third average pooling layer is used to perform average pooling on the full features to obtain the third... Scale characteristics.

6. The radar target and jamming signal separation method based on lightweight neural network according to claim 1, characterized in that, The inter-level connection module includes three sequentially connected modules: a first upsampling module, a second upsampling module, and a third upsampling module. The first upsampling module is used to perform upsampling on the fused features and based on the global self-attention map. The dot product of scale features is upsampled by a factor of two, and the sampling result is compared with... Multiply the scale features by a dot product to obtain the first recovered multiscale features; The second upsampling module is used to perform a four-fold upsampling on the first recovered multi-scale feature based on the global self-attention map, and the sampling result is compared with... The second recovered multi-scale feature is obtained by multiplying the scale features by the dot product. The third upsampling module is used to upsample the second recovered multi-scale feature by eight times based on the global self-attention map, and the sampling result is multiplied by the full features of the preprocessed signal to obtain the third recovered multi-scale feature; wherein, the Scale features, the aforementioned Scale features and the The scale feature is the multi-scale feature.

7. The radar target and jamming signal separation method based on lightweight neural network according to claim 6, characterized in that, The decoder module includes a first cascaded local attention layer, a second cascaded local attention layer, and a third cascaded local attention layer connected in sequence; The first cascaded local attention layer is used to extract local attention features based on the first recovery multi-scale features and the second recovery multi-scale features to obtain the first local attention features; The second cascaded local attention layer is used to extract local attention features based on the first local attention features and the third recovery multi-scale features to obtain the second local attention features; The third cascaded local attention layer is used to extract local attention features based on the second local attention features and the full features of the preprocessed signal to obtain the third local attention features.

8. The radar target and jamming signal separation method based on lightweight neural network according to claim 7, characterized in that, The decoder module is used to modulate the recovered multi-scale feature pairs using the global self-attention map to obtain the separated target echo signal, including: The third local attention feature is mapped to the same spatial dimension as the time-domain echo signal to obtain the separated target echo signal.

9. The radar target and jamming signal separation method based on lightweight neural network according to claim 1, characterized in that, The step of estimating specific information about the target based on the separated target echo signal includes: The signal type of the time-domain echo signal is determined, and the pulse compression processing of the separated target echo signal is performed according to the corresponding pulse compression method to obtain the pulse compression signal; The pulse compression signal is subjected to coherent processing to obtain a coherent processed signal; wherein, the coherent processed signal includes specific information about the target.

Images