Pulse stream de-interleaving method and device, equipment and storage medium

Abstract

The invention discloses a pulse stream de-interleaving method, device and equipment and a storage medium, and belongs to the technical field of signal sorting, and the method comprises the steps: recovering the features lost due to pulse overlapping in a specific pulse stream through the processing of overlapped pulses, then mining and learning a long-term time pattern in the pulse stream through employing LSTM, tracking the change features of the specific pulse stream, and obtaining the de-interleaving feature of the specific pulse stream. Therefore, effective separation of the pulses is realized. According to the method, high sorting accuracy can be obtained under high pulse loss and overlapping probability.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal sorting, and in particular relates to a pulse stream deinterleaving method, device, equipment and storage medium. Background Art

[0002] The pulse stream deinterleaving method based on sequence information plays an important role in the main sorting process and can effectively sort out radar individuals.

[0003] However, in today's high-density pulse data environment, pulse overlap and pulse loss are serious phenomena. The pulse stream deinterleaving method used in conventional sequence information is prone to failure and low accuracy during the search phase, which seriously affects the effectiveness of the electronic reconnaissance system. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the prior art and provide a pulse stream deinterleaving method, device, equipment and storage medium, which restores the features lost due to pulse overlap in a specific pulse stream by processing overlapping pulses, and then uses LSTM to mine and learn the long-term time patterns in the pulse stream and track the changing characteristics of the specific pulse stream, thereby achieving effective separation of the pulse.

[0005] The object of the present invention is achieved through the following technical solutions:

[0006] A pulse stream deinterleaving method, the method comprising:

[0007] Perform bidirectional training on the LSTM prediction model for each type of pulse stream to obtain a forward LSTM prediction model and a reverse LSTM prediction model for each type of pulse stream;

[0008] The trained bidirectional LSTM model is used for deinterleaving, and the possible features of overlapping pulses are calculated through stack operations. The starting prediction pulse is selected to start the prediction and matching of each pulse, and the pulses are predicted and matched one by one to separate the same pulse before and after the starting prediction pulse.

[0009] Furthermore, the method further comprises:

[0010] Each type of radar pulse stream is represented by a continuous two-dimensional vector, which includes the PRI characteristics and PW characteristics of the pulse.

[0011] Furthermore, the bidirectional training of the LSTM prediction model for each type of pulse stream specifically includes:

[0012] The input pulse stream is segmented into segments of preset length to obtain a number of samples, where the feature of the next pulse is the label of the feature of the previous pulse;

[0013] For each type of pulse data, all samples in the training set are traversed for training to obtain the forward LSTM prediction model corresponding to each type of pulse data;

[0014] For each type of pulse data, the end time is taken as the start time, and the PRI feature of each pulse is reversely calculated as the reverse pulse data training set;

[0015] All samples in the reverse pulse data training set are traversed for training to obtain the reverse LSTM prediction model corresponding to each type of pulse data.

[0016] Furthermore, the method for selecting the starting prediction pulse specifically includes:

[0017] Calculate the latent features of each pulse;

[0018] Inputting the potential features into the trained bidirectional LSTM model, and calculating the matching error between the real pulse and the predicted pulse;

[0019] The forward matching error of each pulse is obtained through the forward LSTM, and the forward sequence number of each pulse is obtained by sorting them from small to large, and the pulse with the smallest forward sequence number is taken as the starting pulse.

[0020] Furthermore, the method further comprises:

[0021] If there is a pulse loss, the predicted value is used to replace the lost pulse feature, and an error threshold is set to determine whether the next pulse is lost. If all matching errors calculated within the matching range are greater than the error threshold, the next pulse is considered to be lost.

[0022] On the other hand, the present invention further provides a pulse stream deinterleaving device, the device comprising:

[0023] A model training module, wherein the model training module performs bidirectional training of the LSTM prediction model on each type of pulse stream to obtain a forward LSTM prediction model and a reverse LSTM prediction model for each type of pulse stream;

[0024] A pulse separation module uses a trained bidirectional LSTM model for deinterleaving, calculates possible features of overlapping pulses through stack operations, selects a starting prediction pulse to start predicting and matching pulses one by one, predicts and matches pulses one by one, and separates the same pulse before and after the starting prediction pulse.

[0025] On the other hand, the present invention further provides a computer device, comprising a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement any one of the above-mentioned pulse stream deinterleaving methods.

[0026] On the other hand, the present invention further provides a computer-readable storage medium, in which a computer program is stored. The computer program is loaded and executed by a processor to implement any of the above-mentioned pulse stream deinterleaving methods.

[0027] The beneficial effects of the present invention are:

[0028] (1) The present invention recovers the features lost due to pulse overlap in a specific pulse stream by processing overlapping pulses, and then uses LSTM to mine and learn the long-term time patterns in the pulse stream and track the changing characteristics of the specific pulse stream, thereby achieving effective separation of the pulses and achieving a higher sorting accuracy under high pulse loss and overlap probabilities.

[0029] (2) Compared with the traditional method, the present invention can reduce the probability of pulse loss by retaining the overlapping pulses and performing correlation processing on them to restore the characteristics of the target pulse lost due to pulse overlap, thereby effectively reducing the probability of pulse loss.

[0030] (3) The present invention uses LSTM to mine and learn the long-term temporal patterns in the target pulse stream, so that the sorting process can adaptively move closer to the target pulse stream pattern and gradually complete self-correction when a wrong selection is made due to pulse loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flowchart of a pulse stream deinterleaving method according to an embodiment of the present invention;

[0032] Figure 2 is a schematic diagram representing the pulse flow of an embodiment of the present invention;

[0033] Figure 3 Schematic diagram of the LSTM prediction network training framework according to an embodiment of the present invention;

[0034] Figure 4 is a schematic diagram of a stack of TOA and TOD according to an embodiment of the present invention;

[0035] Figure 5 is a schematic diagram of starting pulse selection in an embodiment of the present invention;

[0036] Figure 6 This is a schematic diagram of forward and reverse search matching in an embodiment of the present invention;

[0037] Figure 7 It is a structural block diagram of a pulse stream deinterleaving device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0039] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0040] In a high-density pulse data environment, pulse overlap and pulse loss are serious. The pulse stream deinterleaving method used in conventional sequence information is prone to failure and low accuracy during the search phase, which seriously affects the effectiveness of the electronic reconnaissance system.

[0041] In order to solve the above technical problems, the following embodiments of the pulse stream deinterleaving method, device, equipment and storage medium of the present invention are proposed.

[0042] Example 1

[0043] This embodiment provides a pulse stream deinterleaving method based on a bidirectional LSTM prediction model. In this embodiment, each pulse is firstly transformed into a two-dimensional vector {PRI n ,PW n} characterization, each type of radar pulse stream can be represented by a series of continuous two-dimensional vectors. Figure 2 ,like Figure 2 Shown is a schematic diagram representing the pulse flow of this embodiment.

[0044] Reference Figure 1 ,like Figure 1 FIG. 1 is a flowchart of a pulse stream deinterleaving method according to an embodiment of the present invention. The method specifically comprises the following steps:

[0045] Step 1: Perform bidirectional training of the LSTM prediction model for each type of pulse stream. Figure 3 ,like Figure 3 The figure shows a schematic diagram of the LSTM prediction network training framework of this embodiment. Figure 3 The following is a typical LSTM many-to-many prediction scenario. The specific training method is as follows:

[0046] Step 1.1: Assume that the original clean pulse characteristic sequence is I = {I 1 ,I 2 ,…,I N}, where the elements The data is standardized using the classic z-sore standardization method. The standardized result is X = {X 1 ,X 2 ,L,X N},in The corresponding formula is as follows:

[0047]

[0048] Step 1.2: Segment the input pulse stream into segments of length L to obtain several samples, the first sample is X 1 ={X 1 ,X 2 ,…,X L}, the label corresponding to this sample is Label 1 ={X 2 ,X 3 ,…,X L+1}, that is, the feature of the next pulse is the label of the feature of the previous pulse, which reflects the mechanism of single-step prediction.

[0049] Step 1.3: For each type of pulse data, traverse all samples in the training set for full training to obtain the forward LSTM prediction model corresponding to each type of pulse data.

[0050] Step 1.4: For each type of pulse data, take the end time as the start time, and reversely calculate the PRI feature of each pulse as the reverse pulse data training set. By training with the above steps, the reverse LSTM prediction model corresponding to each type of pulse data can be obtained.

[0051] Step 2: Use the trained bidirectional LSTM model to perform pulse deinterleaving. This includes the following sub-steps:

[0052] Step 2.1: Overlap the stack of pulses.

[0053] by Describe the received aliased interleaved pulse stream sequence, perform stack operations on the overlapping pulses in the test phase, and calculate the possible characteristics of the overlapping pulses, refer to Figure 4 ,like Figure 4 The stack diagram of TOA and TOD in this embodiment is shown. The characterization vector of the overlapping pulse consists of four components:

[0054]

[0055] Step 2.2: Selection of the start pulse.

[0056] Reference Figure 5 ,like Figure 5 FIG. 1 is a schematic diagram of the starting pulse selection of this embodiment. Figure 5The figure shows the selection of the starting pulse in a certain aliased interleaved pulse stream, assuming that the forward LSTM and the reverse LSTM are trained with the pulse stream corresponding to the first pulse in the figure (hereinafter referred to as pulse stream A). The specific steps of starting pulse selection are as follows:

[0057] Step 2.2.1: Calculate the latent features for each pulse.

[0058] First, calculate the range Kmin~Kmax of the previous or next similar pulse:

[0059]

[0060] In the formula, TOA max Indicates the maximum TOA value in the received pulse stream, N test represents the number of pulses in the test set, then P test Indicates the density of the pulse flow, floor(·) is the upward rounding function, ceil(·) is the downward rounding function, PRI min and PRI max Represent the minimum and maximum PRI values ​​of pulse stream A respectively.

[0061] In the range of Kmin to Kmax pulses, calculate the potential PRI characteristics of the current pulse. Figure 4 For example, the potential features of the current pulse can be expressed as a matrix:

[0062]

[0063] If there are overlapping pulses within the range of Kmin to Kmax pulses, the possible PRI is calculated based on all TOAs in the TOA stack of overlapping pulses to represent the potential characteristics of the current pulse. All pulses in this pulse stream are traversed to obtain the potential characteristics of each pulse.

[0064] It should be noted that, in this embodiment, the first Kmax pulses are discarded because the potential PRI features cannot be calculated.

[0065] Step 2.2.2: Calculate the matching error of the predicted pulse.

[0066] Input all possible features of the current pulse into the trained LSTM network to obtain the possible predicted pulse matrix:

[0067]

[0068] In the next Kmin~Kmax pulse range, the matching error between the real pulse and the possible predicted pulse matrix is ​​calculated respectively. The matching error is calculated as follows:

[0069]

[0070] Where λ 1 and λ 2 It can represent the feature accuracy coefficient, which respectively represents the prediction accuracy of the PRI feature and the PW feature. It can be adjusted according to the actual situation and is set to 0.5 here.

[0071] If there is an overlapping pulse within the next Kmin~Kmax pulse range, the matching error is calculated based on all possible characteristics of the overlapping pulse. The minimum value of all matching errors of the current pulse is taken as the matching error standard error of the pulse. i , and then sort the matching error criteria of all pulses.

[0072] Step 2.2.3: Take the pulse with the smallest serial number as the starting pulse.

[0073] The forward matching error standard error of each pulse is obtained through the forward LSTM i f , sorted from small to large to get the positive sequence number I of each pulse f Similarly, the reverse matching error standard error of each pulse is obtained through the reverse LSTM i b , after sorting, the reverse sequence number I of each pulse is obtained b Then, the serial number of each pulse is expressed as I=I f +I b , take the pulse with the smallest sequence number I as the starting pulse, that is, the pulse of pulse stream A, and use this pulse as a springboard to perform prediction matching forward and backward.

[0074] Step 2.3: Pulse-by-pulse prediction and matching.

[0075] Reference Figure 6 ,like Figure 6 The figure shows the forward and reverse search and matching schematic diagram of this embodiment. Input the features of the starting pulse into the forward LSTM network to obtain the predicted features of the next pulse, and calculate the matching errors of the next Kmin~Kmax pulses according to the formula in step 2.2.2. The pulse corresponding to the minimum matching error value is the next A pulse. Then continuously input all the selected A pulse features into the forward LSTM network to obtain the subsequent predicted features, and repeat the previous steps to select the A pulse after the starting pulse in the received pulse stream. Similarly, input the features of the starting pulse into the reverse LSTM network, perform prediction and matching in reverse, and select all A pulses before the starting pulse. In this way, the A pulse stream is separated from the aliased and interwoven pulse stream.

[0076] When encountering overlapping pulses, all possible features of the overlapping pulses are treated as independent pulses for matching and updating. If the pulse is lost due to other reasons, the predicted value is used to replace the lost pulse feature. An error threshold δ is set to determine whether the next pulse is lost. That is, if all matching errors calculated within the matching range are greater than the error threshold δ, the next pulse is considered to be lost, and the predicted value is used to replace the lost value.

[0077] This embodiment recovers the features lost due to pulse overlap in a specific pulse stream by processing overlapping pulses, and then uses LSTM to mine and learn the long-term time pattern in the pulse stream, and tracks the changing characteristics of the specific pulse stream, thereby achieving effective separation of the pulse. This method has two advantages over traditional methods: First, it reduces the probability of pulse loss. It effectively reduces the probability of pulse loss by retaining overlapping pulses and performing relevant processing on them to recover the features of the target pulse lost due to pulse overlap. Second, it is the powerful adaptive feature of deep learning. LSTM is used to mine and learn the long-term time pattern in the target pulse stream, so that the sorting process can adaptively move closer to the target pulse stream pattern, and gradually complete self-correction when a wrong selection is made due to pulse loss. Therefore, this method can achieve a higher sorting accuracy under high pulse loss and overlap probabilities.

[0078] Example 2

[0079] Reference Figure 7 ,like Figure 7 FIG. 1 is a block diagram of the structure of the pulse stream deinterleaving device of this embodiment, and the device specifically includes the following structures:

[0080] Model training module: The model training module performs bidirectional training of the LSTM prediction model for each type of pulse stream to obtain a forward LSTM prediction model and a reverse LSTM prediction model for each type of pulse stream;

[0081] Pulse separation module,The pulse separation module uses the trained bidirectional LSTM model for deinterleaving, calculates the possible characteristics of overlapping pulses through stack operations, and selects the starting prediction pulse to start the prediction and matching of each pulse, predicts and matches the pulses one by one, and separates the same pulse before and after the starting prediction pulse.

[0082] Example 3

[0083] This preferred embodiment provides a computer device, which can implement the steps in any embodiment of the pulse stream deinterleaving method provided in the embodiments of the present application. Therefore, the beneficial effects of the pulse stream deinterleaving method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0084] Example 4

[0085] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling related hardware through instructions, and the instructions can be stored in a computer-readable storage medium and loaded and executed by a processor. To this end, an embodiment of the present invention provides a storage medium, which stores multiple instructions, and the instructions can be loaded by a processor to execute the steps of any embodiment of the pulse stream deinterleaving method provided by the embodiment of the present invention.

[0086] The storage medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0087] Since the instructions stored in the storage medium can execute the steps in any pulse stream deinterleaving method embodiment provided in the embodiments of the present invention, the beneficial effects that can be achieved by any pulse stream deinterleaving method provided in the embodiments of the present invention can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A pulse stream deinterleaving method, characterized in that: The method comprises: Perform bidirectional training on the LSTM prediction model for each type of pulse stream to obtain a forward LSTM prediction model and a reverse LSTM prediction model for each type of pulse stream; The trained bidirectional LSTM model is used for deinterleaving, and the possible features of overlapping pulses are calculated through stack operations. The starting prediction pulse is selected to start the prediction and matching of each pulse, and the pulses are predicted and matched one by one to separate the same pulse before and after the starting prediction pulse.

2. The pulse stream deinterleaving method according to claim 1, characterized in that: The method further comprises: Each type of radar pulse stream is represented by a continuous two-dimensional vector, which includes the PRI characteristics and PW characteristics of the pulse.

3. The pulse stream deinterleaving method according to claim 1, characterized in that: The bidirectional training of the LSTM prediction model for each type of pulse stream specifically includes: The input pulse stream is segmented into segments of preset length to obtain a number of samples, where the feature of the next pulse is the label of the feature of the previous pulse; For each type of pulse data, all samples in the training set are traversed for training to obtain the forward LSTM prediction model corresponding to each type of pulse data; For each type of pulse data, the end time is taken as the start time, and the PRI feature of each pulse is reversely calculated as the reverse pulse data training set; All samples in the reverse pulse data training set are traversed for training to obtain the reverse LSTM prediction model corresponding to each type of pulse data.

4. The pulse stream deinterleaving method according to claim 1, characterized in that: The method for selecting the starting prediction pulse specifically includes: Calculate the latent features of each pulse; Inputting the potential features into the trained bidirectional LSTM model, and calculating the matching error between the real pulse and the predicted pulse; The forward matching error of each pulse is obtained through the forward LSTM, and the forward sequence number of each pulse is obtained by sorting them from small to large, and the pulse with the smallest forward sequence number is taken as the starting pulse.

5. The pulse stream deinterleaving method according to claim 1, characterized in that: The method further comprises: If there is a pulse loss, the predicted value is used to replace the lost pulse feature, and an error threshold is set to determine whether the next pulse is lost. If all matching errors calculated within the matching range are greater than the error threshold, the next pulse is considered to be lost.

6. A pulse stream deinterleaving device, characterized in that: The device comprises: A model training module, wherein the model training module performs bidirectional training of the LSTM prediction model on each type of pulse stream to obtain a forward LSTM prediction model and a reverse LSTM prediction model for each type of pulse stream; A pulse separation module uses a trained bidirectional LSTM model for deinterleaving, calculates possible features of overlapping pulses through stack operations, selects a starting prediction pulse to start predicting and matching pulses one by one, predicts and matches pulses one by one, and separates the same pulse before and after the starting prediction pulse.

7. A computer device, characterized in that: The computer device comprises a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the pulse stream deinterleaving method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which is loaded and executed by a processor to implement the pulse stream deinterleaving method according to any one of claims 1 to 5.

Images