A radar signal sorting method based on tag-based multi-Bernoulli filters

Abstract

This invention relates to the field of radar signal processing technology, specifically to a radar signal sorting method based on a tag-based multi-Bernoulli filter. The method includes: modeling each potential radar source as a tagged Bernoulli element, which contains information on survival probability and spatial distribution; predicting the state and survival of existing targets based on the target state and motion model of the previous time step; updating the existence probability of the Bernoulli element by calculating the likelihood of each "predicted target-observation" pair and associating it with each target for new measurements; generating new Bernoulli elements with a certain probability for observations not used to update Bernoulli elements; and removing Bernoulli elements with an existence probability below a threshold in the next step. This invention, by applying the tag-based multi-Bernoulli method to radar signal sorting, achieves robustness against missed detections and false alarms, while avoiding the need for complex target management where the survival of a target is entirely determined by its existence probability.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, specifically to a radar signal sorting method based on a tag-based multi-Bernoulli filter. Background Technology

[0002] With the rapid development of modern electronic information technology, the electromagnetic environment in modern electronic warfare has become unprecedentedly complex. Radar signal sorting, as a key technology for signal separation, deinterleaves the Pulse Description Word (PDW) extracted from intercepted signals and clusters similar pulse signals to obtain all pulse clusters belonging to the same radar. This enables independent modeling of the signal characteristics of each radar radiation source, and its performance has a significant impact on subsequent radiation source identification, tracking, and intent analysis.

[0003] Currently, mainstream sorting algorithms are mainly divided into two categories: sorting algorithms based on Pulse Repetition Interval (PRI) features and clustering sorting algorithms based on multi-parameter feature fusion. The former includes methods such as Cumulative Difference Histogram (CDIF) and Sequential Difference Histogram (SDIF). These methods find the radar PRI and PRI period based on the Time Difference of Arrival (DTOA) in the pulse stream, and then search for target radar pulses from the pulse stream. However, when facing large-scale jitter or slip signals, the PRI spectral lines are not only difficult to focus but also easily submerged in noise. Typical clustering methods for radar signal sorting include spatial density clustering, fuzzy clustering, model-based clustering, and density peak clustering. However, these clustering methods often require iterative optimization, have high computational complexity, are sensitive to the distribution of pulse PDW, and parameter selection often still relies on prior knowledge. Summary of the Invention

[0004] The purpose of this invention is to provide a radar signal sorting method based on a tag-based multi-Bernoulli filter to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A radar signal sorting method based on tag-based multi-Bernoulli filters includes the following steps:

[0007] S1. Model each potential radar radiation source as a tagged Bernoulli element; wherein the Bernoulli element includes a uniquely identified tag, the survival rate of the Bernoulli element, and a radar pulse descriptor.

[0008] S2. After detecting the next signal, predict each existing Bernoulli element according to the motion model, and update the state distribution and survival rate of each Bernoulli element.

[0009] S3. Combine the predicted LMB distribution with the observations to generate the LMB posterior distribution; calculate the likelihood of each "predicted target-observation" pair and determine the most likely target of the observation in a probabilistic manner; the survival rate of Bernoulli elements that are successfully associated with the observations will increase, and the state distribution will be corrected according to the observations.

[0010] S4. Observations that were not used to update any existing targets are used to generate new Bernoulli elements, i.e., new radar radiation sources, with a certain probability according to the birth model.

[0011] S5. Bernoulli elements that fail to be associated with any observation will have a reduced survival rate. When the survival rate is below a threshold, they will be removed in the next step.

[0012] Preferably, step S1 specifically includes:

[0013] Each potential radar source is modeled as a labeled Bernoulli element, and for each state... Add tags This yields a labeled state vector. The label set of set X is composed of Given; if X and If the potentials are equal, then each element of the random finite set X has a different label.

[0014] Therefore These are indicators of different labels; the LMB random finite set consists of a parameter set. Given, where L is the label space;

[0015] The density of an LMB random finite set is: ;

[0016] in, ;

[0017] For a delta-generalized labeled Bernoulli random finite set, its density is:

[0018] ;

[0019] Among the tags For tags, For each target in the current hypothesis, the measurement-target association history is provided.

[0020] Preferably, step S2 specifically includes:

[0021] Predicting each existing Bernoulli element based on the motion model, wherein the state transition matrix of the motion model is an identity matrix, and the specific prediction formula is as follows:

[0022] ;

[0023] ;

[0024] ;

[0025] in, As a label, it is unaffected by predictions. Represents the state-related survival probability, trajectory The survival probability is determined by Give, The single-target transfer density is determined by Give;

[0026] Predicting multi-object density in the form of δ-GLMB during the update step:

[0027] ;

[0028] in, Represents the predicted label space and And there are

[0029] ;

[0030] ;

[0031] Since the target tag must be a persistent tag Space or new target The space, therefore for each label , or .

[0032] Preferably, step S3 specifically includes:

[0033] Combining the predicted LMB distribution with observations yields an updated and more accurate posterior LMB distribution: ;

[0034] in, Represents the mapping space ;and

[0035] ;

[0036] ;

[0037] ;

[0038] The spatial likelihood is: ;

[0039] in, This represents the state-related detection probability of the target. It is the probability of not being detected. Indicates the spatial clutter intensity of the Poisson distribution;

[0040] For closed recursion, the LMB filter passes through a parameter... The LMB density is used to approximate the multi-object posterior density, where,

[0041] ;

[0042] ;

[0043] ;

[0044] The optimal a posteriori hypothesis is obtained by solving the extended cost matrix, which includes not only the uncertainty of the association of the measurement but also the appearance and disappearance of objects;

[0045] First, enumeration , and For each pair Define a P tuple ,in:

[0046] ;

[0047] To obtain tags in the following ways and :

[0048] ;

[0049] ;

[0050] If for , and ,set up:

[0051] ;

[0052] in, Corresponding to a persistent goal, Indicates the goals of new students; in addition, Indicates the index of the measurement, where This indicates the disappearance of the target. This indicates a missed detection;

[0053] Thus, the formula Rewritten as:

[0054] ;in, ;

[0055] By solving Achieve optimal allocation;

[0056] in, For the allocation matrix, The cost matrix;

[0057] ;

[0058] Where M is the number of observations, and P is the sum of the existing number of targets R and the number of new targets.

[0059] Preferably, step S4 specifically includes:

[0060] Observations not used to update any existing targets will generate new Bernoulli elements, i.e., new radar sources, with a certain probability, based on the birth model: ;

[0061] in, .

[0062] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0063] This method incorporates prior knowledge of the environment into the model in a probabilistic form by modeling the existence and parameter evolution of radar radiation sources as labeled Bernoulli random finite sets. The method exhibits adaptability to the number of radiation sources and parameter variations, and works effectively with only weak parameter priors. While maintaining a structured model of physical reality, it achieves weak prior adaptation to complex and unknown environments. Attached Figure Description

[0064] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0065] Figure 1 This is a flowchart of the method of the present invention;

[0066] Figure 2 This is a flowchart of the label-based multi-Bernoulli filter of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Please see Figures 1-2 The present invention provides the following technical solution:

[0069] Example 1: As Figure 1 , Figure 2 As shown in the figure, the specific process of a radar signal sorting method based on a tag-based multi-Bernoulli filter in this embodiment is as follows:

[0070] S1, model each potential radar radiation source as a tagged Bernoulli element, each Bernoulli element including a uniquely determined tag l, the survival rate r of the Bernoulli element, and a radar pulse descriptor;

[0071] For each state Add tags This yields a labeled state vector. The label set of set X is composed of... Given. If X and If the cardinality is equal, then each element of the random finite set X has a different label. Therefore... These are indicators of different labels. The LMB random finite set consists of a parameter set. Given, where L is the label space. The density of the LMB random finite set is given by the following equation: ;

[0072] in .

[0073] For a delta-generalized labeled Bernoulli (δ-GLMB) random finite set, its density is given by the following equation:

[0074] ;

[0075] Among the tags For tags, For each target in the current hypothesis, the measurement-target association history is used. The spatial distribution of targets depends only on the association history. Targets with associated histories have the same spatial distribution in each tag set I.

[0076] S2, after detecting the next signal, predict the state distribution and survival rate of each existing Bernoulli element according to the motion model. The state transition matrix of the motion model is an identity matrix. The prediction is performed by the following formula:

[0077] ;

[0078] ;

[0079] ;

[0080] in As a label, it is unaffected by predictions. Represents the state-related survival probability, trajectory The survival probability is determined by Give, The single-target transfer density is determined by Provided.

[0081] The update step requires predicting the multi-object density in δ-GLMB form. To reduce computational complexity, a standard gating algorithm is used to divide the targets and measurements into statistically independent groups. For simplicity, group index i is omitted below. The predicted δ-GLMB density is as follows:

[0082] ;

[0083] in Represents the predicted label space and And there are

[0084] ;

[0085] ;

[0086] Since the target tag must be a persistent tag Space or new target The space, therefore for each label , or .

[0087] S3 combines the predicted LMB distribution with the observations to produce an updated and more accurate LMB posterior distribution. The δ-GLMB prediction is updated as follows:

[0088] .

[0089] in Represents the mapping space .and

[0090] ;

[0091] ;

[0092] ;

[0093] The spatial likelihood is given by the following equation:

[0094] ;

[0095] in This represents the state-related detection probability of the target. It is the probability of not being detected. This represents the spatial clutter intensity of the Poisson distribution.

[0096] For closed recursion, the LMB filter passes through a parameter... The LMB density is used to approximate the multi-object posterior density, where

[0097] ;

[0098] ;

[0099] ;

[0100] The optimal posterior hypothesis can be obtained by solving the extended cost matrix, which includes not only the uncertainty of the measurement association but also the appearance and disappearance of objects.

[0101] First, enumeration , and For each pair Define a P tuple ,in

[0102] ;

[0103] To obtain tags in the following ways and :

[0104] ;

[0105] ;

[0106] If for , and We define:

[0107] ;

[0108] in Corresponding to a persistent goal, This indicates the goals for new students. Furthermore... Indicates the index of the measurement, where This indicates the disappearance of the target. This indicates a missed detection.

[0109] Thus, the formula is: ;

[0110] It can be rewritten as: ;

[0111] in By solving: ;

[0112] To obtain the optimal allocation. For the allocation matrix, The cost matrix;

[0113] ;

[0114] S4, which is not used to update any existing target observations, will generate new Bernoulli elements (i.e., new radar sources) with a certain probability according to the birth model. The LMB distribution of the newborn object is given by the following formula:

[0115] ;

[0116] in, ;

[0117] S5, Bernoulli elements that fail to be associated with any observation, have a reduced survival rate. When the survival rate falls below a certain threshold, they are removed in the next step. This threshold is set to 0.1.

[0118] Through the above steps, this invention achieves signal sorting under structured weak prior conditions. To verify the effectiveness of the method, a simulation experiment was conducted in a typical radar signal environment. Seven different radar signal sources were set up in the experiment, and the sorting was performed using this method. The results are shown in Table 1.

[0119] Table 1. Radar signal sorting performance indicators of the method of the present invention

[0120]

[0121] As can be seen from Table 1, the method of the present invention achieves a purity of 0.849, indicating that the intra-cluster purity of the sorting results is high; the overclustering ratio is 0.0, indicating that the method can completely preserve the pulse sequence of each radar signal and does not have the problem of incorrectly splitting a real radar signal into multiple signals, which is of great significance for maintaining the characteristic integrity of radar signals.

[0122] It is worth noting that the Normalized Mutual Information (NMI) and Adjusted Rand Index (ARI) are relatively low. Analysis reveals this is due to label permutation between the cluster labels output by the clustering algorithm and the actual radar labels. Specifically, while the method of this invention can accurately separate pulses from different radars (proven by high purity), there is a mapping misalignment between the cluster numbers automatically assigned by the algorithm and the preset radar numbers, resulting in low NMI and ARI indices based on label consistency. This label permutation phenomenon is common in unsupervised clustering evaluation and does not affect the actual sorting ability of the method.

[0123] In summary, the method of this invention can effectively separate pulse signals from different radars, and has the advantages of high purity and less over-clustering, and has good application prospects in the field of radar signal sorting.

[0124] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A radar signal sorting method based on tag-based multi-Bernoulli filters, characterized in that, Includes the following steps: S1. Model each potential radar radiation source as a tagged Bernoulli element; wherein the Bernoulli element includes a uniquely identified tag, the survival rate of the Bernoulli element, and a radar pulse descriptor. S2. After detecting the next signal, predict each existing Bernoulli element according to the motion model, and update the state distribution and survival rate of each Bernoulli element. S3. Combine the predicted LMB distribution with the observations to generate the LMB posterior distribution; calculate the likelihood of each "predicted target-observation" pair and determine the most likely target of the observation in a probabilistic manner; the survival rate of Bernoulli elements that are successfully associated with the observations will increase, and the state distribution will be corrected according to the observations. S4. Observations that were not used to update any existing targets are used to generate new Bernoulli elements, i.e., new radar radiation sources, with a certain probability according to the birth model. S5. Bernoulli elements that fail to be associated with any observation will have a reduced survival rate. When the survival rate is below a threshold, they will be removed in the next step.

2. The radar signal sorting method based on a tag-based multi-Bernoulli filter as described in claim 1, characterized in that, Step S1 specifically includes: Each potential radar source is modeled as a labeled Bernoulli element, and for each state... Add tags This yields a labeled state vector. The label set of set X is composed of Given; if X and If the potentials are equal, then each element of the random finite set X has a different label. Therefore These are indicators of different labels; the LMB random finite set consists of a parameter set. Given, where L is the label space; The density of an LMB random finite set is: ; in, ; For a delta-generalized labeled Bernoulli random finite set, its density is: ； Among the tags For tags, For each target in the current hypothesis, the measurement-target association history is provided.

3. The radar signal sorting method based on a tag-based multi-Bernoulli filter as described in claim 1, characterized in that, Step S2 specifically includes: Predicting each existing Bernoulli element based on the motion model, wherein the state transition matrix of the motion model is an identity matrix, and the specific prediction formula is as follows: ； ； ； in, As a label, it is unaffected by predictions. Represents the state-related survival probability, trajectory The survival probability is determined by Give, The single-target transfer density is determined by Give; Predicting multi-object density in the form of δ-GLMB during the update step: ； in, Represents the predicted label space and And there are ； ； Since the target tag must be a persistent tag Space or new target The space, therefore for each label , or .

4. The radar signal sorting method based on a tag-based multi-Bernoulli filter as described in claim 1, characterized in that, Step S3 specifically includes: Combining the predicted LMB distribution with observations yields an updated and more accurate posterior LMB distribution: ; in, Represents the mapping space ;and ； ； ； The spatial likelihood is: ; in, This represents the state-related detection probability of the target. It is the probability of not being detected. Indicates the spatial clutter intensity of the Poisson distribution; For closed recursion, the LMB filter passes through a parameter... The LMB density is used to approximate the multi-object posterior density, where, ； ； ； The optimal a posteriori hypothesis is obtained by solving the extended cost matrix, which includes not only the uncertainty of the association of the measurement but also the appearance and disappearance of objects; First, enumeration , and ; for each pair Define a P tuple ,in: ； To obtain tags in the following ways and : ； ； If for , and ,set up: ； in, Corresponding to a persistent goal, Indicates the goals of new students; in addition, Indicates the index of the measurement, where This indicates the disappearance of the target. This indicates a missed detection; Thus, the formula Rewritten as: ;in, ; By solving Achieve optimal allocation; in, For the allocation matrix, The cost matrix; ； Where M is the number of observations, and P is the sum of the existing number of targets R and the number of new targets.

5. The radar signal sorting method based on a tag-based multi-Bernoulli filter as described in claim 1, characterized in that, Step S4 specifically includes: Observations not used to update any existing targets will generate new Bernoulli elements, i.e., new radar sources, with a certain probability, based on the birth model: ; in, .

Images