A signal intelligent sorting method based on Monte Carlo tree search
Through the multi-level sorting method based on Monte Carlo tree search, the hardware resource management and parameter adjustment problems in unknown radar signal sorting are solved, efficient radar signal sorting is achieved, and the sorting accuracy and accuracy are improved.
Patent Information
- Application Number
- CN202211057309.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-31
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-08-31
AI Technical Summary
The existing radar signal sorting algorithm is difficult to effectively process unknown radar signals, especially in hardware resource allocation and parameter adjustment, which leads to unreasonable subspace division and affects the sorting effect.
Using the intelligent signal sorting method based on Monte Carlo tree search, multi-level sorting is performed through parameters such as DOA, encoding, frequency modulation and pulse width, and combined with PRI-RF and RF-PRI sorting, the final result is output in the fusion judgment. Simplifies hardware resource management and improves sorting accuracy.
It realizes effective sorting of unknown radar signals, improves the sorting accuracy, reduces the probability of missing sorting, and is suitable for pulse blind sorting of unknown radar signals in actual engineering applications.
Smart Images

Figure CN115372905B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of signal sorting in electronic reconnaissance technology, and in particular relates to a signal intelligent sorting method based on Monte Carlo tree search. Background Art
[0002] Existing radar signal sorting algorithms include classic algorithms such as the cumulative difference histogram method (CDIF), the sequence difference histogram method (SDIF), the pulse arrival time (TOA) difference histogram method, the repetition interval estimation method based on the pulse repetition interval (RPI) transformation, and a series of improved algorithms based on the CDIF and SDIF algorithms. However, the above algorithms are all pulse sorting algorithms for known radar signals. There is little research on the pulse sorting problem of unknown radar signals.
[0003] The existing sorting algorithms for unknown radar signals mainly include spatial parameter filtering and network clustering sorting algorithms. The principle of spatial parameter filtering is relatively simple and easy to implement in hardware, but how to effectively divide the subspace is a difficult problem. If the subspace division is too large, it will cause a waste of hardware resources, and if the subspace division is too small, the possibility of missing points will be greatly increased.
[0004] The principle of clustering sorting algorithm based on Kohonen network is complex. During the sorting process, the values of parameters such as the initial number of cluster centers N and the adjustable parameter λ of Minkowsky distance will directly affect the sorting results of the algorithm, making it difficult to apply in practice. Summary of the invention
[0005] In order to solve the problems existing in the background technology, the present invention provides a signal intelligent sorting method based on Monte Carlo tree search, and its specific technical solution is as follows:
[0006] A signal intelligent sorting method based on Monte Carlo tree search comprises the following steps:
[0007] S1: The PDW word of each pulse is formed, including the arrival angle DOA, coding method, frequency modulation method, pulse width PW, carrier frequency RF, and pulse repetition interval PRI;
[0008] S2: Pulse sorting is performed based on the four parameters of DOA, coding, frequency modulation, and PW. Different intervals are divided according to the different ranges of the four parameters. The intervals with the same four parameters are determined as one category for pulse search.
[0009] S3: Divide each sorting result in S2 into two branches, and perform PRI-RF sorting and RF-PRI sorting respectively;
[0010] S4: Branch sorting result fusion decision. RF-PRI sorting and PRI-RF sorting are two parallel sortings after the PW level sorting result. The two sorting results are compared and the sorting result is output after comprehensive decision. The same sorting results are merged and only one result is retained.
[0011] S5: Output the sorting result after the final judgment.
[0012] In the preferred scheme, step S1 simulates the jitter error of the receiver's intra-pulse analysis module for various parameters of the target radiation source in actual engineering applications, and performs different degrees of error deviation on DOA, PW, RF, and PRI, among which the DOA deviation is 1°, the PW deviation is 1%, the RF deviation is 5%, and the RPI deviation is 2%.
[0013] In a preferred solution, the process of step S2 includes:
[0014] P1: First-level sorting. The target angle does not change much in a short period of time. DOA is used for first-level sorting. New angle items are dynamically established based on the angle measurement results and accuracy.
[0015] P2: The second level sorting, sorting the DOA output sorting results by coding method, including but not limited to coding and non-coding;
[0016] P3: The third level sorting is to sort the pulse frequency modulation types, including but not limited to fixed point frequency and agile frequency;
[0017] P4: The fourth level of sorting sorts PW and dynamically establishes new angle items based on PW values and PW measurement accuracy.
[0018] Furthermore, the specific process of step P1 includes:
[0019] H1: Round up or down the PDW data with deviation;
[0020] H2: Perform DOA sorting; set the search range to the full range of 360° space, set the search step to 1°, if the DOA of a certain pulse is found, store the position of the DOA in the full search range and count the number of pulses found by the DOA, and set the decision threshold according to 6% of the number of pulses to be sorted;
[0021] H3: Perform peak threshold judgment. If the peak value is greater than the judgment threshold, it is valid sorting, otherwise it is invalid;
[0022] Perform peak interval judgment. If the interval between two adjacent peaks is greater than 1, both peaks are considered to be valid for sorting. Otherwise, only the first peak of the two peaks is retained for sorting.
[0023] H4: Record the location and size of the peak value that passes the decision threshold, which are the sorted DOA value and the corresponding number of sorted pulses respectively; also record the node number to which each pulse is sorted.
[0024] Further, the specific process of step P4 includes:
[0025] J1: Set the search range to 1us to 120us, and the search step to 1us. If a PW of a pulse is found, the position of the PW in the full search range is stored and the number of pulses found by the PW is counted. The decision threshold is set according to 6% of the number of pulses to be sorted.
[0026] J2: Perform peak threshold judgment. If the peak value is greater than the judgment threshold, it is valid sorting, otherwise it is invalid;
[0027] J3: Perform peak interval judgment. If the interval between two adjacent peaks is greater than 1, both peaks are considered to be valid for sorting. Otherwise, only the first peak of the two peaks is retained for sorting.
[0028] J4: Record the location and size of the peak value that passes the decision threshold, which are the sorted DOA value and the corresponding number of sorted pulses respectively; also record the node number to which each pulse is sorted;
[0029] J5: Based on the information of the four-level sorting, the node eigenvalue is constructed using the node number assigned to each pulse in each sorting, and the valid child nodes after the four-level sorting are obtained according to the eigenvalue information of each node.
[0030] Further, the specific process of PRI-RF sorting in step S3 includes:
[0031] Q1: First, PRI sorting is performed, and the SDIF method is used to determine whether PRI sorting is successful. If successful, RF sorting is performed. If unsuccessful, the PW-level sorting results are classified into one category. The judgment basis for whether PRI sorting is successful or not is whether the number of pulses in the sorting result is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails.
[0032] Q2: Perform RF sorting on the PRI sorting results to determine whether the RF sorting is successful. If successful, the results are divided into categories according to the RF sorting results. If unsuccessful, the PRI-level sorting results are divided into one category. The basis for determining whether the RF sorting is successful or not is whether the number of pulses in the sorting results is greater than a preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails.
[0033] Q3: Output PRI-RF sorting results.
[0034] Further, the specific process of performing RF-PRI sorting in step S3 includes:
[0035] R1: First, RF sorting is performed to determine whether RF sorting is successful. If successful, PRI sorting is performed. If unsuccessful, the PW-level sorting results are classified into one category. The judgment basis for whether RF sorting is successful or not is whether the number of pulses in the sorting result is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails.
[0036] R2: Perform PRI sorting on the RF sorting results, and use SDIF to determine whether the PRI sorting is successful. If successful, the results are divided into categories according to the PRI sorting results. If unsuccessful, the RF-level sorting results are divided into one category. The judgment basis for the success of PRI sorting is whether the number of pulses in the sorting results is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails.
[0037] R3: Output RF-PRI sorting results.
[0038] The beneficial effects achieved by the present invention are:
[0039] The present invention uses the Monte Carlo tree search concept, takes the pulse signals from different target radiation sources as the root nodes of the Monte Carlo tree, and takes different search latitudes as the search decision criteria at different levels of the Monte Carlo tree. The method is simple in principle and can realize the effective sorting of unknown radar signals, providing a new method for the blind sorting of pulses of unknown radar signals. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 Flowchart of the signal intelligent sorting method based on Monte Carlo tree search.
[0041] Figure 2 Schematic diagram of the sorting process based on Monte Carlo tree search.
[0042] Figure 3 Histogram of DOA sorting results.
[0043] Figure 4 Histogram of PW sorting results.
[0044] Figure 5 Histogram of DOA, coded, FM, and PW pulse sorting results.
[0045] Figure 6 The sorting accuracy and missed sorting probability under different DOA conditions.
[0046] Figure 7 The sorting accuracy and missed sorting probability under different PW conditions.
[0047] Figure 8 The sorting accuracy and missed sorting probability under different RF conditions.
[0048] Fig. 9The sorting accuracy and missed sorting probability under different PRI conditions. DETAILED DESCRIPTION
[0049] To facilitate those skilled in the art to understand the present invention, the specific implementation of the present invention is described below in conjunction with the attached tables and drawings.
[0050] The PDW and various parameter descriptions to be used in the present invention are shown in Table 1.
[0051] Table 1 Description of parameters for selecting PDW
[0052]
[0053] Reference Figure 1-Figure 9 , a signal intelligent sorting method based on Monte Carlo tree search, is carried out in the following steps:
[0054] S1: The PDW word of each pulse is formed, including the arrival angle DOA, coding method, frequency modulation method, pulse width PW, carrier frequency RF, and pulse repetition interval PRI;
[0055] S2: Pulse sorting is performed based on the four parameters of DOA, coding, frequency modulation, and PW. Different intervals are divided according to the different ranges of the four parameters. The intervals with the same four parameters are determined as one category for pulse search. Specifically: P1: The first level sorting, the target angle does not change much in a short period of time, and DOA is used for the first level sorting. According to the angle measurement results and accuracy, new angle items are dynamically established; H1: PDW data with deviations are rounded up or down; H2: DOA sorting; Set the search range to the full range of 360° space, set the search step to 1°, and if the DOA of a certain pulse is found, Store the position of the DOA in the entire search range and count the number of pulses searched by the DOA, and set the decision threshold according to 6% of the number of pulses to be sorted; H3: Perform peak threshold judgment, if the peak value is greater than the judgment threshold, it is valid sorting, otherwise it is invalid; perform peak interval judgment, if the interval between two adjacent peaks is greater than 1, then both peaks are considered to be valid sorting, otherwise only the previous peak of the two peaks is retained for sorting; H4: Record the position and peak value of the peak that passes the judgment threshold, the two are the sorted DOA value and the corresponding number of sorted pulses respectively; at the same time, record the node number to which each pulse is sorted.
[0056] P2: The second level sorting, sorting the DOA output sorting results by coding method, including but not limited to coding and non-coding;
[0057] P3: The third level sorting is to sort the pulse frequency modulation types, including but not limited to fixed point frequency and agile frequency;
[0058] P4: The fourth level of sorting, sorting PW, dynamically establishing new angle items based on PW values and PW measurement accuracy. The specific process of P4 includes:
[0059] J1: Set the search range to 1us to 120us, and the search step to 1us. If a PW of a pulse is found, the position of the PW in the full search range is stored and the number of pulses found by the PW is counted. The decision threshold is set according to 6% of the number of pulses to be sorted.
[0060] J2: Perform peak threshold judgment. If the peak value is greater than the judgment threshold, it is valid sorting, otherwise it is invalid;
[0061] J3: Perform peak interval judgment. If the interval between two adjacent peaks is greater than 1, both peaks are considered to be valid for sorting. Otherwise, only the first peak of the two peaks is retained for sorting.
[0062] J4: Record the location and size of the peak value that passes the decision threshold, which are the sorted DOA value and the corresponding number of sorted pulses respectively; also record the node number to which each pulse is sorted;
[0063] J5: Based on the information of the four-level sorting, the node eigenvalue is constructed using the node number assigned to each pulse in each sorting, and the valid child nodes after the four-level sorting are obtained according to the eigenvalue information of each node.
[0064] S3: Divide each sorting result in S2 into two branches, and perform PRI-RF sorting and RF-PRI sorting respectively; the specific process of performing PRI-RF sorting in step S3 includes:
[0065] Q1: First, PRI sorting is performed, and the SDIF method is used to determine whether PRI sorting is successful. If successful, RF sorting is performed. If unsuccessful, the PW-level sorting results are classified into one category. The judgment basis for whether PRI sorting is successful or not is whether the number of pulses in the sorting result is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails.
[0066] Q2: Perform RF sorting on the PRI sorting results to determine whether the RF sorting is successful. If successful, the results are divided into categories according to the RF sorting results. If unsuccessful, the PRI-level sorting results are divided into one category. The basis for determining whether the RF sorting is successful or not is whether the number of pulses in the sorting results is greater than a preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails.
[0067] Q3: Output PRI-RF sorting results.
[0068] The specific process of RF-PRI sorting in step S3 includes:
[0069] R1: First, RF sorting is performed to determine whether RF sorting is successful. If successful, PRI sorting is performed. If unsuccessful, the PW-level sorting results are classified into one category. The judgment basis for whether RF sorting is successful or not is whether the number of pulses in the sorting result is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails.
[0070] R2: Perform PRI sorting on the RF sorting results, and use SDIF to determine whether the PRI sorting is successful. If successful, the results are divided into categories according to the PRI sorting results. If unsuccessful, the RF-level sorting results are divided into one category. The judgment basis for the success of PRI sorting is whether the number of pulses in the sorting results is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails.
[0071] R3: Output RF-PRI sorting results.
[0072] S4: Branch sorting result fusion decision. RF-PRI sorting and PRI-RF sorting are two parallel sortings after the PW level sorting result. The two sorting results are compared and the sorting result is output after comprehensive decision. The same sorting results are merged and only one result is retained.
[0073] S5: Output the sorting result after the final judgment.
[0074] Embodiment 1:
[0075] The present invention is explained below in conjunction with a specific embodiment. The simulated data model in this embodiment is derived from a specific application scenario of a certain type of task. In actual applications, the receiver receives pulse signals from different radiation sources that are mixed together in the time domain. Its internal intra-pulse analysis signal processing module is responsible for processing the received signal and outputting the PDW information of each pulse.
[0076] The PDW to be used in the present invention includes DOA, coding mode, frequency modulation mode, PW, RF, and PRI. In the simulation, 8 types of signal radiation sources are set, and the number of pulses of each radiation source is 10,000. It is simulated that 10,000 signal radiation sources radiate signals outward in the same space and at the same time. The DOA, coding mode, frequency modulation mode, PW, RF, and PRI of each pulse radiation source are shown in Table 2 below.
[0077] Table 2 PDW of each pulse radiation source
[0078]
[0079] In the above table, code "0" represents no code, and code "1" represents code. Frequency modulation "0" represents fixed frequency, and frequency modulation "1" represents agile frequency. According to the flow chart, it is implemented as follows:
[0080] 1. Each pulse PDW is generated;
[0081] In order to simulate the jitter error of the receiver's intra-pulse analysis module in estimating the parameters of the target radiation source in actual engineering applications, different degrees of error deviation are applied to DOA, PW, RF, and PRI, with a DOA deviation of 1°, a PW deviation of 1%, a RF deviation of 5%, and a RPI deviation of 2%.
[0082] 2. Pulse sorting by DOA, coding, FM, and PW;
[0083] (2.1) Rounding up or down the PDW data with deviation;
[0084] (2.2) Perform DOA sorting; set the search range to the full range of 360° space, set the search step to 1°, if the DOA of a certain pulse is found, store the position of the DOA in the full search range and count the number of pulses found by the DOA, set the decision threshold to 6% of the number of pulses to be sorted, and draw a histogram of the DOA sorting results, as shown in Figure 3 shown.
[0085] Perform peak threshold judgment. If the peak value is greater than the judgment threshold, it is valid sorting, otherwise it is invalid.
[0086] Perform peak interval judgment. If the interval between two adjacent peaks is greater than 1, both peaks are considered to be effectively sorted. Otherwise, only the first peak of the two peaks is retained.
[0087] Record the peak position and peak value that pass the decision threshold, which are the sorted DOA value and the corresponding number of sorted pulses. Also record the node number to which each pulse is sorted.
[0088] (2.3) Perform coding sorting: sort all pulses by coding method, the value of node 1 is 1 for coding, and the value of node 2 is 0 for no coding. Record the node number to which each pulse is sorted.
[0089] (2.4) Perform frequency modulation sorting: sort all pulses by frequency modulation, the value of node 1 is 1 for agile frequency, and the value of node 2 is 0 for fixed frequency. Record the node number to which each pulse is sorted.
[0090] (2.5) Perform PW sorting; according to actual engineering experience, the search range is set to 1us to 120us, and the search step is set to 1us. If a PW of a certain pulse is found, the position of the PW in the full search range is stored and the number of pulses found by the PW is counted. The decision threshold is set according to 6% of the number of pulses to be sorted, and a histogram of the PW sorting results is drawn, as shown in the figure. Figure 4 shown.
[0091] Perform peak threshold judgment. If the peak value is greater than the judgment threshold, it is valid sorting, otherwise it is invalid.
[0092] Perform peak interval judgment. If the interval between two adjacent peaks is greater than 1, both peaks are considered to be effectively sorted. Otherwise, only the first peak of the two peaks is retained.
[0093] Record the peak position and peak value that pass the decision threshold, which are the sorted DOA value and the corresponding number of sorted pulses. Also record the node number to which each pulse is sorted.
[0094] 3. Based on the information of the four-level sorting in 2, the node characteristic value is formed by using the node number assigned to each pulse in each sorting. The calculation formula of the node characteristic value is as follows:
[0095] The characteristic value of the fourth-level sorting node = 1000×DOA sorting node number + 100×coding sorting node number + 10×frequency modulation sorting node number + 1×PW sorting node number.
[0096] According to the characteristic value information of each node, the valid child nodes after four-level sorting can be obtained. The four-level sorting results are as follows: Figure 5 shown.
[0097] 4. Perform PRI-RF sorting on each child node obtained in 3;
[0098] (4.1) PRI sorting: Using the SDIF method, a decision threshold is set according to 6% of the number of pulses to be sorted to determine whether the PRI sorting is successful. If successful, RF sorting is performed. If unsuccessful, the PW-level sorting results are classified into one category.
[0099] (4.2) Perform RF sorting; perform RF sorting on the PRI sorting results, set a decision threshold according to 6% of the number of pulses to be sorted, and judge whether the RF sorting is successful. If successful, the RF sorting results are divided into categories; if unsuccessful, the PRI level sorting results are divided into one category.
[0100] (4.3) Output PRI-RF sorting results.
[0101] 5. Perform RF-PRI sorting on each child node obtained in 3;
[0102] (5.1) Perform RF sorting: First, perform RF sorting. Set a decision threshold according to 6% of the number of pulses to be sorted to determine whether RF sorting is successful. If successful, perform PRI sorting. If unsuccessful, classify the PW-level sorting results into one category.
[0103] (5.2) Perform PRI sorting on the RF sorting results, using SDIF, and set a decision threshold according to 6% of the number of pulses to be sorted to determine whether the PRI sorting is successful. If successful, the results are divided into categories according to the PRI sorting results. If unsuccessful, the RF level sorting results are divided into one category.
[0104] (5.3) Output RF-PRI sorting results.
[0105] 6. The branch sorting results are merged and judged, the same sorting results are merged, and the sorting results after the final judgment are output. The sorting results are shown in Table 3 below. In the present invention, the definitions of the pre-sorting pulse number, the actual sorting pulse number, the correct sorting pulse number, the missed sorting pulse number, and the wrong sorting pulse number are as follows:
[0106] The number of pre-sorted pulses is the number of pulses actually transmitted by different radar radiation sources captured by the receiver;
[0107] The actual number of sorted pulses is the total number of pulses sorted by different radar radiation sources after the pulse sorting of the present invention;
[0108] The number of correctly sorted pulses is the number of correctly sorted pulses from different radar radiation sources after the pulse sorting of the present invention;
[0109] The number of missed sorting pulses is the number of pulses missed by a radar radiation source after the pulse sorting of the present invention;
[0110] The number of wrongly sorted pulses is the number of pulses wrongly sorted to a certain radar radiation source after the pulse sorting of the present invention.
[0111] The calculation formula of sorting accuracy and missed sorting probability is as follows:
[0112] Sorting accuracy = number of correct sorting pulses / actual number of sorting pulses;
[0113] Missed sorting probability = number of missed sorting pulses / number of pre-sorting pulses.
[0114] Table 3 Sorting results summary
[0115]
[0116] It can be seen from Table 3 that under the conditions of DOA deviation of 1°, PW deviation of 1%, RF deviation of 5%, and PRI deviation of 2%, the sorting accuracy of unknown radar signals can reach more than 90% when this method is used. For individual radar target radiation sources, the sorting accuracy is higher than 99%, which can achieve effective sorting of unknown radar signals.
[0117] 7. Conduct sorting performance simulation analysis under different parameter estimation deviation conditions
[0118] By setting the DOA deviation to 0.1°-3°, PW deviation to 0.1%-2%, RF deviation to 1%-10%, and RPI deviation to 0.1%-5%, the sorting performance of the method under different parameter estimation deviation conditions can be obtained. The simulation results are shown in Figure 6-Figure 9 shown.
[0119] It can be seen from the simulation results that under the conditions that the DOA deviation is not higher than 1°, the PW deviation is not higher than 1%, the RF deviation is not higher than 10%, and the RPI deviation is not higher than 2%, the accuracy of pulse sorting of unknown radar signals using the method of the present invention is higher than 90%, and the probability of pulse leakage sorting is about 10%. The principle and calculation process of the present invention do not involve complex calculations, and provide a new method for blind pulse sorting of unknown radar signals, which has practical engineering application value.
[0120] The above embodiments of the present invention do not constitute a limitation on the protection scope of the present invention. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.
Claims
1. A signal intelligent sorting method based on Monte Carlo tree search, characterized by: It includes the following steps: S1: The PDW word of each pulse is formed, including the arrival angle DOA, coding method, frequency modulation method, pulse width PW, carrier frequency RF, and pulse repetition interval PRI; S2: Pulse sorting is performed based on the four parameters of DOA, coding, frequency modulation, and PW. Different intervals are divided according to the different ranges of the four parameters. The intervals with the same four parameters are determined as one category for pulse search. S3: Divide each sorting result in S2 into two branches, and perform PRI-RF sorting and RF-PRI sorting respectively; S4: Branch sorting result fusion decision. RF-PRI sorting and PRI-RF sorting are two parallel sortings after the PW level sorting result. The two sorting results are compared and the sorting result is output after comprehensive decision. The same sorting results are merged and only one result is retained. S5: Output the sorting result after the final judgment; The specific process of PRI-RF sorting in step S3 includes: Q1: First, PRI sorting is performed, and the SDIF method is used to determine whether PRI sorting is successful. If successful, RF sorting is performed. If unsuccessful, the PW-level sorting results are classified into one category. The judgment basis for whether PRI sorting is successful or not is whether the number of pulses in the sorting result is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails. Q2: Perform RF sorting on the PRI sorting results to determine whether the RF sorting is successful. If successful, the results are divided into categories according to the RF sorting results. If unsuccessful, the PRI-level sorting results are divided into one category. The basis for determining whether the RF sorting is successful or not is whether the number of pulses in the sorting results is greater than a preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails. Q3: Output PRI-RF sorting results; The specific process of RF-PRI sorting in step S3 includes: R1: First, RF sorting is performed to determine whether RF sorting is successful. If successful, PRI sorting is performed. If unsuccessful, the PW-level sorting results are classified into one category. The judgment basis for whether RF sorting is successful or not is whether the number of pulses in the sorting result is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails. R2: Perform PRI sorting on the RF sorting results, and use SDIF to determine whether the PRI sorting is successful. If successful, the results are divided into categories according to the PRI sorting results. If unsuccessful, the RF-level sorting results are divided into one category. The basis for judging whether the PRI sorting is successful or not is whether the number of pulses in the sorting results is greater than the preset threshold. If it is greater than the preset threshold, the sorting is successful, otherwise the sorting fails. R3: Output RF-PRI sorting results.
2. The method for intelligent signal sorting based on Monte Carlo tree search according to claim 1, characterized in that: In step S1, in order to simulate the jitter error of the receiver's intra-pulse analysis module for each parameter of the target radiation source in actual engineering applications, different degrees of error deviation are performed on DOA, PW, RF, and PRI, where the DOA deviation is 1°, the PW deviation is 1%, the RF deviation is 5%, and the RPI deviation is 2%.
3. The method for intelligent signal sorting based on Monte Carlo tree search according to claim 1, characterized in that: The process of step S2 includes: P1: First-level sorting. The target angle does not change much in a short period of time. DOA is used for first-level sorting. New angle items are dynamically established based on the angle measurement results and accuracy. P2: The second level sorting, sorting the DOA output sorting results by coding method, including but not limited to coding and non-coding; P3: The third level sorting is to sort the pulse frequency modulation types, including but not limited to fixed point frequency and agile frequency; P4: The fourth level of sorting sorts PW and dynamically establishes new angle items based on PW values and PW measurement accuracy.
4. The method for intelligent signal sorting based on Monte Carlo tree search according to claim 3, characterized in that: The specific process of step P1 includes: H1: Round up or down the PDW data with deviation; H2: Perform DOA sorting; set the search range to the full range of 360° space, set the search step to 1°, if the DOA of a certain pulse is found, store the position of the DOA in the full search range and count the number of pulses found by the DOA, and set the decision threshold according to 6% of the number of pulses to be sorted; H3: Perform peak threshold judgment. If the peak value is greater than the judgment threshold, it is valid sorting, otherwise it is invalid; Perform peak interval judgment. If the interval between two adjacent peaks is greater than 1, both peaks are considered to be valid for sorting. Otherwise, only the first peak of the two adjacent peaks is retained for sorting. H4: Record the location and size of the peak value that passes the decision threshold, which are the sorted DOA value and the corresponding number of sorted pulses respectively; also record the node number to which each pulse is sorted.
5. The method for intelligent signal sorting based on Monte Carlo tree search according to claim 3, characterized in that: The specific process of step P4 includes: J1: Set the search range to 1us to 120us, and the search step to 1us. If a PW of a pulse is found, the position of the PW in the entire search range is stored and the number of pulses found by the PW is counted. The decision threshold is set according to 6% of the number of pulses to be sorted. J2: Perform peak threshold judgment. If the peak value is greater than the judgment threshold, it is valid sorting, otherwise it is invalid; J3: Perform peak interval judgment. If the interval between two adjacent peaks is greater than 1, both peaks are considered to be valid for sorting. Otherwise, only the first peak of the two adjacent peaks is retained for sorting. J4: Record the location and size of the peak value that passes the decision threshold, which are the sorted DOA value and the corresponding number of sorted pulses respectively; also record the node number to which each pulse is sorted; J5: Based on the information of the four-level sorting, the node eigenvalue is constructed using the node number assigned to each pulse in each sorting, and the valid child nodes after the four-level sorting are obtained according to the eigenvalue information of each node.
