Near-distance friend machine link instant identification method based on Kolmogorov-Arnold network

Abstract

The invention provides a short-distance friend machine link instant identification method based on a Kolmogorov-Arnold network. The method comprises the following steps: receiving an array broadband baseband signal on site and carrying out channel time alignment; performing time-frequency occupancy detection on the broadband array data, and segmenting according to a burst boundary to obtain candidate link segments; the MUSIC arrival angle is calculated for the fragment in a windowing mode, an arrival angle track is formed, the direction stability is calculated and compared with a threshold value, and direction stability gating is carried out; extracting a starting waveform shape, a short-time carrier offset and a power fluctuation rhythm in the fragment through the fragment, and splicing into a link characteristic time sequence; inputting the data into a Kolmogorov-Arnold network to generate a link representation vector, fusing direction stability in a confidence degree judgment head, and outputting a friend machine recognition result in combination with a similarity threshold value of a registered friend machine feature library; generating an avoiding frequency band and an avoiding direction range and restraining a suppression frequency band and a transmitting direction of the countering equipment; according to the method, the reliability of instant identification and avoidance configuration of the friend machine link can be improved.

Description

Technical Field

[0001] This invention relates to the technical field of radio spectrum monitoring and array signal processing, and in particular to a method for real-time identification of short-range friendly links based on the Kolmogorov-Arnold network. Background Technology

[0002] Key defense sites or major event locations typically host multiple types of low-altitude drones and their remote control and image transmission wireless links, including unknown incoming targets, compliant drones such as those used by media or law enforcement, and patrol / interception drones owned by security systems. The electromagnetic environment at these sites is characterized by congestion on the same frequency, short-lived signal bursts, and link parameters that vary depending on the manufacturer, firmware, and regional frequency band configuration. If countermeasures equipment is activated in an omnidirectional or wide-coverage manner, failure to promptly identify registered friendly drone links could lead to uncontrolled forced landings or mission interruptions. Therefore, it is crucial to quickly confirm friendly drone links on-site and support avoidance configuration.

[0003] Current on-site handling typically combines spectrum monitoring with rule base management: energy detection or time-frequency analysis is performed on broadband signals to mark occupied areas and segment suspected link fragments according to burst boundaries; parameters such as center frequency, bandwidth, and duration are extracted from the fragments for comparison with whitelisted frequency bands or known device configurations. Some solutions further employ protocol / standard identification, distinguishing common remote control and image transmission systems through demodulation or feature discrimination; other solutions use RF fingerprinting, constructing fingerprints from features such as link initiation transients, carrier offset, and power changes over time, and obtaining representation vectors through machine learning networks, which are then matched with a registered feature library to output identification results. In array scenarios, angle of arrival estimation can also be performed on the signal for localization or to assist in separation.

[0004] Under conditions of close-range, same-frequency congestion, simultaneous occupancy by multiple sources and multipath reflections can easily lead to aliasing within segments. Relying solely on time-frequency segmentation and static whitelists is insufficient to reliably map to "the same link from the same transmitter." When using only RF fingerprints or learned representations, the lack of constraints on segment spatial consistency makes the extracted transients, carrier offsets, and power rhythms susceptible to contamination by signals from other directions, resulting in unreliable matching in the registration database. When only angle-of-arrival information is used as an aid, short-term fluctuations may make it difficult to establish a basis for determining link identity, and the identification results often fail to form a closed-loop linkage with the frequency band / direction avoidance configuration of countermeasures devices.

[0005] Therefore, a method for real-time identification of close-range friendly machine links that can overcome the shortcomings of the existing technology is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method for real-time identification of close-range friendly links based on the Kolmogorov-Arnold network. The core technical problem to be solved by this application is: how to realize real-time confirmation of registered friendly links for candidate link segments received by the array under conditions of close-range co-frequency congestion and short-term signal bursts in key protection or major event sites, and to directly convert the confirmation result into frequency band and transmission direction avoidance constraints that can be executed by countermeasures equipment, so as to avoid accidentally damaging friendly links.

[0007] The real-time identification method for near-range friendly links based on the Kolmogorov-Arnold network according to embodiments of the present invention includes:

[0008] S1. Receive the broadband baseband signal from the array antenna at the protection site and complete the channel time alignment to obtain the broadband array baseband data;

[0009] S2. Perform time-frequency occupancy detection on the broadband array baseband data and segment it according to burst boundaries to obtain candidate link segments, and determine the center frequency band, occupied bandwidth and segment duration for the candidate link segments;

[0010] S3. Calculate the MUSIC angle of arrival for candidate link segments and form an angle of arrival trajectory along the segment time sequence. Calculate the directional stability based on the angle of arrival trajectory and compare it with the directional stability threshold to obtain the directional gating result. When the directional gating result is passed, extract the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment from the candidate link segment according to the segment time sequence. Then, splice the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment in time sequence to form the link feature time sequence.

[0011] S4. Input the link feature time series into the Kolmogorov-Arnold network. The Kolmogorov-Arnold network transforms the link feature time series into a link representation vector through a decomposable one-dimensional function mapping unit. The directional stability is fused in the confidence discriminant head to obtain the recognition confidence. The similarity between the link representation vector and the registered friendly machine feature library is calculated and compared with the recognition threshold to obtain the friendly machine recognition result. The friendly machine recognition result includes the friendly machine identifier, recognition confidence, and unknown marker.

[0012] S5. Based on the friendly machine identifier in the friendly machine identification result, and combined with the center frequency band, occupied bandwidth and angle of arrival trajectory of the candidate link segment, generate avoidance constraints. The avoidance constraints include the avoidance frequency band and the range of avoidance direction.

[0013] S6. Based on the avoidance constraints, the suppression frequency band and transmission direction of the countermeasure equipment are constrained and configured to generate countermeasure control commands.

[0014] Optionally, S1 is as follows:

[0015] The broadband baseband signals received by the array antennas at the site of major events or key area protection are separated according to the array channels and formed into array channel-level inputs according to the array channel sequence configured on site;

[0016] Using the array channel-level input as the data source, each array channel is time-aligned to obtain time-aligned array channel data;

[0017] The time-aligned array channel data is aggregated according to the sampling order and organized according to the array channel order to generate broadband array baseband data.

[0018] Using broadband array baseband data as output, it provides input data for time-frequency occupancy detection, MUSIC direction estimation module and feature composition module, so that the formation of candidate link segments and the calculation of arrival angle trajectory are carried out on a unified data source.

[0019] Optionally, S2 is as follows:

[0020] The broadband array baseband data is input into the time-frequency occupancy detection, the energy distribution is calculated according to the time window and frequency grid, and the continuous occupancy area is marked according to the occupancy judgment threshold.

[0021] The continuous occupancy area is segmented into sudden boundaries according to the time series, and adjacent occupancy is merged or split according to the boundary segmentation threshold to determine the start and end time of each candidate link segment.

[0022] For each candidate link segment, the center frequency band, occupied bandwidth, and segment duration are calculated. The center frequency band is determined based on the weighted center of the frequency grid, the occupied bandwidth is determined based on the range of continuously occupied frequencies, and the segment duration is determined based on the start and end time difference.

[0023] The candidate link segments, their center frequency band, occupied bandwidth, and segment duration are provided to the MUSIC direction estimation module and feature composition module, so that the arrival angle trajectory and link feature timing are established on a unified segment definition.

[0024] Optionally, S3 specifically refers to:

[0025] Using the array channel data corresponding to the candidate link segment as input, the segment time sequence is divided into time windows, and the data within the time window is organized according to the array channel order. The MUSIC angle of arrival is calculated for each time window to obtain the MUSIC angle of arrival for each time window, and the correspondence between the time window and the array channel is maintained for the continuity determination of the angle of arrival trajectory.

[0026] The arrival angles of MUSIC in each time window are connected sequentially to form the arrival angle trajectory, and the arrival angle trajectory is used as the data source for directional stability calculation, while keeping the start and end times of the arrival angle trajectory consistent with those of the candidate link segments.

[0027] Time series statistics are performed on the arrival angle trajectory to calculate directional stability. The directional stability is obtained based on the time variation range of the arrival angle trajectory and the time proportion of the dominant direction. The directional stability reflects the directional consistency of the arrival angle within the segment duration.

[0028] The directional stability is compared with the directional stability threshold to obtain the directional gating result. The directional gating result is used to distinguish between candidate link segments dominated by a single direction and candidate link segments with mixed multiple directions.

[0029] When the direction gating result is passed, the continuous sampling at the beginning of the passed candidate link segment is used as input to extract the shape of the initial power-on waveform, and the sampling order is kept consistent with the segment timing. The shape of the initial power-on waveform is used as the first segment of the link feature timing.

[0030] When the direction gating result is passed, the carrier offset is calculated by dividing the time window according to the duration of the segment, and the short-time carrier offset is obtained. The short-time carrier offset is based on the carrier offset of each time window and is kept aligned with the time sequence of the initial power-on waveform.

[0031] When the directional gating result is passed, the passed candidate link segment is used as input, and the power fluctuation rhythm within the segment is calculated according to the same time window. The initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment are spliced ​​together in time order to form the link characteristic timing sequence. The link characteristic timing sequence is not incorporated into the directional stability and is used as the input of the Kolmogorov-Arnold network. The directional stability is used as an independent input of the confidence discriminant head.

[0032] Optionally, time-series statistics are performed on the arrival angle trajectory, directional stability is calculated, and the directional stability is compared with a directional stability threshold to obtain a directional gating result. The directional stability is calculated by a joint function based on the time variation range of the arrival angle trajectory and the time proportion of the dominant direction. Specifically, the joint function is:

[0033] ;

[0034] in, Indicates directional stability. Indicates the total number of time windows. Indicates the time window index. Indicates the first The arrival angle of the music within a time window Indicates the dominant direction. This indicates the threshold for determining whether the directions are consistent. Represents the absolute value of the angle difference, used to characterize the magnitude of the angle deviation. This indicates an indicator function; it takes the value of 1 if the condition within the parentheses is true, and takes the value of zero if the condition is false. This indicates the range of time variation in the arrival angle trajectory. This represents the normalized upper limit of the range of time variation. This indicates that the larger of the values ​​within the parentheses is selected.

[0035] Optionally, S4 specifically refers to:

[0036] The link feature time series is fed as input into the input layer of the Kolmogorov-Arnold network, while maintaining the consistency of the 128-dimensional components of the link feature time series with the time order.

[0037] In the first level of decomposable one-dimensional function mapping layer, the connection from each input neuron to each hidden neuron is decomposable one-dimensional function mapping unit. Each decomposable one-dimensional function mapping unit consists of 8 one-dimensional basis function neurons and outputs 1 scalar response. 64 hidden neurons are set to weight and converge 128 scalar responses to obtain 64-dimensional intermediate representation.

[0038] In the second-level decomposable one-dimensional function mapping layer, a 64-dimensional intermediate representation is used as input, and a decomposable one-dimensional function mapping unit with the same structure as the first level is connected to 32 output neurons, and the output layer obtains a 32-dimensional link representation vector.

[0039] The 32-dimensional link representation vector and directional stability are input into the confidence discriminant head, which has an input dimension of 33. The confidence discriminant head includes a fully connected layer with 16 hidden neurons and an output layer with 1 output neuron. It performs directional stability discrimination on the link representation vector and outputs the recognition confidence.

[0040] The similarity between the link representation vector and each 32-dimensional registered vector in the registered friend machine feature library is calculated, and the maximum similarity is selected and compared with the recognition threshold to form a similarity judgment result;

[0041] The similarity judgment result and the recognition threshold are used to generate the friendly machine recognition result. When the maximum similarity meets the recognition threshold, a friendly machine identifier is given and the unknown is marked as no. When the maximum similarity does not meet the recognition threshold, the unknown is marked as yes and the recognition confidence is retained.

[0042] The friendly machine identifier, identification confidence level, and unknown marker in the friendly machine identification results are used as outputs for the generation of avoidance constraints and the configuration of countermeasure control commands.

[0043] Optionally, the similarity between the link representation vector and each 32-dimensional registered vector in the registered friend machine feature library is calculated, and the maximum similarity is selected and compared with the recognition threshold to form a similarity judgment result. The similarity is calculated by a metric function on the link representation vector and the 32-dimensional registered vector and weighted by the recognition confidence. The metric function is specifically:

[0044] ;

[0045] Among them, Indicates the candidate link segment index is The link representation vector and the registration item index are The similarity between the registered vectors, Indicates the vector component index. Link representation vector variables The One portion, Represents the registration vector variable The One portion, Indicates the confidence level of identification. This represents the confidence-weighted threshold. This indicates that the larger of the values ​​within the parentheses is taken, which is used to limit the confidence weight to the range of zero to one.

[0046] Optional, S5 specifically includes:

[0047] The friendly aircraft identification result is used as input to parse the friendly aircraft identifier. The center frequency band, occupied bandwidth and angle of arrival trajectory of the candidate link segment are used as the basis for generating avoidance constraints. The friendly aircraft identifier is used as the association key to distinguish avoidance strategies.

[0048] The center frequency band and the occupied bandwidth are defined. The center frequency band is used as the frequency band center. The frequency range of the avoidance band is determined according to the upper and lower boundaries of the occupied bandwidth, and consistency with the center frequency band is maintained.

[0049] Using the time window sequence of the arrival angle trajectory as input, the dominant direction is extracted and the time variation range is calculated. The avoidance direction range is determined based on the dominant direction and the time variation range, and the consistency with the arrival angle trajectory is maintained.

[0050] The avoidance frequency band and avoidance direction range are combined to form an avoidance constraint, which is then associated with the friendly aircraft identifier for use in configuring the suppression frequency band and transmission direction constraints of the countermeasure equipment.

[0051] Optionally, step S6 specifically includes:

[0052] Using avoidance constraints as input, we analyze the avoidance frequency band and avoidance direction range, and establish constraint mapping with countermeasure devices as configuration objects.

[0053] The suppression frequency band of the countermeasures equipment is constrained and configured, the avoidance frequency band is set as the prohibited range of the suppression frequency band, and the suppression frequency band is set outside the prohibited range;

[0054] The transmission direction of the countermeasures device is constrained and configured, the avoidance direction range is set as the prohibited range of the transmission direction, and the transmission direction is set outside the prohibited range. At the same time, the suppression frequency band and the transmission direction are jointly constrained.

[0055] The joint constraint configuration is converted into counter-control instructions for execution.

[0056] The beneficial effects of this invention are:

[0057] (1) This proposal proposes an improved orientation stability gating method based on MUSIC angle of arrival trajectory. By calculating the MUSIC angle of arrival in windowed form within candidate link segments and forming the angle of arrival trajectory, the orientation stability is obtained by constructing a joint function based on the "dominant orientation time ratio" and the "angle of arrival time variation range". The gating result is then compared with the orientation stability threshold. This design enables link feature extraction to be performed only on segments that meet the orientation consistency condition, thereby reducing the interference of multi-source superposition, reflection and mixed signals on fingerprint features in the same-frequency congestion scenario from the source. At the same time, the orientation stability is not incorporated into the link feature vector, but is used as an independent input to the subsequent confidence discrimination head. This allows spatial consistency information to be used for both segment selection and final confidence formation, which is different from the existing processing paths that only perform orientation positioning or fingerprint matching without spatial consistency constraints.

[0058] (2) This proposal proposes a novel link representation learning and confidence fusion method based on the Kolmogorov-Arnold network. The initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment are concatenated in time sequence to form a 128-dimensional link feature time series. A two-level decomposable one-dimensional function mapping structure is used to map it into a 32-dimensional link representation vector. Each connection is transformed into a trainable one-dimensional nonlinear transformation by a mapping unit containing 8 one-dimensional basis function neurons. The representation is then weighted and converged by hidden neurons to obtain a compact representation. Subsequently, the 32-dimensional representation vector is concatenated with the directional stability to form a 33-dimensional input. The confidence is output by the confidence discriminator. When calculating the similarity with the registered friend machine feature library, confidence weighting is introduced. Combined with the recognition threshold and unknown labeling mechanism, the recognition output simultaneously includes the friend machine identifier, recognition confidence, and unknown determination. Thus, under the condition of short burst links, the "similarity matching result" and "spatial consistency confidence" are expressed in a unified way, which is different from the existing practice that makes it difficult to reflect the reliability of the segment by making decisions based on a single similarity threshold.

[0059] (3) This proposal puts forward a closed-loop method for real-time identification and countermeasure avoidance of close-range friendly links. First, channel time alignment is completed at the array receiver to form a unified broadband array baseband data source. Then, candidate link segments and their center frequency band, occupied bandwidth and segment duration are determined by time-frequency occupancy detection and burst boundary segmentation. This ensures that the direction-finding trajectory, characteristic timing and segment boundary are consistent, reducing the uncertainty caused by cross-module time drift and inconsistent segment definitions. After outputting the friendly link identification results, the avoidance frequency band and avoidance direction range are generated by combining the segment center frequency band / occupied bandwidth and the angle of arrival trajectory. These are used to constrain the suppression frequency band and transmission direction of the countermeasure equipment to form an executable countermeasure control command. This closed loop links "friendly link confirmation" and "suppression configuration avoidance" on the same segment evidence chain, so that on-site handling can be constrained and configured around the goal of "avoiding accidental damage to registered friendly links", rather than just staying at the level of detection or identification result display. Attached Figure Description

[0060] 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:

[0061] Figure 1 This is a flowchart of a method for real-time identification of near-range friendly machine links based on the Kolmogorov-Arnold network proposed in this invention;

[0062] Figure 2 This is a flowchart of the channel time alignment and broadband array baseband data generation process for a method for real-time identification of near-range friendly links based on Kolmogorov-Arnold networks proposed in this invention.

[0063] Figure 3 The flowchart shows the time-frequency occupancy detection and candidate link segment formation of a method for real-time identification of near-range friendly links based on Kolmogorov-Arnold network proposed in this invention.

[0064] Figure 4 The flowchart shows the MUSIC directional stability gating and link feature timing structure of a real-time identification method for near-range friendly machine links based on Kolmogorov-Arnold networks proposed in this invention.

[0065] Figure 5 This is a flowchart of the Kolmogorov-Arnold network identification and similarity determination process for a real-time identification method for near-range friendly machine links based on the Kolmogorov-Arnold network proposed in this invention.

[0066] Figure 6This is a flowchart illustrating the obstacle avoidance constraint generation process for a real-time identification method for near-range friendly machine links based on Kolmogorov-Arnold networks proposed in this invention.

[0067] Figure 7 This is a flowchart illustrating the countermeasure device constraint configuration and countermeasure control command generation process for a real-time identification method for near-range friendly machine links based on Kolmogorov-Arnold networks proposed in this invention.

[0068] Figure 8 This is a schematic diagram of the on-site spatial closed-loop avoidance countermeasure of a near-range friendly machine link real-time identification method based on Kolmogorov-Arnold network proposed in this invention.

[0069] Figure 9 This is a schematic diagram of the Kolmogorov-Arnold Network architecture, which is the basis for the proposed method of real-time identification of near-range friendly machine links based on the Kolmogorov-Arnold Network. Detailed Implementation

[0070] In Example 1, reference Figures 1 to 9 A method for real-time identification of close-range friendly links based on the Kolmogorov-Arnold network, comprising:

[0071] S1. Receive the broadband baseband signal from the array antenna at the protection site and complete the channel time alignment to obtain the broadband array baseband data;

[0072] S2. Perform time-frequency occupancy detection on the broadband array baseband data and segment it according to burst boundaries to obtain candidate link segments, and determine the center frequency band, occupied bandwidth and segment duration for the candidate link segments;

[0073] S3. Calculate the MUSIC angle of arrival for candidate link segments and form an angle of arrival trajectory along the segment time sequence. Calculate the directional stability based on the angle of arrival trajectory and compare it with the directional stability threshold to obtain the directional gating result. When the directional gating result is passed, extract the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment from the candidate link segment according to the segment time sequence. Then, splice the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment in time sequence to form the link feature time sequence.

[0074] S4. Input the link feature time series into the Kolmogorov-Arnold network. The Kolmogorov-Arnold network transforms the link feature time series into a link representation vector through a decomposable one-dimensional function mapping unit. The directional stability is fused in the confidence discriminant head to obtain the recognition confidence. The similarity between the link representation vector and the registered friendly machine feature library is calculated and compared with the recognition threshold to obtain the friendly machine recognition result. The friendly machine recognition result includes the friendly machine identifier, recognition confidence, and unknown marker.

[0075] S5. Based on the friendly machine identifier in the friendly machine identification result, and combined with the center frequency band, occupied bandwidth and angle of arrival trajectory of the candidate link segment, generate avoidance constraints. The avoidance constraints include the avoidance frequency band and the range of avoidance direction.

[0076] S6. Based on the avoidance constraints, the suppression frequency band and transmission direction of the countermeasure equipment are constrained and configured to generate countermeasure control commands.

[0077] In this embodiment, step S1 specifically includes:

[0078] The broadband baseband signal received by the array antenna is defined as a variable. Define the number of array channels as a variable. Define the array channel order configured on-site as a variable. Define the raw data segment of each array channel as a variable. ,in Indicates the array channel index, using an array channel separation device for... Perform channel separation, by variable The data from each channel is organized into array channel-level inputs in a specific order, ensuring that the array channel-level inputs maintain consistency with the field configuration in terms of channel order and sampling order.

[0079] To achieve channel time alignment, the reference channel is defined as a variable. Define the synchronization marker position of each array channel as a variable. , with variables As input, the synchronization marker position is obtained using short-time energy detection, specifically by defining the short-time energy sequence as a variable. The time window length for short-time energy calculation is defined as a variable. Define the synchronization determination threshold as a variable. , for variables The sampling sequence according to the time window length Divide and calculate variables Time series, and variables To determine the threshold, the first sampling index of the rising edge of the short-time energy is used as a variable. Synchronization marker variable of reference channel Based on the reference channel, the channel time offset is estimated according to the difference in the synchronization mark sampling index between each channel and the reference channel. The channel time offset is defined as a variable. According to variables For variables The sampling index is shifted and interpolated to maintain waveform continuity and frequency content, resulting in time-aligned array channel data. This aligns all channels on a global sampling time reference while preserving the array channel order and variables. Consistent;

[0080] The time-aligned array channel data is aggregated according to the sampling order and organized according to the array channel order to generate broadband array baseband data. The sampling length of the data block is defined as a variable. Define the broadband array baseband data as a variable. ,variable Structurally includes Each array channel and Each sampling point, according to variable The channels are arranged sequentially, and the sampling points are organized chronologically to ensure cross-channel temporal consistency, which is directly referenced in subsequent processing stages. To guarantee linear connection with subsequent steps, variables are... As a unified data source, it is provided to the time-frequency occupancy detection module, the MUSIC orientation estimation module, and the feature construction module, respectively. Time-frequency occupancy detection uses variables... The input is used for occupancy determination and burst boundary segmentation, ensuring that candidate link segments are formed based on the same data source. The MUSIC direction estimation module uses variables... The array channel data is used as input to calculate the angle of arrival (AHA) and form an AHA trajectory along the segment time sequence, ensuring that the AHA trajectory is consistent with the candidate link segment on the time boundary. The feature construction module uses variables... The initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment are extracted from the input according to the segment time sequence and spliced ​​in time sequence to form the link characteristic time sequence, which establishes the input of the Kolmogorov-Arnold network on the basis of unified sampling and channel alignment.

[0081] Through the above implementation, the formation of candidate link segments, the calculation of arrival angle trajectories, and the extraction of link feature timing are all performed under the same variable. This is done on the same platform to avoid time drift and channel order inconsistencies introduced by cross-data source operations. (Variables) When providing data to the MUSIC direction estimation module, the array channel order and channel time are kept aligned to ensure that the time window of the arrival angle trajectory is aligned with the start and end times of the candidate link segments. When providing data to the feature composition module, the sampling order and segment boundaries are kept consistent to ensure that the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment can be spliced ​​into a fixed-length link feature time sequence in chronological order. The above organization and alignment provide a unified input basis for the Kolmogorov-Arnold network's mapping of link feature time sequences and the confidence discriminator's fusion of direction stability. It also provides consistency assurance for the similarity matching module to perform similarity calculation and identification threshold comparison on the same time reference. Through this specific implementation, the linear flow of broadband array baseband data in acquisition, alignment, and organization meets the requirements for spatial and temporal consistency under conditions of close proximity, co-frequency congestion, and short-term occurrence. This enables subsequent steps to complete the arrival angle trajectory calculation, direction gating, and link feature time sequence composition on a unified and determinable data source.

[0082] In this embodiment, step S2 specifically includes:

[0083] Define the broadband array baseband data as a variable. Define the number of array channels as a variable. Define the sampling rate as a variable. Define the time window length as a variable. Define the frequency grid as a variable Define the frequency grid spacing as a variable. , will the The center frequency of each frequency grid is defined as a variable. Define the occupancy determination threshold as a variable. , for variables By time window length Sliding frame division is performed to obtain a time window sequence. Spectral estimation is then performed on the data from each array channel within each time window. Specifically, a Discrete Fourier Transform is applied to the sampled sequence within the time window, and the squared amplitude is calculated to obtain the power spectrum of each channel. The power spectra of each channel are summed according to the same frequency index to obtain the array synthesized power spectrum for that time window. The array synthesized power spectrum is then gridded according to frequency. Aggregation, specifically, involves summing the power of spectral points falling within the same frequency grid to obtain the energy distribution variable. ,variable Use time window index and frequency raster index as indexes;

[0084] With variables To determine the threshold for occupancy, the variable... The determination is performed on a time-window, frequency-grid basis to generate occupancy marker variables. When the corresponding energy is not less than the variable The occupancy flag is set to occupied when the current state is active, and set to unoccupied otherwise, based on the variable. To form a continuous occupancy area, specifically, adjacent occupancy states are connected into an occupancy time interval on the time window index, and the occupancy time intervals of adjacent frequency grids are spliced ​​on the frequency grid index to obtain a continuous occupancy area that is connected in time and frequency. Each continuous occupancy area includes the occupancy time window index range and the occupancy frequency grid index range.

[0085] The continuous occupancy area is segmented into sudden boundary segments according to the time series, and the boundary segmentation threshold is defined as a variable. ,variable This represents the maximum allowed idle interval duration for merging. For each consecutive occupied area, an occupied time interval sequence is extracted, and the idle interval duration between adjacent occupied time intervals is calculated. The idle interval duration is calculated by multiplying the number of unoccupied time windows between adjacent occupied time intervals by the time window length. We get that when the idle interval duration is no greater than the variable... When adjacent occupied time intervals are merged, if the idle interval length is greater than the variable... At the idle interval, burst boundaries are formed, and the start and end times corresponding to each burst boundary are defined as variables. With variables ,in For the segment index, the start and end times are calculated by multiplying the start and end time window indices by the time window length. And combined with sampling rate The conversion yields that the time range corresponding to each burst boundary and the occupancy frequency grid range of the continuous occupancy area within that time range are collectively defined as candidate link segment variables. ;

[0086] Candidate link segment variables To calculate the center frequency band, occupied bandwidth, and segment duration in units, the center frequency band is defined as a variable. For candidate link segment variables Variables within the covered time window The energy weight of each frequency grid is obtained by summing over the time dimension, and the frequency variable is determined by the center of the frequency grid. To obtain the frequency values, an energy weight is used to perform a weighted average of the frequency values ​​to obtain the variable. Define the occupied bandwidth as a variable In candidate link segment variables Within the occupied frequency grid range, the minimum and maximum center frequencies are taken, and the difference between them is used as a variable. Define the segment duration as a variable. , with variables With variables The difference is obtained from the variable ;

[0087] Candidate link segment variables and its center band variables Bandwidth usage variables and segment duration variable The MUSIC orientation estimation module and feature composition module are provided. The MUSIC orientation estimation module is based on variables. The start and end times are used to divide the time sequence of segments into time windows and calculate the angle of arrival, ensuring that the time boundaries of the angle of arrival trajectory are consistent with the variables. ,variable Consistent, feature-based modules are based on variables The start and end times are extracted to obtain the initial power-on waveform shape, short-time carrier offset and power fluctuation rhythm within the segment, and spliced ​​together to form the link characteristic timing sequence, ensuring that the link characteristic timing sequence and the angle of arrival trajectory are established on a unified segment definition.

[0088] In this embodiment, step S3 specifically includes:

[0089] Define the candidate link segments output in step S2 as variables. ,in For the segment index, the array channel data corresponding to the candidate link segment is defined as a variable. ,variable The data is obtained from broadband array baseband data within the start and end time range of candidate link segments, while maintaining the array channel order. The number of array channels is defined as a variable. The center frequency band of the candidate link segment is defined as a variable. The segment duration of the candidate link segment is defined as a variable. The length of the time window used for segment temporal division is defined as a variable. Define the total number of time windows as a variable. ,variable Based on segment duration With time window length The divisibility relation is determined for the variable. By time window length The process involves segmentation to obtain a time window sequence arranged in time fragments. For each time window, the sequence is organized according to the array channel order. Channel observation data ensures a one-to-one correspondence between the array channel data and the array channel sequence within each time window;

[0090] For each time window, calculate the MUSIC angle of arrival, and then... The array covariance matrix for each time window is defined as a variable. ,in For time window index, variable Within this time window The channel observation data is obtained by multiplying the sampling points and averaging them over a time window, for the variables. Perform eigenvalue decomposition, dividing the signal subspace and noise subspace according to the magnitude of the eigenvalues, and define the noise subspace as a variable. Define the angle scan grid as a variable. Construct an array guide vector for each angle in the angle scan grid, and then connect the array guide vector with the variable. Orthogonality calculations were performed to obtain the MUSIC spectral value sequence. The angle with the largest spectral value was selected as the MUSIC arrival angle for that time window, and this MUSIC arrival angle was defined as a variable. By maintaining variables Time window index The correspondence between the time window and the array channel remains unchanged and is used to determine the continuity of the arrival angle trajectory;

[0091] The arrival angles of the music in each time window are concatenated sequentially to form the arrival angle trajectory, and the arrival angle trajectory is defined as a variable. ,variable Depend on to Obtained by arranging in chronological order of time windows, while preserving variables. Start and end times and candidate link segment variables The start and end times are consistent, using variables As a data source for directional stability calculation, for variables In time series statistics, the dominant direction is defined as a variable. By adjusting variables Angular scanning grid Histogram analysis was performed on the above, and the angle that appeared most frequently was selected as the variable. Define the direction consistency determination threshold as a variable. The time range of the arrival angular trajectory is defined as a variable. , where variables By variables The difference between the maximum and minimum angles of arrival is obtained, and the normalized upper limit of the time variation range is defined as a variable. ,variable Angle-scanned grid The difference between the maximum and minimum scan angles is used to normalize the angle change into a dimensionless weight, and the directional stability is defined as a variable. A joint function is used to jointly calculate the time percentage of the dominant direction and the time variation range of the arrival angle trajectory:

[0092] ;

[0093] in, Indicates directional stability. Indicates the total number of time windows. Indicates the time window index. Indicates the first The arrival angle of the music within a time window Indicates the dominant direction. This indicates the threshold for determining whether the directions are consistent. Represents the absolute value of the angle difference, used to characterize the magnitude of the angle deviation. This indicates an indicator function; it takes the value of 1 if the condition within the parentheses is true, and takes the value of zero if the condition is false. This indicates the range of time variation in the arrival angle trajectory. This represents the normalized upper limit of the range of time variation. This indicates that the larger of the values ​​within the parentheses is selected.

[0094] Define the directional stability threshold as a variable. Define the direction gating result as a variable. , will variables With variables Comparison, when variables Not less than variable Time variable Set to pass when variable Less than variable Time variable Set to fail, thereby distinguishing candidate link segment variables dominated by a single direction. Candidate link segment variables with multi-directional mixing Only when the variable To ensure that the link feature timing extraction process is initiated upon successful completion, the link feature timing is established based on directional consistency constraints.

[0095] When variables To achieve this, the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment are extracted and spliced ​​together to form a link characteristic timing sequence. The initial power-on waveform shape is defined as a variable. From candidate link segment variables A fixed-length sampling sequence is formed by truncating continuous sampling points at the beginning and then normalizing it according to the amplitude before using it as a variable. The fixed length of the sampling points is sixty-four, making the variable... To characterize the transient waveform shape during the link establishment phase and maintain a sampling order consistent with the segment timing, the short-time carrier offset is defined as a variable. Candidate link segment variables The time window was divided into thirty-two time windows based on the segment length. For each time window, the process was first based on variables. Down-conversion is performed to obtain a baseband complex sampling sequence. Then, conjugate multiplication is performed on adjacent sampling points, and the results are accumulated and averaged within a time window to obtain the average phase increment. The phase value of the average phase increment is obtained using arctangent operation, and combined with the sampling rate, it is converted into the carrier offset value for that time window. The carrier offset values ​​of the thirty-two time windows are arranged in chronological order to form a variable. Define the power fluctuation rhythm within the segment as a variable. , for variables The average power within each of the same thirty-two time windows is calculated, and the thirty-two average power values ​​are normalized and arranged in chronological order to form variables. , make the variable Characterizes the rhythm of power change within a segment duration and is related to variables The time window boundaries are consistent;

[0096] Define link characteristic timing as a variable , will variables ,variable With variables Concatenate them in chronological order to form a fixed-length vector as a variable. , where variables For a 64-dimensional sampling sequence, variables For a 32-dimensional carrier offset sequence, the variables are... For a 32-dimensional power fluctuation sequence, the variables are... The dimension is 128, and the variables are... Variables not included ,variable As input to the Kolmogorov-Arnold network, variables As an independent input to the confidence discriminant head, it is used in subsequent steps, thus enabling the spatial consistency constraint to be continuously referenced in both the generation of orientation gating results and the generation of recognition confidence.

[0097] In this embodiment, step S4 specifically includes:

[0098] Define the link characteristic timing output in step S3 as a variable. ,in For candidate link segment indexes, variables The vector is 128-dimensional, and its 128 components are arranged in chronological order according to the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment. The directional stability output in step S3 is defined as a variable. Define the Kolmogorov-Arnold network as a variable Define the confidence discriminant head as a variable. Define the registered friend machine feature database as a variable. Define the recognition threshold as a variable. In step S41, the variable Variables are fed in order of components. The input layer has 128 input neurons, which receive variables one by one. The 128-dimensional components enable the input neuron index and variables to... The chronological order must remain consistent;

[0099] Perform mapping computation on the first-level decomposable one-dimensional function mapping layer, defining the number of hidden neurons in the first-level hidden layer as a variable. ,in Taking sixty-four, we define the first-level decomposable one-dimensional function mapping unit as a variable. ,in Indicates the input neuron index. This represents the index of the first-level hidden neuron for each input component. With each hidden neuron ,Will Input the corresponding decomposable one-dimensional function mapping unit Eight one-dimensional basis function neurons are set inside a decomposable one-dimensional function mapping unit, so that the eight one-dimensional basis function neurons respectively target... Perform a trainable one-dimensional nonlinear mapping and output eight basis function responses. Weighted summation of these eight basis function responses yields a scalar response, which is then used as the input neuron. To hidden neurons The connection output, hidden neuron The scalar responses from 128 input neurons are weighted and pooled to output a hidden representation component. The outputs of 64 hidden neurons are then arranged in index order to form a 64-dimensional intermediate representation, defined as a variable. ;

[0100] Perform mapping computation on the second-level decomposable one-dimensional function mapping layer, defining the number of second-level output neurons as a variable. ,in Take thirty-two, and define the second-level decomposable one-dimensional function mapping unit as a variable. ,in Representing variables Component index, This represents the index of the second-level output neuron, for the variable. To each output neuron The connections employ a decomposable one-dimensional function mapping unit structure consistent with the first level, where each connection consists of eight one-dimensional basis function neurons generating basis function responses, which are then weighted and converged to obtain a scalar response. The output neuron... The sixty-four scalar responses are weighted and aggregated to output a single link representation component. These thirty-two link representation components are then arranged in index order to form a thirty-two-dimensional link representation vector, defined as a variable. ;

[0101] The link representation vector and directional stability are input into the confidence discriminant head, which outputs the recognition confidence score. The input vector of the confidence discriminant head is defined as a variable. ,variable By variables The thirty-two components and variables The input is obtained by concatenating the data in a fixed order. The input dimension is 33. The confidence discriminant head has a fully connected hidden layer, and the number of hidden neurons is defined as a variable. ,in Take sixteen, the fully connected hidden layer pairs of variables Weighted aggregation and pre-defined nonlinear activation are performed to obtain a 16-dimensional hidden output. An output neuron is set in the output layer. The 16-dimensional hidden output is weighted and aggregated, and then normalized to obtain a one-dimensional recognition confidence score, defined as a variable. and variables The value range is limited to zero to one;

[0102] The similarity matching module calculates the similarity between the link representation vector and the registered friendly machine feature library. The registered friendly machine feature library is defined as a set of several registered items, each containing a friendly machine identifier and a 32-dimensional registration vector. The friend machine identifier for each registered item is defined as a variable. , will the Each registration vector is defined as a variable. , where variables Input variables based on the link characteristics and timing of the corresponding friendly machine link The output is obtained later, and the variable A 32-dimensional vector and its relation to variables With consistent dimensions, the confidence-weighted threshold is defined as a variable. ,variable The value range is set to between zero and one, and less than one. This is used to map the identification confidence score to a weighted similarity coefficient for each registered vector. First calculate the variables separately. With variables The similarity of the registered item is set to zero when the magnitude of either vector is zero. When the magnitudes of both vectors are not zero, the similarity is calculated using a metric function to obtain the similarity value.

[0103] ;

[0104] in, Indicates the candidate link segment index is The link representation vector and the registration item index are The similarity between the registered vectors, Indicates the vector component index. Link representation vector variables The One portion, Represents the registration vector variable The One portion, Indicates the confidence level of identification. This represents the confidence-weighted threshold. This indicates that the larger of the values ​​within the parentheses is taken, which is used to limit the confidence weight to the range of zero to one;

[0105] Traversing variables Of all registered items, the one with the highest similarity was selected as the variable. And record the identifier of the friendly machine that generates the highest similarity, defined as a variable. ;

[0106] variables With variables By comparing the results of the identification of the other machine, the unknown markers are defined as variables. When the variable Not less than variable When, the variable Set to no and output the friend machine identifier as a variable. When the variable Less than variable When, the variable Setting it to not output the friendly machine identifier, and defining the friendly machine identification result as a variable. ,variable At least include identification confidence variables With unknown marker variables When the unknown flag is not specified, the variable containing the friend identifier is included. , will variables This output is provided for use in the configuration of avoidance constraint generation and countermeasure control commands.

[0107] In this embodiment, step S5 specifically includes:

[0108] Define the friendly machine identification result output in step S4 as a variable. ,in For candidate link segment indexing, define the friendly machine identifier as a variable. Define the identification confidence level as a variable. Define the unknown marker as a variable ,variable ,variable ,variable By variables The analysis revealed that the center frequency band of the candidate link segment was defined as a variable. The bandwidth occupied by candidate link segments is defined as a variable. Define the arrival angle trajectory as a variable. , where variables With variables The variable is obtained from the output of step S2. The output from step S3 defines the avoidance constraint as a variable. Define the association key of the avoidance constraint as a variable. ;

[0109] variables Parsing as input, when variables If not, from the variable Reading variables As a friendly machine identifier, and the variable Set as variable , make the variable As the association key for avoidance strategy, when the variable In this case, avoidance constraints associated with the friendly machine identifier are not generated, and the variables are... ,variable With variables The input to the avoidance constraint generation unit serves as the basis for the generation of avoidance constraints, ensuring that the avoidance constraint generation process remains consistent with the candidate link segment definition.

[0110] The center frequency band and occupied bandwidth are defined, and the lower frequency boundary of the avoided frequency band is defined as a variable. Define the upper frequency boundary of the avoidance band as a variable. , with variables As the center of the frequency band, the variables Divide by Obtain the half bandwidth and apply the half bandwidth to the variables respectively. On both sides, we obtain variables With variables , where variables Pick Frequency value after subtracting half bandwidth, variable Pick Adding the frequency value after half the bandwidth makes the center and variable of the avoidance frequency band... Consistency, and ensuring that the avoided frequency band covers the occupied bandwidth of the candidate link segment, will change the variables. With variables Records the frequency range to be avoided, for use by countermeasure equipment when configuring frequency band constraints for suppression.

[0111] Using the time window sequence of the arrival angular trajectory as input, the dominant direction is extracted and the time variation range is calculated. The dominant direction is defined as a variable. Define the range of time variation as a variable. , for variables An angle histogram was performed on the arrival angle sequence within the time window. Each arrival angle was assigned to a corresponding angle interval according to a preset angle step, and the frequency of occurrence was accumulated. The angle with the highest frequency was selected as the variable. , for variables Take the maximum and minimum angles of arrival and calculate the difference to obtain the variable. Define the lower boundary of the angle of the avoidance direction range as a variable. Define the upper angular boundary of the avoidance direction range as a variable. , with variables As the center of direction, the variable Divide by Obtain the half-range and apply the half-range to the variable respectively. On both sides, we obtain variables With variables , where variables Pick The angle value after subtracting half the range, variable Pick Adding the angle value after half the range makes the avoidance direction range related to the variable. The direction of change is consistent with that of the variable, and remains consistent with the direction of change. The correspondence between time window sequences is not disrupted;

[0112] The avoidance frequency band and the avoidance direction range are combined to form avoidance constraints, and the variables are... The data is organized as structured data, and structured data contains at least variables. ,variable ,variable ,variable ,variable With variables and variables With variables Perform associative storage, and store variables The output is sent to the countermeasure equipment configuration interface, enabling the countermeasure equipment to control variables when generating suppression frequency bands and transmission direction commands. With variables As a boundary to avoid suppressing frequency bands, variables With variables As a clearance boundary for the launch direction, and as a variable Use it as a friend machine identifier association key to complete the avoidance constraint configuration.

[0113] In this embodiment, step S6 specifically includes:

[0114] Define the avoidance constraint output in step S5 as a variable. ,in For candidate link segment indexing, define the association key of the avoidance constraint as a variable. Define the lower frequency boundary of the avoidance band as a variable. Define the upper frequency boundary of the avoidance band as a variable. Define the lower boundary of the angle of the avoidance direction range as a variable. Define the upper angular boundary of the avoidance direction range as a variable. Define the set of countermeasures equipment as a variable. , will the Each countermeasure device is defined as a variable. The lower boundary of the available suppression frequency range of the countermeasures equipment is defined as a variable. The upper boundary of the available suppression frequency range of the countermeasures equipment is defined as a variable. The lower boundary of the usable launch direction range of the countermeasure device is defined as a variable. The upper boundary of the range of usable launch directions of the countermeasures device is defined as a variable. Define the frequency protection bandwidth as a variable. Define the direction protection angle width as a variable. ,variable With variables The countermeasure equipment configuration strategy is preset to a fixed value;

[0115] variables As input, the variables are parsed. ,variable ,variable ,variable ,variable Establish constraint mapping with the countermeasure equipment as the configuration object, specifically for variables. Each countermeasure device variable in Initiate a capability parameter read, retrieving the lowest configurable suppression frequency supported by the countermeasures device as a variable. Read the highest configurable suppression frequency supported by the countermeasure device as a variable. Read the minimum launch pointing angle supported by the countermeasure device as a variable. Read the maximum launch pointing angle supported by the countermeasure device as a variable. Define the suppression bandwidth requirement of the countermeasures equipment as a variable. Read the bandwidth parameter from the suppression mode configuration of the countermeasures device as a variable. When the suppression mode configuration does not specify a bandwidth parameter, the variable will be... Set the minimum suppression bandwidth supported by the countermeasures device, write the above analysis results and capability parameters into the same mapping record, and make the key of the mapping record a variable. The value of the mapping record contains variables. ,variable ,variable ,variable ,variable ,variable ,variable ,variable With variables This is used for subsequent constraint configuration;

[0116] Constraints are configured on the suppression frequency bands of the countermeasures equipment, and variables are set for each countermeasures equipment. Set the prohibited range to an interval. And the prohibited range will be narrowed down to the range of suppression frequencies available for countermeasure equipment. Within this range, the permitted suppression frequency bands will be defined as those located outside the prohibited area and within... Within the frequency range, the permitted suppression band consists of a continuous frequency range to the left of the prohibited range and a continuous frequency range to the right of the prohibited range. The width of each of the two continuous frequency ranges is calculated and compared with the variable... For comparison, select an interval with a width not less than the variable. The continuous frequency range is used as a candidate suppression frequency band. When multiple continuous frequency ranges meet the conditions, the continuous frequency range with the largest range width is selected as the candidate suppression frequency band, and the final suppression frequency band is defined as a variable. , where variables With variables By truncating the selected candidate frequency bands with a length as a variable The continuous frequency range is obtained;

[0117] The transmission direction of the countermeasures equipment is constrained and configured, and the suppression frequency band and transmission direction are jointly constrained. This applies to the variables of each countermeasures device. Set the prohibited range for the launch direction as an interval. And the prohibited range will be narrowed down to the range of launch directions that the countermeasures equipment can use. Within this range, the permitted launch direction interval will be defined as being outside the prohibited area and located within... Within the angular range, the permitted launch direction range consists of a continuous angular range to the left of the prohibited range and a continuous angular range to the right of the prohibited range. The launch pointing angle of the countermeasure device is defined as a variable. The midpoint of the selected allowable launch direction range is taken as the variable. and variables Restricted to variables With variables Within a closed interval, the joint constraint is defined as a variable. ,variable Includes variables With variables and variables With variables Perform associated storage;

[0118] The joint constraint configuration is converted into counter-control instructions for execution, and the counter-control instructions are defined as variables. , will variables The organization field is a structured field, and the structured field must at least contain a countermeasure device identifier and an association key variable. Suppression band lower boundary variables Suppression band upper boundary variables , emission pointing angle variable And write the instruction enable flag to trigger execution, and set the variable The output is sent to the control interface of the countermeasure device, and the countermeasure device uses variables... Configure the suppression frequency band and transmission direction and execute.

[0119] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for real-time identification of near-range friendly machine links based on Kolmogorov-Arnold networks, characterized in that, include: S1. Receive the broadband baseband signal from the array antenna at the protection site and complete the channel time alignment to obtain the broadband array baseband data; S2. Perform time-frequency occupancy detection on the broadband array baseband data and segment it according to burst boundaries to obtain candidate link segments, and determine the center frequency band, occupied bandwidth and segment duration for the candidate link segments; S3. Calculate the MUSIC angle of arrival for candidate link segments and form an angle of arrival trajectory along the segment time sequence. Calculate the directional stability based on the angle of arrival trajectory and compare it with the directional stability threshold to obtain the directional gating result. When the directional gating result is passed, extract the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment from the candidate link segment according to the segment time sequence. Then, splice the initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment in time sequence to form the link feature time sequence. S4. Input the link feature time series into the Kolmogorov-Arnold network. The Kolmogorov-Arnold network transforms the link feature time series into a link representation vector through a decomposable one-dimensional function mapping unit. The directional stability is fused in the confidence discriminant head to obtain the recognition confidence. The similarity between the link representation vector and the registered friendly machine feature library is calculated and compared with the recognition threshold to obtain the friendly machine recognition result. The friendly machine recognition result includes the friendly machine identifier, recognition confidence, and unknown marker. S5. Based on the friendly machine identifier in the friendly machine identification result, and combined with the center frequency band, occupied bandwidth and angle of arrival trajectory of the candidate link segment, generate avoidance constraints. The avoidance constraints include the avoidance frequency band and the range of avoidance direction. S6. Based on the avoidance constraints, the suppression frequency band and transmission direction of the countermeasure equipment are constrained and configured to generate countermeasure control commands.

2. The method for real-time identification of near-range friendly links based on Kolmogorov-Arnold networks according to claim 1, characterized in that, S1 specifically refers to: The broadband baseband signals received by the array antennas at the site of major events or key area protection are separated according to the array channels and formed into array channel-level inputs according to the array channel sequence configured on site; Using the array channel-level input as the data source, each array channel is time-aligned to obtain time-aligned array channel data; The time-aligned array channel data is aggregated according to the sampling order and organized according to the array channel order to generate broadband array baseband data. Using broadband array baseband data as output, it provides input data for time-frequency occupancy detection, MUSIC direction estimation module and feature composition module, so that the formation of candidate link segments and the calculation of arrival angle trajectory are carried out on a unified data source.

3. The method for real-time identification of near-range friendly links based on Kolmogorov-Arnold networks according to claim 1, characterized in that, S2 specifically refers to: The broadband array baseband data is input into the time-frequency occupancy detection, the energy distribution is calculated according to the time window and frequency grid, and the continuous occupancy area is marked according to the occupancy judgment threshold. The continuous occupancy area is segmented into sudden boundaries according to the time series, and adjacent occupancy is merged or split according to the boundary segmentation threshold to determine the start and end time of each candidate link segment. For each candidate link segment, the center frequency band, occupied bandwidth, and segment duration are calculated. The center frequency band is determined based on the weighted center of the frequency grid, the occupied bandwidth is determined based on the range of continuously occupied frequencies, and the segment duration is determined based on the start and end time difference. The candidate link segments, their center frequency band, occupied bandwidth, and segment duration are provided to the MUSIC direction estimation module and feature composition module, so that the arrival angle trajectory and link feature timing are established on a unified segment definition.

4. The method for real-time identification of near-range friendly links based on Kolmogorov-Arnold networks according to claim 1, characterized in that, S3 specifically refers to: Using the array channel data corresponding to the candidate link segment as input, the segment time sequence is divided into time windows, and the data within the time window is organized according to the array channel order. The MUSIC angle of arrival is calculated for each time window to obtain the MUSIC angle of arrival for each time window, and the correspondence between the time window and the array channel is maintained for the continuity determination of the angle of arrival trajectory. The arrival angles of MUSIC in each time window are connected sequentially to form the arrival angle trajectory, and the arrival angle trajectory is used as the data source for directional stability calculation, while keeping the start and end times of the arrival angle trajectory consistent with those of the candidate link segments. Time series statistics are performed on the arrival angle trajectory to calculate directional stability. The directional stability is obtained based on the time variation range of the arrival angle trajectory and the time proportion of the dominant direction. The directional stability reflects the directional consistency of the arrival angle within the segment duration. The directional stability is compared with the directional stability threshold to obtain the directional gating result. The directional gating result is used to distinguish between candidate link segments dominated by a single direction and candidate link segments with mixed multiple directions. When the direction gating result is passed, the continuous sampling at the beginning of the passed candidate link segment is used as input to extract the shape of the initial power-on waveform, and the sampling order is kept consistent with the segment timing. The shape of the initial power-on waveform is used as the first segment of the link feature timing. When the direction gating result is passed, the carrier offset is calculated by dividing the time window according to the duration of the segment, and the short-time carrier offset is obtained. The short-time carrier offset is based on the carrier offset of each time window and is kept aligned with the time sequence of the initial power-on waveform. When the directional gating result is passed, the passed candidate link segment is used as input, and the power fluctuation rhythm within the segment is calculated according to the same time window. The initial power-on waveform shape, short-time carrier offset, and power fluctuation rhythm within the segment are spliced ​​together in time order to form the link characteristic timing sequence. The link characteristic timing sequence is not incorporated into the directional stability and is used as the input of the Kolmogorov-Arnold network. The directional stability is used as an independent input of the confidence discriminant head.

5. The method for real-time identification of near-range friendly links based on the Kolmogorov-Arnold network according to claim 4, characterized in that, The arrival angle trajectory is subjected to time series statistics, directional stability is calculated, and the directional stability is compared with a directional stability threshold to obtain the directional gating result. The directional stability is calculated by a joint function based on the time variation range of the arrival angle trajectory and the time proportion of the dominant direction. The joint function is specifically: ； in, Indicates directional stability. Indicates the total number of time windows. Indicates the time window index. Indicates the first The arrival angle of the music within a time window Indicates the dominant direction. This indicates the threshold for determining whether the directions are consistent. Represents the absolute value of the angle difference, used to characterize the magnitude of the angle deviation. This indicates an indicator function; it takes the value of 1 if the condition within the parentheses is true, and takes the value of zero if the condition is false. This indicates the range of time variation in the arrival angle trajectory. This represents the normalized upper limit of the range of time variation. This indicates that the larger of the values ​​within the parentheses is selected.

6. The method for real-time identification of near-range friendly links based on Kolmogorov-Arnold networks according to claim 1, characterized in that, S4 specifically refers to: The link feature time series is fed as input into the input layer of the Kolmogorov-Arnold network, while maintaining the consistency of the 128-dimensional components of the link feature time series with the time order. In the first-level decomposable one-dimensional function mapping layer, the connection from each input neuron to each hidden neuron is decomposable one-dimensional function mapping unit. Each decomposable one-dimensional function mapping unit consists of 8 one-dimensional basis function neurons and outputs 1 scalar response. 64 hidden neurons are set to weight and converge 128 scalar responses to obtain 64-dimensional intermediate representations. In the second-level decomposable one-dimensional function mapping layer, a 64-dimensional intermediate representation is used as input, and a decomposable one-dimensional function mapping unit with the same structure as the first level is connected to 32 output neurons, and the output layer obtains a 32-dimensional link representation vector. The 32-dimensional link representation vector and directional stability are input into the confidence discriminant head, which has an input dimension of 33. The confidence discriminant head includes a fully connected layer with 16 hidden neurons and an output layer with 1 output neuron. It performs directional stability discrimination on the link representation vector and outputs the recognition confidence. The similarity between the link representation vector and each 32-dimensional registered vector in the registered friend machine feature library is calculated, and the maximum similarity is selected and compared with the recognition threshold to form a similarity judgment result; The similarity judgment result and the recognition threshold are used to generate the friendly machine recognition result. When the maximum similarity meets the recognition threshold, a friendly machine identifier is given and the unknown is marked as no. When the maximum similarity does not meet the recognition threshold, the unknown is marked as yes and the recognition confidence is retained. The friendly machine identifier, identification confidence level, and unknown marker in the friendly machine identification results are used as outputs for the generation of avoidance constraints and the configuration of countermeasure control commands.

7. The method for real-time identification of near-range friendly links based on Kolmogorov-Arnold networks according to claim 6, characterized in that, The similarity is calculated between the link representation vector and each 32-dimensional registered vector in the registered friend machine feature library. The maximum similarity is then compared with the recognition threshold to form a similarity determination result. The similarity is calculated by a metric function on the link representation vector and the 32-dimensional registered vector, and weighted by the recognition confidence. The metric function is specifically: ； in, Indicates the candidate link segment index is The link representation vector and the registration item index are The similarity between the registered vectors, Indicates the vector component index. Link representation vector variables The One portion, Represents the registration vector variable The One portion, Indicates the confidence level of identification. This represents the confidence-weighted threshold. This indicates that the larger of the values ​​within the parentheses is taken, which is used to limit the confidence weight to the range of zero to one.

8. The method for real-time identification of near-range friendly links based on Kolmogorov-Arnold networks according to claim 1, characterized in that, S5 specifically refers to: The friendly aircraft identification result is used as input to parse the friendly aircraft identifier. The center frequency band, occupied bandwidth and angle of arrival trajectory of the candidate link segment are used as the basis for generating avoidance constraints. The friendly aircraft identifier is used as the association key to distinguish avoidance strategies. The center frequency band and the occupied bandwidth are defined. The center frequency band is used as the frequency band center. The frequency range of the avoidance band is determined according to the upper and lower boundaries of the occupied bandwidth, and consistency with the center frequency band is maintained. Using the time window sequence of the arrival angle trajectory as input, the dominant direction is extracted and the time variation range is calculated. The avoidance direction range is determined based on the dominant direction and the time variation range, and the consistency with the arrival angle trajectory is maintained. The avoidance frequency band and avoidance direction range are combined to form an avoidance constraint, which is then associated with the friendly aircraft identifier for use in configuring the suppression frequency band and transmission direction constraints of the countermeasure equipment.

9. The method for real-time identification of near-range friendly links based on Kolmogorov-Arnold networks according to claim 1, characterized in that, Step S6 is as follows: Using avoidance constraints as input, we analyze the avoidance frequency band and avoidance direction range, and establish constraint mapping with countermeasure devices as configuration objects. The suppression frequency band of the countermeasures equipment is constrained and configured, the avoidance frequency band is set as the prohibited range of the suppression frequency band, and the suppression frequency band is set outside the prohibited range; The transmission direction of the countermeasures device is constrained and configured, the avoidance direction range is set as the prohibited range of the transmission direction, and the transmission direction is set outside the prohibited range. At the same time, the suppression frequency band and the transmission direction are jointly constrained. The joint constraint configuration is converted into counter-control instructions for execution.

Images