Signal interference suppression method, electronic device, and storage medium

Abstract

This application discloses a signal interference suppression method, electronic device, and storage medium. The method includes: acquiring electromagnetic environment sensing data, wherein the electromagnetic environment sensing data includes multi-dimensional feature data, including spatial domain feature data, polarization domain feature data, frequency domain feature data, and energy domain feature data; performing predictive analysis on the electromagnetic environment sensing data using a target interference analysis model to obtain interference suppression decision results, wherein the target interference analysis model is used to map the multi-dimensional feature data into interference type labels and suppression strategy parameter vectors; and performing interference suppression operations based on the interference suppression decision results. This application solves the technical problems of poor adaptability, poor timeliness, and low accuracy in related signal interference suppression methods.

Description

Technical Field

[0001] This application relates to the field of satellite communication technology, and more specifically, to a signal interference suppression method, electronic device, and storage medium. Background Technology

[0002] With the rapid deployment of low-Earth orbit satellite constellations, spaceborne phased arrays face challenges from complex electromagnetic environments. Among related technologies, traditional static electromagnetic compatibility (EMC) designs cannot respond to new dynamic interference in orbit. While adaptive nulling antennas can cope with dynamic interference, they suffer from long convergence times, severe main lobe distortion, and insufficient utilization of the polarization domain. Cognitive radio spectrum sensing only acquires situational information and is isolated from beamforming execution, lacking millisecond-level closed-loop linkage. In summary, related technologies still suffer from poor adaptability, timeliness, and accuracy.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a signal interference suppression method, an electronic device, and a storage medium to at least solve the technical problems of poor adaptability, poor timeliness, and low accuracy of signal interference suppression methods in the related art.

[0005] According to one aspect of the embodiments of this application, a signal interference suppression method is provided, comprising: acquiring electromagnetic environment sensing data, wherein the electromagnetic environment sensing data includes multi-dimensional feature data, including: spatial domain feature data, polarization domain feature data, frequency domain feature data, and energy domain feature data; performing predictive analysis on the electromagnetic environment sensing data using a target interference analysis model to obtain interference suppression decision results, wherein the target interference analysis model is used to map the multi-dimensional feature data into interference type labels and suppression strategy parameter vectors; and performing interference suppression operations based on the interference suppression decision results.

[0006] Optionally, acquiring electromagnetic environment sensing data includes: performing beam scanning of the entire airspace using a sensing subarray within a sensing time slot to obtain raw scanning data; performing signal conversion operations on the raw scanning data to obtain signal conversion results, wherein the signal conversion operations include at least one of the following: signal amplification operation, signal mixing operation, and signal sampling operation; and performing multi-dimensional feature extraction based on the signal conversion results to obtain electromagnetic environment sensing data.

[0007] Optionally, the target interference analysis model includes a feature preprocessing module and a prediction inference module. The target interference analysis model is used to predict and analyze electromagnetic environment sensing data to obtain interference suppression decision results. This includes: using the feature preprocessing module to normalize the electromagnetic environment sensing data to obtain normalization results; and using the prediction inference module to predict and analyze the normalization results to obtain interference suppression decision results. The prediction inference module is composed of a one-dimensional convolutional neural network.

[0008] Optionally, the spatial domain feature data includes: the azimuth and elevation angles of the interfering signal; the polarization domain feature data includes: the polarization ellipse tilt angle and axial ratio of the interfering signal; the frequency domain feature data includes: the center frequency and bandwidth of the interfering signal; and the energy domain feature data includes: the power intensity and time duty cycle of the interfering signal. The interfering type label includes at least one of the following: narrowband continuous interference, wideband continuous interference, narrowband pulse interference, wideband pulse interference, and spoofing signal interference. The suppression strategy parameter vector includes: recommended null depth, polarization switching flag, time-domain scheduling duty cycle detection threshold, and priority level.

[0009] Optionally, the interference suppression operation performed based on the interference suppression decision results includes: performing spatial beamforming operation and polarization adaptive control operation based on the interference type label and the suppression strategy parameter vector. The spatial beamforming operation is used to calculate a complex weight vector based on the recommended null depth to control the amplitude and phase distribution of the phased array elements. The polarization adaptive control operation is used to adjust the amplitude ratio and phase difference of the biorthogonal polarization channels in the feed network according to the polarization state of the interference signal.

[0010] Optionally, performing interference suppression operations based on the interference suppression decision results includes: in response to the interference type label indicating that the interference signal is pulse interference, performing a time-domain resource cooperative scheduling operation based on the time-domain scheduling duty cycle detection threshold and the time duty cycle of the interference signal, wherein the time-domain resource cooperative scheduling operation is used to generate an RF switch control signal.

[0011] Optionally, the signal interference suppression method further includes: acquiring signal-to-interference-plus-noise ratio (SIR) data after interference suppression operation at a preset frequency; and, in response to the fact that the improvement in SIR data after multiple consecutive interference suppression operations is less than a preset improvement threshold, fine-tuning the output layer weights of the target interference analysis model using online training samples to generate an updated target interference analysis model, wherein the online training samples include: electromagnetic environment sensing data corresponding to preset historical frames and actual suppression strategy parameters.

[0012] According to another aspect of the embodiments of this application, a signal interference suppression device is also provided, comprising: a first acquisition module, configured to acquire electromagnetic environment sensing data, wherein the electromagnetic environment sensing data includes multi-dimensional feature data, including: spatial domain feature data, polarization domain feature data, frequency domain feature data, and energy domain feature data; an analysis module, configured to perform predictive analysis on the electromagnetic environment sensing data using a target interference analysis model to obtain interference suppression decision results, wherein the target interference analysis model is used to map the multi-dimensional feature data into interference type labels and suppression strategy parameter vectors; and an execution module, configured to perform interference suppression operations according to the interference suppression decision results.

[0013] Optionally, the first acquisition module is further configured to: perform beam scanning of the entire spatial domain using a sensing subarray within the sensing time slot to obtain raw scanning data; perform signal conversion operations on the raw scanning data to obtain signal conversion results, wherein the signal conversion operations include at least one of the following: signal amplification operation, signal mixing operation, and signal sampling operation; and perform multi-dimensional feature extraction based on the signal conversion results to obtain electromagnetic environment sensing data.

[0014] Optionally, the target interference analysis model includes a feature preprocessing module and a prediction inference module. The analysis module is also used to: normalize the electromagnetic environment sensing data using the feature preprocessing module to obtain the normalization result; and perform prediction analysis on the normalization result using the prediction inference module to obtain the interference suppression decision result. The prediction inference module is composed of a one-dimensional convolutional neural network.

[0015] Optionally, the spatial domain feature data includes: the azimuth and elevation angles of the interfering signal; the polarization domain feature data includes: the polarization ellipse tilt angle and axial ratio of the interfering signal; the frequency domain feature data includes: the center frequency and bandwidth of the interfering signal; and the energy domain feature data includes: the power intensity and time duty cycle of the interfering signal. The interfering type label includes at least one of the following: narrowband continuous interference, wideband continuous interference, narrowband pulse interference, wideband pulse interference, and spoofing signal interference. The suppression strategy parameter vector includes: recommended null depth, polarization switching flag, time-domain scheduling duty cycle detection threshold, and priority level.

[0016] Optionally, the execution module is further configured to: perform spatial beamforming operation and polarization adaptive control operation according to the interference type label and the suppression strategy parameter vector, wherein the spatial beamforming operation is used to calculate a complex weight vector based on the recommended null depth to control the amplitude and phase distribution of the phased array elements, and the polarization adaptive control operation is used to adjust the amplitude ratio and phase difference of the biorthogonal polarization channels in the feed network according to the polarization state of the interference signal.

[0017] Optionally, the execution module is further configured to: in response to an interference type label indicating that the interference signal is pulse interference, perform a time-domain resource cooperative scheduling operation based on a time-domain scheduling duty cycle detection threshold and the time duty cycle of the interference signal, wherein the time-domain resource cooperative scheduling operation is used to generate an RF switch control signal.

[0018] Optionally, the signal interference suppression device further includes: a second acquisition module, used to acquire signal-to-interference-plus-noise ratio (SIR) data after interference suppression operation at a preset frequency; and a fine-tuning module, used to fine-tune the output layer weights of the target interference analysis model using online training samples in response to the fact that the improvement amount of the SIR data after multiple consecutive interference suppression operations is less than a preset improvement threshold, thereby generating an updated target interference analysis model. The online training samples include: electromagnetic environment sensing data corresponding to preset historical frames and actual suppression strategy parameters.

[0019] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0020] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0021] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0022] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0023] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0024] In this embodiment, electromagnetic environment sensing data containing multi-dimensional feature data in the spatial domain, polarization domain, frequency domain, and energy domain is acquired. Then, the multi-dimensional feature data is mapped into interference type labels and suppression strategy parameter vectors using a target interference analysis model. Based on the decision result, interference suppression operations are performed, achieving the goal of accurately mapping the multi-dimensional features of the electromagnetic environment to the interference suppression strategy. This improves the adaptability, timeliness, and accuracy of signal interference suppression, thereby solving the technical problems of poor adaptability, poor timeliness, and low accuracy of signal interference suppression methods in related technologies. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0026] Figure 1 This is a schematic diagram of an optional signal interference suppression system according to an embodiment of this application;

[0027] Figure 2 This is a flowchart of an optional signal interference suppression method according to an embodiment of this application;

[0028] Figure 3 This is a schematic diagram of an optional one-dimensional convolutional neural network according to an embodiment of this application;

[0029] Figure 4 This is a schematic diagram of an optional signal interference suppression method according to an embodiment of this application;

[0030] Figure 5 This is a structural block diagram of an optional signal interference suppression device according to an embodiment of this application. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] With the rapid deployment of large-scale low-Earth orbit (LEO) constellations and the expansion of frequency bands to higher frequencies, spaceborne communication systems face increasingly complex electromagnetic interference (EMI) challenges. Currently, relevant anti-interference technologies suffer from the following limitations: First, traditional static EMC design relies on pre-launch planning and lacks on-orbit adaptive capabilities, making it difficult to respond to dynamically changing interference sources. Second, while adaptive nulling techniques can cope with dynamic interference, their convergence speed is slow, they are prone to main lobe distortion, and they do not fully utilize the polarization domain degrees of freedom, limiting their suppression effectiveness. Third, relevant spectrum sensing technologies only output situational information and are isolated from beamforming execution, lacking a millisecond-level closed-loop linkage mechanism, resulting in high response latency. Finally, under limited spaceborne resources, traditional algorithms have high computational complexity, and relevant spaceborne artificial intelligence (AI) solutions often use fixed weights, failing to evolve automatically according to changes in the on-orbit environment. Therefore, these technologies struggle to achieve rapid, accurate, and on-orbit adaptive EMI suppression under resource-constrained conditions.

[0034] According to an embodiment of this application, a method embodiment for signal interference suppression is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0035] Figure 1 This is a schematic diagram of an optional signal interference suppression system according to an embodiment of this application, such as... Figure 1 As shown in the figure, the overall architecture of the system consists of an electromagnetic environment sensing module, a lightweight neural network processing unit, a fast beamforming and null generator, a polarization adaptive controller, a time-domain resource cooperative scheduler, an on-orbit incremental learning controller, and a clock synchronization and reference unit.

[0036] The electromagnetic environment sensing module reuses M array elements located at the edge of the spaceborne phased array antenna array, for example, M=8, and is equipped with an independent down-conversion channel including a low-noise amplifier, mixer, intermediate frequency filter, automatic gain control (AGC) amplifier and 14-bit high-speed analog-to-digital converter (ADC). It performs beam scanning of the entire airspace in a dedicated sensing time slot and outputs multi-channel digital intermediate frequency signals.

[0037] The lightweight neural network processing unit is implemented based on a radiation-hardened field-programmable gate array (FPGA) and is internally divided into a feature preprocessing submodule, a one-dimensional convolutional neural network inference engine, and a policy mapping submodule. After receiving the digital signal from the sensing module, this unit directly outputs a suppression policy parameter vector containing the null depth, polarization switching flag, and time-domain scheduling threshold. For example, this application can utilize the policy mapping submodule to look up the suppression parameters in a table.

[0038] The fast beamforming and null generator calculates the amplitude and phase weights of each array element in real time based on the interference arrival angle and recommended null depth output by the neural network. It controls the 6-bit digital phase shifter and 5-bit digital attenuator of each channel through the Serial Peripheral Interface (SPI) bus to form a deep null.

[0039] The polarization adaptive controller is located in the antenna feed network layer and includes independent excitation control circuits for dual polarization channels. It achieves agile polarization state by adjusting the amplitude ratio and phase difference of the two orthogonal polarization channels.

[0040] The time-domain resource cooperative scheduler includes a fast radio frequency switch array and a pulse detection synchronization circuit. Based on the time duty cycle parameter output by the neural network, it quickly disconnects the damaged channel or switches to the backup beam during the duration of the interference pulse and restores communication during the pulse gap.

[0041] The on-orbit incremental learning controller includes a static random-access memory (SRAM) for temporary model parameter storage, a loss calculation unit, and gradient update logic, which is used to receive intermediate features of the neural network and perform on-orbit fine-tuning of the final layer weights.

[0042] The modules described above provide a unified clock reference through clock synchronization and a reference unit to ensure timing synchronization. For example, the unified clock reference can be 100MHz.

[0043] The sensor module's output is connected to the neural network processing unit's input. The neural network processing unit's strategy output is connected to the beamformer, polarization controller, and time-domain scheduler's control inputs, respectively. The beamformer and polarization controller's outputs are connected to the antenna array element's feed network. The time-domain scheduler's output is connected to the control terminals of each channel's RF switches. The on-orbit incremental learning controller's data input is connected to the neural network processing unit's intermediate feature output, and its update output is connected to the neural network processing unit's weight configuration port, thus forming a seamless closed loop from data acquisition to execution feedback.

[0044] Figure 2 This is a flowchart of an optional signal interference suppression method according to an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0045] Step S21: Obtain electromagnetic environment sensing data, wherein the electromagnetic environment sensing data includes multi-dimensional feature data, which includes: spatial domain feature data, polarization domain feature data, frequency domain feature data and energy domain feature data.

[0046] Step S22: Use the target interference analysis model to predict and analyze the electromagnetic environment sensing data to obtain interference suppression decision results. The target interference analysis model is used to map multi-dimensional feature data into interference type labels and suppression strategy parameter vectors.

[0047] Step S23: Perform interference suppression operation based on the interference suppression decision result.

[0048] The aforementioned electromagnetic environment sensing data can be a collection of multi-channel digital intermediate frequency signals formed by the edge array elements of the spaceborne phased array antenna in a dedicated sensing time slot, after down-conversion, filtering and analog-to-digital conversion. This data reflects the original physical form of electromagnetic signals in various directions of space at the current moment.

[0049] The aforementioned multidimensional feature data can be a structured data vector generated after performing feature extraction algorithms on electromagnetic environment sensing data. This vector integrates quantitative information of interference signals in four dimensions: spatial angle, polarization state, frequency distribution, and energy intensity, providing standardized input for subsequent intelligent analysis.

[0050] The aforementioned spatial characteristic data can be the angle of arrival parameters of the interference signal obtained by performing covariance matrix eigenvalue decomposition and spectral peak search on the electromagnetic environment sensing data through the Multiple Signal Classification (MUSIC) algorithm. Specifically, these parameters include azimuth and elevation angles, which are used to determine the spatial pointing of the interference source relative to the phased array antenna.

[0051] The aforementioned polarization domain characteristic data can be obtained by comparing the amplitude ratio and phase difference of the signals received by the dual polarization channels and calculating the polarization ellipse tilt angle and axial ratio through trigonometric function operations. These data are used to describe the orthogonal or coupled relationship between the polarization state of the interference signal and the polarization state of the communication signal.

[0052] The aforementioned frequency domain characteristic data can be the spectral distribution information obtained after performing a fast Fourier transform on the electromagnetic environment sensing data. Specifically, it includes the center frequency and bandwidth of the interference signal, which is used to identify the frequency range and spectral width of the interference signal.

[0053] The aforementioned energy domain characteristic data can be the power intensity and time duty cycle of the interference signal obtained through signal processing. The power intensity reflects the absolute energy level of the interference signal, and the time duty cycle reflects the pulse characteristics of the interference signal on the time axis, which is used to distinguish between pulse interference and continuous interference.

[0054] For example, when acquiring electromagnetic environment sensing data containing multi-dimensional feature data, a sensing subarray is formed by multiple array elements located at the edge of the spaceborne phased array antenna. Within the sensing time slot of the communication frame, a full-space scanning signal is received through a low-noise amplifier, mixer, and analog-to-digital converter. Spatial feature data is obtained through covariance matrix eigenvalue decomposition, polarization domain feature data is obtained through dual-polarization channel amplitude ratio and phase difference calculation, frequency domain feature data is obtained through fast Fourier transform, and energy domain feature data is obtained through zero-crossing detection and amplitude calibration. Finally, multi-dimensional feature data including azimuth angle, elevation angle, polarization ellipse tilt angle, axial ratio, center frequency, bandwidth, power intensity, and time duty cycle are constructed, providing high-dimensional electromagnetic situation input for subsequent interference analysis.

[0055] The aforementioned target interference analysis model can be a spaceborne inference engine built on a lightweight one-dimensional convolutional neural network. The model contains convolutional layers, pooling layers, and fully connected layers. Its weight parameters have been trained and quantized on the ground and are used to directly output interference classification results and control parameters from the input multi-dimensional feature data through nonlinear mapping.

[0056] The aforementioned interference suppression decision results can be the set of execution instructions output by the target interference analysis model after processing multi-dimensional feature data. These results include classification labels and numerical control parameters used to guide the specific anti-interference actions of the antenna array, and serve as a decision link between the perception layer and the execution layer.

[0057] The above interference type labels It can output classification enumeration values ​​for the target interference analysis model to characterize the time-frequency characteristics of the interference signal, specifically including: narrowband continuous interference, wideband continuous interference, narrowband pulse interference, wideband pulse interference, and spoofing signal interference. This label reflects the essential attributes of the interference signal in terms of frequency bandwidth, time continuity, or signal intent.

[0058] The aforementioned suppression strategy parameter vector G can be a configuration vector containing numerical control commands output by the target interference analysis model, specifically including: recommended null depth, polarization switching flag, time-domain scheduling duty cycle detection threshold, and priority level. The recommended null depth is used to limit the lower limit of gain suppression that the beamforming algorithm must achieve in the interference direction. The polarization switching flag indicates whether to switch the communication beam polarization state to a state orthogonal to the interference polarization. The time-domain scheduling duty cycle detection threshold is the critical ratio used by the pulse detection synchronization circuit to determine whether to initiate the RF switch's time-domain avoidance action. The priority level indicates the urgency of responding to this interference event.

[0059] For example, after acquiring electromagnetic environment sensing data, a target interference analysis model is used to predict and analyze the electromagnetic environment sensing data to obtain interference suppression decision results. The target interference analysis model maps multidimensional feature data into interference type labels and suppression strategy parameter vectors. The normalized multidimensional feature data is input into a lightweight one-dimensional convolutional neural network embedded in the FPGA. After local correlation patterns are extracted by convolutional layers and dimensionality reduction by pooling layers, features are fused through fully connected layers. Finally, the output fully connected layer outputs interference type labels through a Softmax activation function, and in parallel, it regresses to output a suppression strategy parameter vector consisting of recommended null depth, polarization switching flag, temporal scheduling duty cycle detection threshold, and priority level, thus achieving a direct mapping from electromagnetic features to control commands.

[0060] The aforementioned interference suppression operation can be a technical action that, based on the interference suppression decision results, weakens or isolates interference signals by physical means by adjusting the amplitude and phase weights, polarization state, or radio frequency path on / off state of the antenna array.

[0061] After generating the interference suppression decision result, the interference suppression operation is executed according to the decision result. For example, the suppression strategy parameter vector from the interference suppression decision result is distributed to different execution modules: the recommended null depth is input to the fast beamformer, and a complex weight vector is calculated using a weighted least squares algorithm to drive the phase shifter and attenuator to generate deep nulls in the interference direction. The polarization switching flag is input to the polarization adaptive controller to adjust the excitation amplitude ratio and phase difference of the feed network, making the communication beam polarization state orthogonal to the interference polarization. The time-domain scheduling duty cycle detection threshold is input to the time-domain resource coordination scheduler to control the RF switch to disconnect the damaged channel or switch to the backup beam during pulse interference, thereby collaboratively completing physical isolation and signal suppression of the interference signal in the spatial, polarization, and time dimensions.

[0062] Based on the above steps S21 to S23, this embodiment of the application acquires electromagnetic environment sensing data containing multi-dimensional feature data in the spatial domain, polarization domain, frequency domain, and energy domain, and uses a target interference analysis model to map the multi-dimensional feature data into interference type labels and suppression strategy parameter vectors. Then, based on the decision result, interference suppression operations are performed, achieving the purpose of accurate mapping between multi-dimensional features of the electromagnetic environment and interference suppression strategies. This improves the adaptability, timeliness, and accuracy of signal interference suppression, thereby solving the technical problems of poor adaptability, poor timeliness, and low accuracy of signal interference suppression methods in related technologies.

[0063] Optionally, acquiring electromagnetic environment sensing data includes: performing beam scanning of the entire airspace using a sensing subarray within a sensing time slot to obtain raw scanning data; performing signal conversion operations on the raw scanning data to obtain signal conversion results, wherein the signal conversion operations include at least one of the following: signal amplification operation, signal mixing operation, and signal sampling operation; and performing multi-dimensional feature extraction based on the signal conversion results to obtain electromagnetic environment sensing data.

[0064] The aforementioned sensing time slot can be a specific time interval within a communication frame used for electromagnetic environment sensing rather than communication service transmission. During this interval, the system suspends the transmission and reception of the communication beam and switches the radio frequency channel to sensing mode. This time slot allocation is controlled by frame synchronization pulses generated by the clock synchronization unit.

[0065] For example, the system divides each communication frame into a communication time slot and a sensing time slot. For instance, in an embodiment of this application, the duration of one frame is... =10ms, of which the communication time slot occupies 9.8ms and the sensing time slot occupies 200μs.

[0066] The aforementioned sensing subarray can be an independent receiving unit composed of multiple array elements selected from the edge of the spaceborne phased array antenna array. In this application, eight edge array elements can be selected to form a subarray for performing a full-space scanning task.

[0067] The aforementioned full airspace can cover the complete spatial angular range of azimuth from 0° to 360° and elevation from 0° to 90°.

[0068] The aforementioned beam scanning can be a process of rapidly switching the beam direction of the sensing subarray using a pre-stored scanning codebook, and lingering on each direction for a specific time to acquire the signal in that direction.

[0069] The raw scanning data mentioned above can be unprocessed analog or digital signal streams acquired by the sensing subarray during its dwell time in each beam direction.

[0070] The above signal conversion operations can be a series of operations to preprocess the raw scan data to extract usable information, including signal amplification, signal mixing, and signal sampling.

[0071] The above signal conversion result can be a digital intermediate frequency signal formed after the original scanning data is processed by a low-noise amplifier, mixer and analog-to-digital converter. This signal retains the amplitude, phase and frequency information of the original electromagnetic wave, and has a quantized data format suitable for subsequent algorithm calculations by the digital signal processing unit.

[0072] The above signal amplification operation can be performed by using a low-noise amplifier to amplify the received weak electromagnetic signal to improve the signal-to-noise ratio, while introducing the lowest possible self-noise to ensure that the signal characteristics are not overwhelmed by noise during subsequent processing.

[0073] The aforementioned signal mixing operation involves nonlinearly mixing the received radio frequency (RF) signal with the local oscillator (LO) signal in a mixer to generate a new frequency signal containing the sum and difference of the original signal frequency and the LO frequency. Subsequent filtering selects the difference frequency signal, down-converting the high-frequency RF signal to a lower intermediate frequency (IF) so that the analog-to-digital converter (ADC) can perform digitization at a lower sampling rate. For example, the center frequency of the LO signal is fc = 12.5 GHz.

[0074] The above signal sampling operation can be performed by using an analog-to-digital converter to discretize the instantaneous voltage value of the analog intermediate frequency signal according to a fixed sampling clock period, and converting the measured analog voltage value into a multi-bit digital code to generate a digital signal sequence that can be directly processed by digital logic circuits such as FPGA.

[0075] The above-mentioned multidimensional feature extraction can be a process of calculating and extracting numerical parameters reflecting the characteristics of interference signals in different physical dimensions from the signal conversion results.

[0076] When acquiring electromagnetic environment sensing data, a sensing subarray is used to perform beam scanning across the entire spatial domain within the sensing time slot to obtain raw scan data. For example, the system clock synchronization unit triggers the sensing subarray to operate within the sensing time slot of the communication frame. Based on a pre-stored 8×N scanning codebook, the sensing subarray sequentially switches to multiple discrete beam directions (e.g., 64 directions) across the entire spatial domain with a preset dwell time, such as 2 μs, completing a scan of all directions within 200 μs. Spatial electromagnetic signals are received at each beam direction, and the received continuous signals are temporarily stored as raw scan data, completing a rapid full-space coverage scan.

[0077] After acquiring the raw scan data, a signal conversion operation is performed on the raw scan data to obtain the signal conversion result. For example, the raw scan data is sequentially amplified by a low-noise amplifier (LNA) to increase signal strength, then enters a mixer for signal mixing, down-converting the high-frequency signal to a fixed intermediate frequency, such as 70MHz. After passing through a bandpass filter, a high-speed analog-to-digital converter (ADC) performs signal sampling, quantizing the analog IF signal into a digital IF signal, ultimately forming a signal conversion result that can be read by the digital signal processing module. For example, the bandwidth of the bandpass filter can be 20MHz, and a 14-bit ADC can be used to sample at a rate of 160 MSPS.

[0078] Multidimensional feature extraction is performed on the signal conversion results to obtain electromagnetic environment sensing data. For example, the digital signal processing module executes multiple algorithms in parallel to analyze the signal conversion results, including using the MUSIC algorithm to calculate the covariance matrix and search for spectral peaks to extract spatial domain feature data, such as the direction of arrival (DOA). Polarization domain feature data, such as the polarization ellipse tilt angle and axial ratio, is calculated using the amplitude ratio and phase difference of the dual-polarization channels. Frequency domain feature data, such as the center frequency and bandwidth, is extracted using Fast Fourier Transform. Energy domain feature data, such as power intensity and time duty cycle, is extracted through zero-crossing detection and amplitude calibration. Finally, the above four-dimensional feature data are integrated to form electromagnetic environment sensing data.

[0079] For example, when extracting spatial feature data, the MUSIC algorithm based on the covariance matrix is ​​used. This algorithm calculates the covariance matrix of the signals between channels and performs eigenvalue decomposition to separate the signal subspace from the noise subspace. For instance, an 8×8 dimensional covariance matrix can be used. ,in, This represents a vector composed of signals from each channel of the receiving array.

[0080] Subsequently, a peak search formula is constructed using the orthogonality between the steering vector and the noise subspace to search for the peak position of the power spectral function, thereby accurately obtaining the azimuth and elevation angles of the interference signal. The peak search formula is:

[0081]

[0082] in, This represents the spatial spectral function value of the MUSIC algorithm. Indicates azimuth. Indicates pitch angle, This represents the array guide vector corresponding to the spatial angle. This represents a matrix composed of eigenvectors from the noise subspace. Furthermore, the DOA estimation accuracy is better than 1°.

[0083] For example, when performing polarization domain feature parameter extraction, left-handed circular polarization (LHCP) and right-handed circular polarization (RHCP) signals are received separately using dual polarization channels. The amplitude and phase data of each channel are obtained, the ratio of the amplitudes of the left and right circular polarization channels and the phase difference are calculated, and then the polarization ellipse tilt angle of the interference signal is derived through trigonometric function operations. Compared to the shaft ratio The calculation results are then quantified into eight typical polarization states: left-handed circular polarization, right-handed circular polarization, horizontal linear polarization, vertical linear polarization, +45° linear polarization, -45° linear polarization, left-handed elliptical polarization, and right-handed elliptical polarization. The polarization ellipse tilt angle is also considered. Compared to the shaft ratio The calculation formulas are as follows:

[0084]

[0085]

[0086] in, This indicates the signal amplitude of the left-hand circularly polarized (LHCP) channel. This indicates the phase of the signal received by the left-hand circularly polarized channel. This indicates the signal amplitude of the right-hand circularly polarized (RHCP) channel. This indicates the phase of the signal received by the right-hand circularly polarized channel. This represents the phase difference between the left-hand and right-hand channels, i.e. .

[0087] For example, when extracting frequency and energy domain feature parameters, a 2048-point Fast Fourier Transform (FFT) is performed on the digital intermediate frequency signal of each channel to obtain the spectral amplitude distribution, and the frequency point corresponding to the spectral peak is searched as the interference center frequency. The frequency width corresponding to a 3 dB decrease in peak amplitude is calculated as the interference bandwidth. The peak amplitude is converted into interference power intensity after calibration by an analog-to-digital converter. The unit is dBm. Simultaneously, zero-crossing detection is performed on the signal's time-domain envelope to statistically determine the ratio of the signal's effective duration to its period, thereby obtaining the time duty cycle reflecting the pulse characteristics. ,in, The value is between 0 and 1, indicating narrowband continuous interference. ≈1, pulse interference <0.3.

[0088] Finally, the system integrates the azimuth and elevation angles in the spatial domain, the polarization ellipse tilt and axial ratio in the polarization domain, the center frequency and bandwidth in the frequency domain, and the power intensity and time duty cycle in the energy domain to construct a four-dimensional feature vector containing eight dimensions of information. This serves as standardized input data for subsequent lightweight neural network interference feature extraction and classification.

[0089] Based on the above optional embodiments, this application embodiment performs beam scanning of the entire airspace using a sensing subarray within the sensing time slot to obtain raw scanning data. Signal conversion is then performed on the raw scanning data to obtain signal conversion results. Multidimensional feature extraction is performed based on the signal conversion results to obtain electromagnetic environment sensing data. This achieves the goal of comprehensively acquiring and standardizing multidimensional features of the electromagnetic environment, thereby improving the accuracy of interference signal feature extraction and reducing the system's hardware resource consumption. It also solves the technical problem that related technologies struggle to simultaneously acquire high-precision multidimensional features of airspace, polarization, frequency, and energy under limited spaceborne resources.

[0090] Optionally, the target interference analysis model includes a feature preprocessing module and a prediction inference module. The target interference analysis model is used to predict and analyze electromagnetic environment sensing data to obtain interference suppression decision results. This includes: using the feature preprocessing module to normalize the electromagnetic environment sensing data to obtain normalization results; and using the prediction inference module to predict and analyze the normalization results to obtain interference suppression decision results. The prediction inference module is composed of a one-dimensional convolutional neural network.

[0091] The aforementioned feature preprocessing module is a functional unit that standardizes the input data. Its core function is to normalize the electromagnetic environment sensing data, mapping the original features with huge differences in different dimensions or magnitudes to a unified numerical range, so as to eliminate the negative impact of data distribution differences on subsequent model training and improve the model convergence speed and stability.

[0092] The aforementioned prediction and inference module can be a functional unit for feature extraction and classification prediction based on deep learning algorithms. In this embodiment, the module is composed of a one-dimensional convolutional neural network, which is used to receive the normalization processing result, extract local correlation patterns by sliding the convolution kernel on the feature dimension, and process them through pooling and fully connected layers, and finally output the interference suppression decision result, including the interference type label and the corresponding suppression strategy parameter vector.

[0093] The above normalization result can be a multidimensional feature vector that has been processed by the feature preprocessing module, eliminating the differences in dimensions and being at a uniform scale. This result is provided as a standardized input to the next processing stage.

[0094] The aforementioned interference suppression decision results can be a comprehensive control command output by the prediction inference module, which includes interference type labels and suppression strategy parameter vectors. The suppression strategy parameter vectors include recommended null depth, polarization switching flag, time-domain scheduling duty cycle threshold, and priority level, which are used to guide the beamformer, polarization controller, and time-domain resource coordinating scheduler to perform specific anti-interference actions.

[0095] When using a target interference analysis model to predict and analyze electromagnetic environment sensing data and obtain interference suppression decision results, a feature preprocessing module is used to normalize the electromagnetic environment sensing data F, obtaining a normalized result. This eliminates dimensional differences and numerical range deviations between different feature dimensions in the original data, providing standardized input data for the neural network. The normalization formula is:

[0096]

[0097] in, and These are the mean and standard deviation of the k-th feature on the training set, respectively, and are pre-stored in the FPGA's read-only memory (ROM). The normalized feature vector... The dimension is 8×1.

[0098] Subsequently, a prediction and inference module composed of a one-dimensional convolutional neural network is used to extract features and perform predictive analysis on the normalization results. The local correlation between features is captured through one-dimensional convolution operations, and after layer-by-layer mapping, the interference suppression decision results, including the interference type and specific suppression parameters, are finally generated.

[0099] Figure 3 This is a schematic diagram of an optional one-dimensional convolutional neural network according to an embodiment of this application, such as... Figure 3 As shown, this neural network specifically includes an input layer, convolutional layers, pooling layers, flattening layers, and fully connected layers. The input layer receives an 8×1 feature vector, which is obtained by normalizing four-dimensional electromagnetic features extracted by the electromagnetic environment perception module, including spatial angle of arrival, polarization ellipse tilt angle, axial ratio, center frequency, bandwidth, power intensity, and time duty cycle.

[0100] The first convolutional layer (Conv1) has a kernel size of 3, a stride of 1, and 16 output channels. It uses the ReLU activation function to map the 8-dimensional features to a 16-dimensional feature space to extract local correlation patterns. Then, a max pooling layer (Pool1) with a window size of 2 and a stride of 2 is used to reduce dimensionality while preserving salient features, outputting a 16×4 feature map. The second convolutional layer (Conv2) further has a kernel size of 3, a stride of 1, and 32 output channels. It also uses the ReLU activation function, and then a max pooling layer (Pool2) with the same parameters is used to compress the feature map to 32×1. At this point, the feature dimension is compressed to one dimension.

[0101] The flattening layer flattens the 32×1 feature map into a 32×1 one-dimensional vector, which is then fed into the first fully connected layer (FC1). This layer has an input dimension of 32 and an output dimension of 16, employs the ReLU activation function, and introduces a dropout mechanism with a packet loss rate of 0.2 to prevent overfitting. The output fully connected layer (FC2) serves as the final classification and regression unit, with an input dimension of 16, and outputs two results in parallel: the first is a C-dimensional probability distribution of interference types, which is selected by using the Softmax activation function to select the interference type label corresponding to the highest probability. Where C=5, corresponding to 5 interference type labels: narrowband continuous interference, wideband continuous interference, narrowband impulse interference, wideband impulse interference, and spoofing interference. The activation function outputs the probability value for each category. , j=1~5.

[0102] The second is the four-dimensional policy parameter vector. The recommended zero-depression depth, polarization switching flag, temporal scheduling duty cycle detection threshold, and priority level are obtained through regression analysis. This indicates the recommended null depth, in dB, determined by a neural network based on the interference power intensity. The results were obtained through a comprehensive regression analysis of the types. , indicates the polarization switching flag, 0 means no switching, 1 means switching to orthogonal polarization. , represents the time-domain scheduling duty cycle detection threshold, which is only effective for impulse interference. The number indicates the priority level, with 1 being the highest and 5 being the lowest.

[0103] The total number of parameters in the network is: Conv1(3×1×16+16)=64, Conv2(3×16×32+32)=1568, FC1(32×16+16)=528, FC2(16×5+5+16×4+4)=85+68=153. This lightweight neural network has less than 2.4k parameters and less than 10 computations. 5The number of multiply-accumulate operations (MACs) is calculated. All weight parameters are quantized to INT8 (integer data type) precision and stored in the FPGA's block random access memory (RAM) after ground training is completed.

[0104] The inference process adopts a pipelined parallel architecture. The Conv1 layer requires 3 clock cycles to process 8 data points, the Conv2 layer requires 2 clock cycles, and the fully connected layer requires 4 clock cycles. The entire inference process takes less than 0.8ms at a 100MHz clock speed, thus enabling rapid identification and suppression strategy generation for interference signals in complex electromagnetic environments with extremely low power consumption and computing power consumption.

[0105] Based on the above optional embodiments, this application embodiment normalizes the electromagnetic environment sensing data using a feature preprocessing module to obtain a normalized processing result, and then uses a prediction and reasoning module to predict and analyze the normalized processing result to obtain an interference suppression decision result, thereby realizing the automated and real-time conversion from multi-dimensional sensing data to specific anti-interference control commands.

[0106] Optionally, the spatial domain feature data includes: the azimuth and elevation angles of the interfering signal; the polarization domain feature data includes: the polarization ellipse tilt angle and axial ratio of the interfering signal; the frequency domain feature data includes: the center frequency and bandwidth of the interfering signal; and the energy domain feature data includes: the power intensity and time duty cycle of the interfering signal. The interfering type label includes at least one of the following: narrowband continuous interference, wideband continuous interference, narrowband pulse interference, wideband pulse interference, and spoofing signal interference. The suppression strategy parameter vector includes: recommended null depth, polarization switching flag, time-domain scheduling duty cycle detection threshold, and priority level.

[0107] The aforementioned azimuth angle can be the angle between the projection of the beam propagation direction of the interfering signal onto the horizontal plane and the reference direction when the signal arrives at the phased array antenna array. It is used to determine the spatial position of the interference source in the horizontal direction and is a key geometric parameter for calculating the spatial steering vector and generating spatial nulls. The reference direction is usually true north or the array normal direction.

[0108] The aforementioned elevation angle can be the angle between the beam propagation direction of the interference signal and the horizontal plane when the interference signal arrives at the phased array antenna array. It is used to determine the spatial position of the interference source in the vertical direction. Combined with the azimuth angle, it constitutes the complete direction of arrival (DOA) of the interference signal in three-dimensional space, providing precise angular constraints for the beamforming algorithm.

[0109] The aforementioned polarization ellipse tilt angle can be the angle between the major axis of the ellipse trajectory formed by the endpoint of the interference signal electric field vector changing in space over time and the reference axis, such as the horizontal axis. It is used to describe the spatial orientation of the polarization state of the interference signal and is an important basis for calculating the optimal receiving polarization state of the communication beam to achieve polarization orthogonal suppression.

[0110] The aforementioned axial ratio can be the ratio of the length of the major axis to the length of the minor axis of the polarization ellipse of the interference signal. It is used to quantify the degree of ellipticity of the polarization state of the interference signal. An axial ratio of 1 indicates circular polarization, and an infinite axial ratio indicates linear polarization. This parameter, together with the tilt angle of the polarization ellipse, fully characterizes the polarization characteristics of the interference signal and guides the polarization controller to adjust the feed network to generate an orthogonal polarization state.

[0111] The aforementioned center frequency can be the frequency point where the spectrum energy distribution of the interference signal is most concentrated. It is used to identify the frequency band where the interference signal is located, and is the basis for judging whether the interference signal is co-channel interference or adjacent-channel interference. It is also a key reference for setting the frequency domain filter and the down-conversion local oscillator frequency.

[0112] The bandwidth mentioned above can be the effective width of the spectrum energy distribution of the interference signal. It is usually defined as the frequency range where the power spectral density drops by 3dB. It is used to distinguish between narrowband interference and broadband interference. The bandwidth information helps to assess the coverage of the interference on the communication frequency band and the number of subcarriers affected.

[0113] The aforementioned power intensity can be the average power level of the interference signal at the receiver, usually expressed in dBm. It is used to quantify the strength of the interference signal and is an important basis for determining the zero-depression depth requirement and triggering the time-domain protection mode in the suppression strategy. High power intensity may lead to saturation of the RF front end, requiring stronger suppression measures.

[0114] The duty cycle mentioned above can be the ratio of the time the interference signal is in the transmission state within a unit period to the entire period. It is used to characterize the time distribution characteristics of the interference signal. A high duty cycle usually corresponds to continuous interference, while a low duty cycle corresponds to pulse interference. This parameter directly affects whether the time-domain resource coordination scheduler enables the radio frequency switch for pulse avoidance.

[0115] The aforementioned narrowband continuous interference can be an interference signal with a narrow spectral width that persists in time. Its energy is concentrated near a specific frequency point, mainly causing a deterioration in the signal-to-interference-plus-noise ratio at a specific frequency point in the communication system. It is usually suppressed by spatial nulling and frequency filtering.

[0116] The aforementioned broadband continuous interference can be interference signals with a wide spectral width and continuous existence in time. Its energy is distributed over a wide frequency range, possibly covering the entire communication bandwidth, causing a decrease in the signal-to-interference-plus-noise ratio across the entire frequency band or over a large area of ​​the communication system. It usually requires comprehensive suppression by combining spatial, polar, and power control methods.

[0117] The aforementioned narrowband pulse interference can be an interference signal with a narrow spectral width but intermittently transmitted in the form of pulses in time. Its peak power is high but its average power may be low. Its main impact on the receiver is to cause instantaneous saturation or blockage at the moment the pulse arrives. It is usually avoided by cutting off the signal path during the pulse period through a time-domain switch.

[0118] The aforementioned broadband pulse interference can be an interference signal with a wide spectral width that is intermittently transmitted in the form of pulses in time. It has both the frequency coverage characteristics of broadband interference and the time burst characteristics of pulse interference, causing severe instantaneous impact and frequency band coverage to the receiver. It needs to be jointly suppressed by combining time-domain avoidance and spatial nulling.

[0119] The aforementioned spoofed signal interference can be interference with legitimate signals that deceive the receiver's positioning or demodulation functions. It may mimic the modulation characteristics and protocol format of communication signals, rendering traditional detection methods based on protocol analysis ineffective. It is necessary to rely on physical layer characteristics for identification and suppression, such as polarization, time delay, and Doppler.

[0120] The recommended null depth can be the degree of gain suppression that the beamformer needs to achieve in the interference direction, as predicted and suggested by the neural network. It is usually expressed in dB. This parameter directly constrains the upper limit of the interference direction gain in the weighted least squares algorithm, ensuring that the interference signal is sufficiently attenuated to a level that does not affect the communication quality.

[0121] The aforementioned polarization switching flag can be a control bit output by a neural network, used to indicate whether the polarization adaptive controller performs a polarization state switching operation. When the flag is 1, the controller adjusts the polarization of the communication beam to a state orthogonal to the polarization of the interference signal, thereby further suppressing interference through polarization isolation. When the flag is 0, the original polarization state is maintained.

[0122] The aforementioned time-domain scheduling duty cycle detection threshold can be a critical duty cycle value output by a neural network to determine whether to activate time-domain RF switch control. When the time duty cycle of the interference signal is lower than this threshold, it is determined to be sparse pulse interference, and the system activates the RF switch for pulse avoidance. When it is higher than this threshold, it is determined to be dense interference, and only spatial and polar domain suppression is relied upon to avoid communication interruptions and resource waste caused by frequent switching.

[0123] The aforementioned priority levels can be attributed to the neural network and used for scheduling arbitration when multiple interferences coexist or resource conflicts occur. A smaller value indicates a higher priority, guiding the time-domain resource coordinating scheduler to prioritize suppression actions for higher-priority interferences when multiple interference events occur simultaneously, thus ensuring the reliability of critical communication links.

[0124] This application's embodiments achieve a comprehensive quantitative description of complex electromagnetic environment interference from four dimensions: space, polarization, frequency, and energy, by defining spatial domain feature data, polarization domain feature data, frequency domain feature data, and energy domain feature data. The azimuth and elevation angles of the interference signal in the spatial domain feature data provide precise interference direction constraints for the beamformer, enabling the weighted least squares algorithm to accurately generate nulls pointing towards the interference source. The polarization ellipse tilt angle and axial ratio of the interference signal in the polarization domain feature data allow the polarization adaptive controller to calculate the orthogonal polarization state, suppressing co-polarization interference by switching the communication beam polarization, thereby increasing the degrees of freedom in the polarization domain based on spatial domain filtering and improving the ability to suppress polarization-agile interference. The center frequency and bandwidth of the interference signal in the frequency domain feature data, combined with the power intensity and time duty cycle of the interference signal in the energy domain feature data, enable the lightweight neural network input layer to receive complete multi-dimensional features and accurately distinguish interference types.

[0125] The interference type labels—narrowband continuous interference, wideband continuous interference, narrowband pulse interference, wideband pulse interference, and spoofed signal interference—cover typical and complex interference scenarios, enabling the system to adopt differentiated suppression strategies for interference of different natures. The recommended null depth in the suppression strategy parameter vector directly quantifies the attenuation level the beamformer needs to achieve in the interference direction. The polarization switching flag controls whether the polarization controller performs orthogonal polarization switching. The time-domain scheduling duty cycle detection threshold determines whether the time-domain resource co-scheduler activates the RF switch for pulse avoidance. The priority level is used for arbitration of resource scheduling when multiple interferences coexist.

[0126] Based on the above optional embodiments, this application embodiment achieves closed-loop processing from multi-dimensional electromagnetic feature perception to specific anti-interference control command conversion by comprehensively acquiring spatial domain feature data, polarization domain feature data, frequency domain feature data, and energy domain feature data, combined with interference type label classification and identification, and the generation of suppression strategy parameter vectors. This scheme uses the azimuth and elevation angles of the interference signal to guide spatial null generation, uses the polarization ellipse tilt angle and axial ratio of the interference signal to guide polarization state switching, and uses the center frequency and bandwidth of the interference signal, as well as the power intensity and time duty cycle of the interference signal, to assist in interference type discrimination. Finally, it outputs recommended null depth, polarization switching flag, time-domain scheduling duty cycle detection threshold, and priority level. The mechanism of multi-dimensional feature fusion and multi-dimensional suppression strategy output enables the system to adaptively adjust the beam pattern, polarization state and time-domain switching action in milliseconds against different types of interference, such as narrowband continuous interference, wideband continuous interference, narrowband pulse interference, wideband pulse interference and spoofing signal interference. This significantly improves the real-time performance, depth and robustness of anti-interference under the condition of limited spaceborne resources, and effectively solves the problems of lack of adaptive capability in traditional static EMC design and slow convergence and severe main lobe distortion of existing adaptive nulling antennas.

[0127] Optionally, the interference suppression operation performed based on the interference suppression decision results includes: performing spatial beamforming operation and polarization adaptive control operation based on the interference type label and the suppression strategy parameter vector. The spatial beamforming operation is used to calculate a complex weight vector based on the recommended null depth to control the amplitude and phase distribution of the phased array elements. The polarization adaptive control operation is used to adjust the amplitude ratio and phase difference of the biorthogonal polarization channels in the feed network according to the polarization state of the interference signal.

[0128] The aforementioned spatial beamforming operation can be a technical process that adjusts the excitation amplitude and phase of each element of a phased array antenna to form a gain main lobe in the direction of the desired signal and a gain null in the direction of interference. This operation aims to use the difference in spatial degrees of freedom to achieve the isolation and suppression of interference signals.

[0129] The aforementioned polarization adaptive control operation can be a technical process that adjusts the amplitude ratio and phase difference of the excitation signals of different polarization channels in the antenna feed network, thereby changing the polarization state of the antenna's radiated or received signal to be orthogonal to the polarization state of the interfering signal, and thus further suppressing interference by utilizing polarization degrees of freedom. For example, the polarization channel can be a left-hand circular polarization channel or a right-hand circular polarization channel.

[0130] The recommended null depth can be a parameter value provided by the suppression strategy parameter vector, which is expected to achieve the signal gain attenuation level in the interference direction. This parameter is used as a constraint input to the beamforming algorithm to ensure that the null is deep enough to suppress strong interference.

[0131] The aforementioned complex weight vector can be a vector containing the complex excitation coefficients of each array element. This vector is the core control data of the beamformer and determines the spatial radiation characteristics of the phased array antenna array. The complex excitation coefficients can be composed of amplitude control words and phase control words.

[0132] The aforementioned phased array elements can serve as the basic radiating units for a phased array antenna array. By independently controlling the amplitude and phase distribution of each element, flexible beam pointing and dynamic pattern reconstruction can be achieved.

[0133] The amplitude and phase distribution described above can be a spatial arrangement and combination of the amplitude and phase angle of the excitation signal of each element in the phased array antenna array. Changing the amplitude and phase distribution changes the far-field radiation pattern of the antenna, thereby achieving null generation in the direction of interference.

[0134] The aforementioned polarization state can be described as the trajectory shape and spatial orientation of the electric field vector in the interference signal as it changes over time in a plane perpendicular to the propagation direction. During electromagnetic wave propagation, the trajectory of the endpoints of the electric field vector typically forms an ellipse. The geometric characteristics and spatial angles of this ellipse together constitute the polarization state of the interference signal. Specifically, the polarization state of the interference signal is determined by two key parameters: first, the tilt angle of the polarization ellipse, i.e., the angle between the major axis of the ellipse and the reference axis, used to describe the spatial rotation angle of the polarization ellipse; second, the axial ratio of the interference signal, i.e., the ratio of the length of the major axis to the length of the minor axis, used to quantify the flatness of the polarization ellipse. An axial ratio of 1 indicates circular polarization, while an infinitely large axial ratio indicates linear polarization.

[0135] The aforementioned feed network can be a passive or active circuit network that distributes radio frequency signals to each element of the phased array antenna and controls its polarization state. In the embodiments of this application, the feed network includes an independent excitation control circuit with dual orthogonal polarization channels to achieve agile polarization state.

[0136] The aforementioned dual orthogonal polarization channels can be two mutually orthogonal polarization channels in the feed network, such as the left-hand circular polarization channel LHCP and the right-hand circular polarization channel RHCP, or the horizontal polarization channel H and the vertical polarization channel V. By independently controlling the signals of the two channels, electromagnetic waves in any polarization state can be synthesized.

[0137] The amplitude ratio mentioned above can be the ratio between the amplitudes of the excitation signals of the two channels in the biorthogonal polarization channel. This ratio determines the ellipticity or linear polarization angle of the synthesized polarized wave.

[0138] The aforementioned phase difference can be the phase offset between the excitation signals of the two channels in the biorthogonal polarization channel. This parameter, together with the amplitude ratio, determines the rotation direction and elliptical tilt angle of the synthesized polarized wave.

[0139] When performing interference suppression operations based on the interference suppression decision results, spatial beamforming and polarization adaptive control operations are performed according to the interference type label and the suppression strategy parameter vector. For example, the system receives the interference suppression decision results output by a lightweight neural network processing unit, parses the results, extracts the interference type label to confirm the nature of the interference, extracts the recommended null depth from the suppression strategy parameter vector for spatial beamforming, and extracts the polarization switching flag and polarization state information for polarization adaptive control, thereby simultaneously or sequentially activating anti-interference mechanisms in both the spatial and polar domains.

[0140] For example, a fast null generation algorithm is executed during spatial beamforming operations. Assuming the spaceborne phased array has N elements, this application can select an 8×8 rectangular grid array with N=64 elements, and the desired communication direction is... The direction of interference is The beamformer needs to solve for an N×1 dimensional complex weight vector W that maximizes the gain in the desired direction, minimizes the gain in the interference direction, and minimizes the dynamic range of each element's amplitude. The maximization of gain in the desired direction satisfies... The minimum gain in the interference direction satisfies g1 is the recommended zero-depression depth.

[0141] This application employs a closed-form solution method based on weighted least squares, the formula of which is:

[0142]

[0143] in, The noise covariance matrix is ​​obtained by estimating the signal-free sampling during the sensing period. A weighting factor for the pairwise elements is introduced here. The weights of the edge elements are greater than those of the center elements to suppress the array edge effect. . For the constraint matrix, Dimension N×2, To constrain the response vector, .

[0144] Using the above formula, the recommended null depth is used as the upper limit of the interference direction gain constraint. The complex weight vector that satisfies this constraint and maximizes the desired direction gain is then solved. Under the constraints of ensuring that the desired direction gain is 1 (normalized) and the interference direction gain is below a set threshold, the total noise power of the array output is minimized, thereby achieving the optimal signal-to-interference-plus-noise ratio (SINR).

[0145] The above solution process involves inverting a 2×2 matrix, with a computational complexity of O(N), thus avoiding the O(N) complexity of inverting an N×N matrix in the traditional Linearly Constrained Minimum Variance (LCMV) beamforming algorithm. The computational complexity is low, making it suitable for real-time implementation on spacecraft.

[0146] The beamformer completes all weight calculations within 0.2ms of receiving the recommended null depth and writes the amplitude control word and phase control word into the phase shifter and attenuator registers of each channel via the SPI bus. The amplitude control word uses 5 bits to represent the attenuation level, corresponding to attenuation levels 0-31. The phase control word uses 6 bits to represent the phase offset, in 5.625° steps, corresponding to 0-360°.

[0147] To prevent null formation from causing main beam distortion, this embodiment adds a first-stage conformal correction step after solving for the weight vector W. Specifically, it calculates the average gain of the current beam pattern within a 3dB width of the main lobe. ,like If the decrease in gain relative to the desired direction exceeds 0.5 dB, then the constraint response vector is fine-tuned. The first term in the algorithm is increased from 1 to 1.03, and W is recalculated until the main lobe gain loss is ≤0.5dB. Experimental results show that this iteration usually converges within 2-3 iterations, with an additional time consumption of <0.1ms.

[0148] For example, when the polarization switching flag g2=1, polarization adaptive control is performed. The polarization controller adjusts the polarization based on the measured disturbance polarization state. Calculations are performed to determine the optimal receiving polarization state of the communication beam, ensuring it is orthogonal to the interfering polarization, thereby further suppressing interference through polarization isolation. Specifically, for a fully polarized wave, the optimal polarization state satisfies the polarization ellipse tilt angle. And axial ratio Subsequently, the polarization controller adjusts the excitation amplitude ratio of the left-hand and right-hand circular polarization channels in the feed network. and phase difference The calculation formulas for achieving the above-mentioned target polarization state are as follows:

[0149]

[0150]

[0151] in, The calculated optimal polarization axis ratio, The excitation amplitude of the left-hand circularly polarized channel. The excitation amplitude of the right-hand circularly polarized channel is... The calculated optimal polarization ellipse tilt angle is given.

[0152] The polarization controller converts the calculated amplitude ratio and phase difference into control words and completes the state switching of the feeder network within 0.1ms, so that the communication beam is orthogonal to the interference signal in the polarization dimension, effectively reducing interference power and improving the overall anti-interference capability and communication link quality of the system.

[0153] After performing the spatial beamforming operation and polarization adaptive control operation, the output of this embodiment includes a 64-dimensional complex weight vector W and a control word set of the polarization feed network. Each element of the complex weight vector includes amplitude and phase, which together drive the phased array antenna to maintain the main lobe in the communication direction, form a deep null in the interference direction, and have the polarization state orthogonal to the interference.

[0154] Based on the above optional embodiments, this application embodiment achieves multi-dimensional collaborative anti-interference in the spatial and polarization domains by performing interference suppression operations according to the interference suppression decision results. The spatial beamforming operation is used to calculate complex weight vectors based on the recommended null depth, controlling the amplitude and phase distribution of the phased array elements, and accurately generating deep nulls pointing towards the interference source in the spatial dimension, effectively suppressing interference energy in specific directions. The polarization adaptive control operation is used to adjust the amplitude ratio and phase difference of the biorthogonal polarization channels in the feed network according to the polarization state of the interference signal, making the communication beam polarization orthogonal to the interference polarization, and further attenuating the interference using polarization isolation. This mechanism fully utilizes spatial and polarization degrees of freedom, significantly improving the interference suppression depth and robustness, effectively reducing main lobe gain loss, maintaining a high signal-to-interference-plus-noise ratio, and achieving efficient adaptive interference suppression.

[0155] Optionally, performing interference suppression operations based on the interference suppression decision results includes: in response to the interference type label indicating that the interference signal is pulse interference, performing a time-domain resource cooperative scheduling operation based on the time-domain scheduling duty cycle detection threshold and the time duty cycle of the interference signal, wherein the time-domain resource cooperative scheduling operation is used to generate an RF switch control signal.

[0156] The aforementioned pulse interference can be an interference signal with discontinuous, intermittent characteristics and a typically low duty cycle. Its characteristic is that the power is extremely high during the pulse duration, which can easily lead to RF link saturation.

[0157] The aforementioned time-domain resource collaborative scheduling operation can be a control process for the system to dynamically adjust the on / off state of the radio frequency channel according to the interference time characteristics.

[0158] The aforementioned RF switch control signal can be a digital control command used to control the on / off state of the fast RF switch array, inserted in the signal path after the low-noise amplifier and before the mixer of each channel.

[0159] For example, the system receives an interference type label from a lightweight neural network processing unit. When the label is identified as "narrowband impulse interference" or "wideband impulse interference", the subsequent time-domain scheduling process is triggered; otherwise, the step is skipped to maintain continuous operation mode.

[0160] In response to an interference type label indicating that the interference signal is impulse interference, a time-domain resource cooperative scheduling operation is performed based on the time-domain scheduling duty cycle detection threshold and the time duty cycle of the interference signal. For example, the time-domain resource cooperative scheduler compares the time duty cycle of the interference signal. The time-domain scheduling duty cycle detection threshold g3 is used. When the time duty cycle of the interference signal is less than the time-domain scheduling duty cycle detection threshold, i.e. The signal is identified as a sparse pulse mode, and fast RF switch control is initiated. When the time duty cycle of the interference signal is greater than or equal to the time-domain scheduling duty cycle detection threshold, i.e. It is determined to be a dense pulse mode, and the time domain switching action is not performed; it is suppressed only by relying on the spatial domain null.

[0161] In sparse pulse mode, the time-domain resource collaborative scheduling operation utilizes a pulse detection synchronization circuit to track the rising and falling edges of interference pulses in real time. This pulse detection synchronization circuit is essentially an envelope detector and a high-speed comparator. When the rising edge of an interference pulse is detected, a disconnect command is sent to the RF switch of that channel within a response time of less than 50 nanoseconds. This RF switch is inserted in the signal path after the LNA and before the mixer, placing the damaged channel in a high-impedance isolation state, blocking strong interference signals from entering subsequent RF links, and preventing LNA saturation. When the falling edge of an interference pulse is detected, a 1-microsecond delay is maintained to allow residual energy to dissipate before a closing command is issued, restoring normal reception. Based on the aforementioned comparison logic and the output of the pulse detection synchronization circuit, the time-domain resource collaborative scheduling operation generates precise RF switch control signals to control the on / off state of the RF switch array, achieving rapid physical isolation of the damaged channel.

[0162] In an optional embodiment, for a multi-beam satellite communication system, when the main beam direction is subjected to pulse interference and the angle between the interference direction and the desired communication direction is <10° (i.e., the spatial domain cannot be effectively distinguished), the time-domain scheduler also controls the beamformer to instantaneously switch the communication beam to an auxiliary backup beam during the pulse duration. This auxiliary backup beam has a redundancy weight set pre-stored in ROM, and its direction is offset from the main beam by 1.5 power beamwidth, thereby avoiding the interference direction while sacrificing a small amount of gain. The backup beam switching is completed within <50μs.

[0163] Based on the above optional embodiments, this application embodiment, in response to the interference type label indicating that the interference signal is pulse interference, performs time-domain resource collaborative scheduling operation and generates radio frequency switch control signal according to the time-domain scheduling duty cycle detection threshold and the time duty cycle of the interference signal. It can quickly disconnect the damaged channel when pulse interference arrives, block the strong interference signal from entering the subsequent radio frequency link, prevent the low noise amplifier from saturating, and restore communication during the pulse gap, thereby effectively obtaining an additional 20 to 30 dB of suppression gain and improving the reliability of the system in a strong interference environment.

[0164] Optionally, the signal interference suppression method further includes: acquiring signal-to-interference-plus-noise ratio (SIR) data after interference suppression operation at a preset frequency; and, in response to the fact that the improvement in SIR data after multiple consecutive interference suppression operations is less than a preset improvement threshold, fine-tuning the output layer weights of the target interference analysis model using online training samples to generate an updated target interference analysis model, wherein the online training samples include: electromagnetic environment sensing data corresponding to preset historical frames and actual suppression strategy parameters.

[0165] The preset frequency can be the time interval or frame rate at which the system periodically acquires signal-to-interference-plus-noise ratio (SINR) data. Specifically, the preset frequency can be 100 frames, or 1 second.

[0166] The above signal-to-interference-plus-noise ratio data It is a metric for measuring communication quality, representing the ratio of expected signal power to the sum of noise and interference residual power, and is used to evaluate the effectiveness of current anti-interference measures.

[0167] The aforementioned improvement can be the difference between the current frame's signal-to-interference-plus-noise ratio (SINR) and the previous frame's SINR, used to quantify the performance improvement brought about by this interference suppression.

[0168] The above-mentioned preset improvement threshold This can be used as a critical value to determine whether the interference suppression effect meets the standard. If the improvement is lower than this value, the suppression effect is considered poor. For example, the preset improvement threshold can be set to 5dB.

[0169] The output layer weights mentioned above can be the parameter matrix inside the fully connected layer in the target interference analysis model, used to map the features of the previous layer to the final output result.

[0170] The aforementioned online training samples can be data pairs dynamically generated during system operation for model updates, including electromagnetic environment perception data and actual suppression strategy parameters.

[0171] The aforementioned preset historical frames can be a set of the most recent communication data frames selected by the system to construct online training samples before triggering the on-orbit incremental learning mechanism.

[0172] The system acquires the signal-to-interference-plus-noise ratio (SIR) data after interference suppression operations according to a preset frequency. For example, after each communication frame ends, the system calculates and records the SIR data of the current frame after interference suppression operations, based on a preset frequency rule, providing real-time data support for subsequent performance evaluation. The formula for calculating the SIR data is:

[0173]

[0174] in, The desired signal power is estimated using the pilot signal. For noise power, This refers to the residual interference power after suppression by the aforementioned spatial beamforming and polarization adaptive control operations.

[0175] The system will use the current frame A frame in history Compare and calculate the improvement amount The formula is: At the same time, the system also records a snapshot of the suppression actions taken in the current frame. The parameters include null, which specifically refers to the angle of arrival (DOA) of the interference signal, consisting of the azimuth and elevation angles.

[0176] If the improvement in the signal-to-interference-plus-noise ratio (SIR) after multiple consecutive interference suppression operations is less than a preset improvement threshold, the output layer weights of the target interference analysis model are fine-tuned using online training samples to generate an updated target interference analysis model. For example, the on-orbit incremental learning controller executes a judgment logic every 100 frames; if the same interference type label... ,continuous =3 times the improvement in signal-to-interference-plus-noise ratio data Less than the preset minimum improvement threshold For example, if the value is 5 dB, it indicates that the current model's suppression strategy for this type of interference has become "passivated" and needs adjustment, triggering the on-orbit incremental learning mechanism. Possible reasons for passivation include: gradual increase in interference power, slow rotation of interference polarization, or systematic bias in DOA estimation caused by interference source migration.

[0177] After the update is triggered, the system reads the sensor features from the most recent 100 frames from the FPGA's block random access memory (BRAM). And the optimal suppression strategy parameters measured in practice These two constitute a set of online training samples. Specifically, by attempting small perturbations of g1 and g2 in subsequent frames and observing the SINR feedback, the current optimal value is obtained through online search. .

[0178] The system will freeze all layer parameters of the neural network except for the output fully connected layer FC2, while keeping the weights unchanged, and only apply the weight matrix of this fully connected layer. Fine-tuning is performed, and the loss function is defined as the mean square error between the predicted output and the actual optimal suppression policy parameters. The calculation formula is as follows:

[0179]

[0180] Where M is the number of samples collected, such as M=100. This is the policy parameter vector for the current model's prediction output for frame t.

[0181] Using the Stochastic Gradient Descent (SGD) algorithm with a learning rate of 0.001 and performing only one iteration, the update formula is as follows:

[0182]

[0183] in, This is the updated weight matrix for the fully connected layer (FC2). This is the weight matrix of the fully connected layer (FC2) before the update. For learning rate, Represents the loss function L Weights of fully connected layers The partial derivatives of the weights, or gradients, are used to indicate the direction and magnitude of weight adjustments.

[0184] New weights The old weights are replaced by writing them into the Block RAM of the FPGA after verification by error check code. For example, Cyclic Redundancy Check (CRC) code can be used. The total data volume of this update process is less than 1k parameters, the weight update calculation time is less than 2ms, and the whole process is completed on the satellite without the need for data interaction with the ground station.

[0185] Through the above operations, the system reads the electromagnetic environment sensing data corresponding to the most recent preset historical frames from the flash memory and combines it with the actual suppression strategy parameters obtained through online search to construct online training samples. Subsequently, the system freezes all layer parameters in the target interference analysis model except for the output layer, and only uses the online training samples to perform stochastic gradient descent iterative updates on the output layer weights of the target interference analysis model based on the mean square error loss function, thereby generating an updated target interference analysis model that can more accurately adapt to changes in the current electromagnetic environment.

[0186] In an optional embodiment, to prevent malicious injection or erroneous updates caused by single-event flips, this embodiment of the application uses Triple Modular Redundancy (TMR) to store the weight parameters of the FC2 layer in three independent SRAM regions and performs a two-out-of-three vote each time it is read, to ensure that single-event flips do not cause erroneous weights to be loaded.

[0187] This application embodiment sets up an update rollback mechanism. The new weights after each update are first stored in a temporary buffer and then verified within the next 10 frames. If the signal-to-interference-plus-noise ratio (SIR) decreases instead of increasing, the new weights are automatically abandoned and rolled back to the old weights before the update, and the update is marked as failed in the system log.

[0188] This application embodiment sets up ground authorization filtering. The system only allows autonomous updates to the weights of the last layer. Significant changes to the deeper structure and weights of the network still require authorization via ground control stations, thus preventing the model from being maliciously and completely tampered with from a system security perspective.

[0189] Based on the above optional embodiments, this application embodiment acquires the signal-to-interference-plus-noise ratio (SIR) data after interference suppression operations at a preset frequency. When the improvement in SIR data after multiple consecutive interference suppression operations is less than a preset improvement threshold, the output layer weights of the target interference analysis model are fine-tuned using online training samples, including electromagnetic environment perception data corresponding to preset historical frames and actual suppression strategy parameters, to generate an updated target interference analysis model. This realizes the system's on-orbit self-evolution capability in complex electromagnetic environments, enabling the model to automatically adapt to dynamic changes such as gradual changes in interference power, polarization rotation, or DOA deviation, gradually improving the accuracy of identification and suppression depth of specific interference sources. At the same time, it avoids the time cost of retraining the model and ensures the long-term robustness of the system under limited spaceborne resources.

[0190] Figure 4 This is a schematic diagram of an optional signal interference suppression method according to an embodiment of this application, as shown below. Figure 4 As shown, this method achieves a complete closed loop from perception to self-learning.

[0191] Step S401, Multi-dimensional Electromagnetic Environment Sensing. The system first divides the communication frame into communication time slots and sensing time slots. Within the sensing time slot, eight array elements at the edge of the array are multiplexed to form a sensing subarray. A rapid scan of 64 beam directions is completed within 200μs using a pre-stored codebook. The received digital intermediate frequency signal is then subjected to covariance matrix eigenvalue decomposition to extract the angle of arrival (DOA) of the interference signal. Simultaneously, the polarization ellipse tilt angle and axial ratio are calculated using dual polarization channels. Combined with FFT analysis, the center frequency, bandwidth, power intensity, and time duty cycle are obtained. Finally, a four-dimensional feature vector F containing the spatial domain, polarization domain, frequency domain, and energy domain is constructed.

[0192] Step S402: Interference feature extraction and classification based on a lightweight neural network. The system first normalizes the four-dimensional feature vector to eliminate dimensional differences, and then performs normalized feature extraction and classification. Input a pre-trained one-dimensional convolutional neural network (1D-CNN), which includes convolutional layers, pooling layers, and fully connected layers. After INT8 quantization to accelerate inference, it directly outputs interference type labels. And the corresponding suppression strategy parameter vector G.

[0193] Step S403: Adaptive beamforming and polarization co-shaping. The beamformer, based on the output interference DOA and recommended null depth, uses a closed-form algorithm based on weighted least squares to quickly calculate the complex weight vector. While ensuring the gain in the desired direction, it forms a null of a specified depth in the interference direction. Iterative correction ensures that the main lobe gain loss is less than 0.5 dB. Simultaneously, the polarization controller adjusts the amplitude ratio and phase difference of the dual-polarization channels in the feed network according to the polarization switching flag g2 and the interference polarization state, making the communication beam polarization orthogonal to the interference polarization, thereby achieving dual suppression in both the spatial and polarization domains.

[0194] Step S404: Dynamic resource scheduling based on interference time characteristics. If the interference type label identifies it as pulse interference and the duty cycle is below the threshold, it is determined to be a sparse pulse mode. The time-domain scheduler uses an envelope detector to track the interference pulse in real time. When the pulse rising edge arrives, the RF switch is quickly disconnected to block strong interference from entering the receiving link and prevent saturation. Reception is restored after a delay following the pulse falling edge, or when it cannot be distinguished in the spatial domain, it switches to a backup beam to achieve interference avoidance in the time dimension. If it is identified as pulse interference and the duty cycle is not below the threshold, it is determined to be a dense pulse mode, and no switching action is performed. If it is not identified as pulse interference, the continuous operation mode is maintained.

[0195] Step S405: On-orbit performance evaluation and incremental self-learning. The system calculates the improvement in signal-to-interference-plus-noise ratio (SNR) for the current frame. When the improvement amount falls below a preset threshold for three consecutive times, the model is considered to be inactive, triggering the incremental learning module. The deep weights of the network are frozen, and the weights of the last fully connected layer are fine-tuned using stochastic gradient descent (SDG) based on mean squared error, using only the most recent 100 frames of data. The updated weights are then written to memory after being confirmed as valid by a triple modulo redundancy check (TMR) and rollback verification mechanism, thereby achieving on-orbit adaptive evolution and reliability assurance of the spaceborne system under limited resources.

[0196] According to another aspect of the embodiments of this application, a signal interference suppression device is also provided. Figure 5 This is a structural block diagram of an optional signal interference suppression device according to an embodiment of this application, such as... Figure 5 As shown. The device includes: a first acquisition module 501, used to acquire electromagnetic environment sensing data, wherein the electromagnetic environment sensing data includes multi-dimensional feature data, including: spatial domain feature data, polarization domain feature data, frequency domain feature data, and energy domain feature data; an analysis module 502, used to perform predictive analysis on the electromagnetic environment sensing data using a target interference analysis model to obtain interference suppression decision results, wherein the target interference analysis model is used to map the multi-dimensional feature data into interference type labels and suppression strategy parameter vectors; and an execution module 503, used to execute interference suppression operations according to the interference suppression decision results.

[0197] Optionally, the first acquisition module 501 is further configured to: perform beam scanning of the entire spatial domain using a sensing subarray within the sensing time slot to obtain raw scanning data; perform signal conversion operations on the raw scanning data to obtain signal conversion results, wherein the signal conversion operations include at least one of the following: signal amplification operation, signal mixing operation, and signal sampling operation; and perform multi-dimensional feature extraction based on the signal conversion results to obtain electromagnetic environment sensing data.

[0198] Optionally, the target interference analysis model includes a feature preprocessing module and a prediction inference module. The analysis module 502 is further used to: normalize the electromagnetic environment sensing data using the feature preprocessing module to obtain the normalization result; and perform prediction analysis on the normalization result using the prediction inference module to obtain the interference suppression decision result. The prediction inference module is composed of a one-dimensional convolutional neural network.

[0199] Optionally, the spatial domain feature data includes: the azimuth and elevation angles of the interfering signal; the polarization domain feature data includes: the polarization ellipse tilt angle and axial ratio of the interfering signal; the frequency domain feature data includes: the center frequency and bandwidth of the interfering signal; and the energy domain feature data includes: the power intensity and time duty cycle of the interfering signal. The interfering type label includes at least one of the following: narrowband continuous interference, wideband continuous interference, narrowband pulse interference, wideband pulse interference, and spoofing signal interference. The suppression strategy parameter vector includes: recommended null depth, polarization switching flag, time-domain scheduling duty cycle detection threshold, and priority level.

[0200] Optionally, the execution module 503 is further configured to: perform spatial beamforming operation and polarization adaptive control operation according to the interference type label and the suppression strategy parameter vector, wherein the spatial beamforming operation is used to calculate a complex weight vector based on the recommended null depth to control the amplitude and phase distribution of the phased array elements, and the polarization adaptive control operation is used to adjust the amplitude ratio and phase difference of the biorthogonal polarization channels in the feed network according to the polarization state of the interference signal.

[0201] Optionally, the execution module 503 is further configured to: in response to the interference type label indicating that the interference signal is pulse interference, perform a time-domain resource cooperative scheduling operation based on the time-domain scheduling duty cycle detection threshold and the time duty cycle of the interference signal, wherein the time-domain resource cooperative scheduling operation is used to generate an RF switch control signal.

[0202] Optionally, the signal interference suppression device further includes: a second acquisition module, used to acquire signal-to-interference-plus-noise ratio (SIR) data after interference suppression operation at a preset frequency; and a fine-tuning module, used to fine-tune the output layer weights of the target interference analysis model using online training samples in response to the fact that the improvement amount of the SIR data after multiple consecutive interference suppression operations is less than a preset improvement threshold, thereby generating an updated target interference analysis model. The online training samples include: electromagnetic environment sensing data corresponding to preset historical frames and actual suppression strategy parameters.

[0203] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0204] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0205] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0206] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0207] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0208] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0209] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0210] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0211] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0212] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0213] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0214] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for suppressing signal interference, characterized in that, include: Acquire electromagnetic environment sensing data, wherein the electromagnetic environment sensing data includes multi-dimensional feature data, which includes: spatial domain feature data, polarization domain feature data, frequency domain feature data and energy domain feature data; The electromagnetic environment sensing data is predicted and analyzed using a target interference analysis model to obtain interference suppression decision results. The target interference analysis model is used to map the multidimensional feature data into interference type labels and suppression strategy parameter vectors. The interference suppression operation is performed based on the interference suppression decision result.

2. The signal interference suppression method according to claim 1, characterized in that, Acquiring the electromagnetic environment sensing data includes: Within the sensing time slot, a sensing subarray is used to perform beam scanning across the entire spatial domain to obtain the raw scanning data; A signal conversion operation is performed on the original scan data to obtain a signal conversion result, wherein the signal conversion operation includes at least one of the following: signal amplification operation, signal mixing operation, and signal sampling operation; Multidimensional feature extraction is performed based on the signal conversion results to obtain the electromagnetic environment sensing data.

3. The signal interference suppression method according to claim 1, characterized in that, The target interference analysis model includes a feature preprocessing module and a prediction inference module. The target interference analysis model is used to predict and analyze the electromagnetic environment sensing data to obtain the interference suppression decision results, including: The electromagnetic environment sensing data is normalized using the feature preprocessing module to obtain the normalized result. The prediction and inference module is used to predict and analyze the normalization result to obtain the interference suppression decision result. The prediction and inference module is composed of a one-dimensional convolutional neural network.

4. The signal interference suppression method according to any one of claims 1 to 3, characterized in that, The spatial domain feature data includes: the azimuth and elevation angles of the interference signal; the polarization domain feature data includes: the polarization ellipse tilt angle and axial ratio of the interference signal; the frequency domain feature data includes: the center frequency and bandwidth of the interference signal; and the energy domain feature data includes: the power intensity and time duty cycle of the interference signal. The interference type label includes at least one of the following: narrowband continuous interference, wideband continuous interference, narrowband pulse interference, wideband pulse interference, and spoofed signal interference; The suppression strategy parameter vector includes: recommended zero-depression depth, polarization switching flag, time-domain scheduling duty cycle detection threshold, and priority level.

5. The signal interference suppression method according to claim 4, characterized in that, Performing interference suppression operations based on the interference suppression decision results includes: Based on the interference type label and the suppression strategy parameter vector, spatial beamforming and polarization adaptive control operations are performed. The spatial beamforming operation is used to calculate a complex weight vector based on the recommended null depth to control the amplitude and phase distribution of the phased array elements. The polarization adaptive control operation is used to adjust the amplitude ratio and phase difference of the biorthogonal polarization channels in the feed network according to the polarization state of the interference signal.

6. The signal interference suppression method according to claim 4, characterized in that, Performing interference suppression operations based on the interference suppression decision results includes: In response to the interference type label indicating that the interference signal is pulse interference, a time-domain resource cooperative scheduling operation is performed based on the time-domain scheduling duty cycle detection threshold and the time duty cycle of the interference signal, wherein the time-domain resource cooperative scheduling operation is used to generate an RF switch control signal.

7. The signal interference suppression method according to claim 1, characterized in that, The signal interference suppression method further includes: The signal-to-interference-plus-noise ratio (SIR) data after the interference suppression operation is obtained according to a preset frequency. In response to the fact that the improvement in signal-to-interference-plus-noise ratio (SIR) after multiple consecutive interference suppression operations is less than a preset improvement threshold, the output layer weights of the target interference analysis model are fine-tuned using online training samples to generate an updated target interference analysis model. The online training samples include: electromagnetic environment sensing data corresponding to preset historical frames and actual suppression strategy parameters.

8. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images