Radar signal modulation recognition method and system driven by cross-domain transfer learning

Abstract

The application provides a radar signal modulation recognition method and system driven by cross-domain migration learning, obtains a time delay characteristic value set of a building reflection path and a background noise intensity change value set; based on the time delay characteristic value set and the background noise intensity change value set, generates multipath propagation parameters and noise distribution parameters, and according to a preset physical environment boundary rule, constructs a physical environment constraint condition, the physical environment constraint condition including a time delay boundary threshold and a noise boundary threshold; based on pre-stored source domain modulation sample data, obtains a source domain baseband signal; according to a time delay quantization value in the multipath propagation parameters, obtains a distortion signal carrying building reflection characteristics; according to the noise distribution parameters, obtains a target domain simulation signal; and by using the physical environment constraint condition, obtains a modulation recognition result of the target domain simulation signal; the technical scheme provided by the application realizes modulation recognition accuracy and generalization ability of a radar signal in a complex environment.

Description

Technical Field

[0001] This application relates to the field of radar signal processing technology, and in particular to a radar signal modulation recognition method and system driven by cross-domain transfer learning. Background Technology

[0002] In general wireless communication environments, radar signal modulation recognition faces challenges from complex electromagnetic environments. This requires recognition systems to be able to adapt to signal distortion under different propagation conditions and maintain stable recognition performance even in the absence of target domain annotation data. Therefore, there is an urgent need for a technical means to effectively transfer source domain knowledge and improve recognition accuracy under changing physical environments.

[0003] To address the aforementioned technical requirements, a mainstream approach in existing technologies is a modulation recognition method based on joint modeling of source domain data augmentation and deep neural networks. This method constructs a large-scale source domain radar signal sample library, uses data augmentation techniques to simulate some environmental interference, trains the augmented data, and finally fine-tunes the target domain signal to achieve the recognition task. However, existing solutions still have some shortcomings. For example, because the data augmentation process does not consider the physical characteristics of electromagnetic waves such as reflection and refraction between buildings, the generated simulated signal differs significantly from the real target domain signal, thus limiting the model's generalization ability. Furthermore, the lack of dynamic modeling of noise intensity and time delay distribution fails to effectively reflect the non-uniformity of noise and multipath effects in the actual environment, causing a decrease in the classifier's recognition accuracy when facing sudden environmental changes. Summary of the Invention

[0004] This application provides a radar signal modulation recognition method and system driven by cross-domain transfer learning, which solves the problems of large differences between analog signals and real target domain signals, insufficient generalization ability of the model, and decreased recognition accuracy of the classifier when the environment changes abruptly.

[0005] In a first aspect, this application provides a radar signal modulation recognition method driven by cross-domain transfer learning, including:

[0006] Obtain the set of time delay characteristic values ​​of the building reflection path and the set of background noise intensity variation values;

[0007] Based on the set of time delay feature values ​​and the set of background noise intensity change values, multipath propagation parameters and noise distribution parameters are generated. According to preset physical environment boundary rules, the multipath propagation parameters and noise distribution parameters are subjected to boundary processing to construct physical environment constraints, which include time delay boundary thresholds and noise boundary thresholds.

[0008] Based on pre-stored source domain modulation sample data, the baseband signal is reconstructed to obtain the source domain baseband signal;

[0009] Based on the time delay quantization value in the multipath propagation parameters, the corresponding reflection path delay is calculated, and the reflection path delay is superimposed on the source domain baseband signal to obtain a distorted signal carrying the building reflection characteristics.

[0010] Based on the noise distribution parameters, random noise is added to the distorted signal to obtain the target domain analog signal;

[0011] Using the physical environment constraints, the source domain baseband signal and the target domain analog signal are processed to obtain the modulation recognition result of the target domain analog signal.

[0012] Optionally, based on pre-stored source domain modulation sample data, the baseband signal is reconstructed to obtain the source domain baseband signal, including:

[0013] Modulation type parameters, symbol mapping parameters, and pulse shape parameters are extracted from the pre-stored source domain modulation sample data;

[0014] Based on the modulation type parameter, the modulation order is determined, and the bit length corresponding to each modulation symbol is determined based on the modulation order.

[0015] Based on the bit length, an original bit sequence is generated, and the original bit sequence is divided into multiple bit groups to generate a discrete bit sequence;

[0016] Based on the symbol mapping parameters, a corresponding discrete modulation symbol is assigned to each bit group to generate a discrete modulation symbol sequence;

[0017] Based on the pulse shape parameters, the discrete modulation symbol sequence is subjected to pulse shaping filtering to obtain the baseband waveform;

[0018] Calculate the carrier frequency offset of the baseband waveform, and compensate and correct the baseband waveform according to the carrier frequency offset to obtain the compensated baseband signal. Use the compensated baseband signal as the source domain baseband signal.

[0019] Optionally, the carrier frequency offset of the baseband waveform is calculated, and the baseband waveform is compensated and corrected according to the carrier frequency offset to obtain a compensated baseband signal. The compensated baseband signal is then used as the source domain baseband signal, including:

[0020] Based on the pulse shape parameters, the center frequency of the spectrum is calculated and used as the reference carrier frequency value.

[0021] The carrier frequency value is extracted from the baseband waveform, and the frequency compensation factor is calculated by combining it with the reference carrier frequency value.

[0022] Based on the frequency compensation factor and the sampling time sequence of the baseband waveform, the phase rotation angle at each sampling moment in the baseband waveform is calculated, and the frequency offset correction signal is obtained by multiplying each complex sample value of the baseband waveform with the corresponding phase rotation angle.

[0023] The frequency offset correction signal is compensated and corrected to generate a compensated baseband signal, which is then used as the source domain baseband signal.

[0024] Optionally, based on the time delay quantization value in the multipath propagation parameters, the corresponding reflection path delay is calculated, and the reflection path delay is superimposed on the source domain baseband signal to obtain a distorted signal carrying the building reflection characteristics, including:

[0025] Obtain the material type of the building environment and query the preset building material reflectance coefficient database to obtain the corresponding reflectance attenuation factor;

[0026] The time delay quantization value corresponding to each reflection path in the multipath propagation parameters is weighted and calculated with the reflection attenuation factor to obtain the corresponding reflection path delay.

[0027] Based on the preset multipath reflection order rule, the reflection path delay is converted into a time axis offset parameter sequence;

[0028] The source domain baseband signal is copied in its entirety to obtain a lossless signal copy;

[0029] Based on the time axis offset parameter sequence, a time delay operation is performed on the lossless signal copy to obtain a time offset signal;

[0030] The time offset signal and the source domain baseband signal are superimposed to obtain a distorted signal carrying the building's reflection characteristics.

[0031] Optionally, random noise is added to the distorted signal according to the noise distribution parameters to obtain a target domain analog signal, including:

[0032] Based on a preset intensity amplitude conversion coefficient, the noise distribution parameters are linearly scaled to obtain a noise amplitude reference value;

[0033] Based on the preset floating ratio parameter, the floating boundary value of the noise amplitude reference value is calculated. Within the range of the floating boundary value, the noise amplitude value is randomly generated to generate a random noise sequence.

[0034] Based on the number of sampling points of the distorted signal, the length of the random noise sequence is adjusted to obtain a length-aligned random noise sequence;

[0035] The length-aligned random noise sequence and the distorted signal are time-aligned to obtain a synchronization noise signal;

[0036] The synchronization noise signal and the distortion signal are superimposed to obtain the target domain analog signal.

[0037] Optionally, using the physical environment constraints, the source domain baseband signal and the target domain analog signal are processed to obtain the modulation identification result of the target domain analog signal, including:

[0038] Multiple source domain feature values ​​are extracted from the segmented data of the source domain baseband signal, multiple target domain feature values ​​are extracted from the segmented data of the target domain analog signal, and all source domain feature values ​​are combined to obtain a set of source domain feature vectors. All target domain feature values ​​are combined to obtain a set of target domain feature vectors.

[0039] Based on the time delay boundary threshold in the physical environment constraints, calculate the maximum allowable distribution offset between the source domain feature vector set and the target domain feature vector set;

[0040] Using the maximum allowable distribution offset and the preset scaling factor, the characteristic value fluctuation tolerance range of the target domain analog signal is calculated. Within the characteristic value fluctuation tolerance range, the target domain feature vector set is adjusted to obtain the adjusted target domain feature vector set.

[0041] Modulation pattern matching is performed on the adjusted target domain feature vector set to obtain the modulation recognition result of the target domain analog signal.

[0042] Optionally, using the maximum permissible distribution offset and a preset scaling factor, the eigenvalue fluctuation tolerance range of the target domain analog signal is calculated. Within the eigenvalue fluctuation tolerance range, the target domain feature vector set is adjusted to obtain an adjusted target domain feature vector set, including:

[0043] The maximum permissible distribution offset and the preset proportional coefficient are multiplied to obtain the characteristic fluctuation tolerance radius of the target domain.

[0044] The target domain feature vector set is decomposed into feature dimensions to obtain numerical sequences of different feature dimensions, and the average value of each numerical sequence is calculated. The average value is used as the center value of the corresponding feature dimension.

[0045] Based on the characteristic fluctuation tolerance radius, calculate the upper and lower tolerance limits of the center value of each characteristic dimension, and combine the upper and lower tolerance limits of all characteristic dimensions to obtain the characteristic value fluctuation tolerance range of the target domain analog signal.

[0046] Within the tolerance range of the eigenvalue fluctuations, each numerical sequence is subjected to boundary constraint processing to obtain multiple boundary-constrained numerical sequences;

[0047] All the numerical sequences after boundary constraints are reconstructed into vectors to generate an adjusted set of target domain feature vectors.

[0048] Secondly, this application provides a radar signal modulation recognition system driven by cross-domain transfer learning, comprising:

[0049] The acquisition module is used to acquire the set of time delay characteristic values ​​of the building reflection path and the set of background noise intensity change values;

[0050] The construction module is used to generate multipath propagation parameters and noise distribution parameters based on the set of time delay feature values ​​and the set of background noise intensity change values, and to perform boundary processing on the multipath propagation parameters and the noise distribution parameters according to preset physical environment boundary rules to construct physical environment constraints, the physical environment constraints including time delay boundary thresholds and noise boundary thresholds;

[0051] The reconstruction module is used to reconstruct the baseband signal based on pre-stored source domain modulation sample data to obtain the source domain baseband signal;

[0052] The calculation module is used to calculate the corresponding reflection path delay based on the time delay quantization value in the multipath propagation parameters, and to superimpose the reflection path delay onto the source domain baseband signal to obtain a distorted signal carrying the building reflection characteristics.

[0053] An addition module is used to add random noise to the distorted signal according to the noise distribution parameters to obtain a target domain analog signal;

[0054] The processing module uses the physical environment constraints to process the source domain baseband signal and the target domain analog signal to obtain the modulation recognition result of the target domain analog signal.

[0055] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement a cross-domain transfer learning-driven radar signal modulation recognition method as described in the first aspect above.

[0056] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a radar signal modulation recognition method driven by cross-domain transfer learning as described in the first aspect.

[0057] In this application, a set of time delay characteristic values ​​and a set of background noise intensity variation values ​​for the building reflection path are obtained; based on the set of time delay characteristic values ​​and the set of background noise intensity variation values, multipath propagation parameters and noise distribution parameters are generated, and boundary processing is performed on the multipath propagation parameters and the noise distribution parameters according to preset physical environment boundary rules to construct physical environment constraints, which include time delay boundary thresholds and noise boundary thresholds; based on pre-stored source domain modulation sample data, the baseband signal is reconstructed to obtain the source domain baseband signal; according to the time delay quantization value in the multipath propagation parameters, the corresponding reflection path delay is calculated, and the reflection path delay is superimposed on the source domain baseband signal to obtain a distorted signal carrying the building reflection characteristics; according to the noise distribution parameters, random noise is added to the distorted signal to obtain a target domain simulated signal; the source domain baseband signal and the target domain simulated signal are processed using the physical environment constraints. The technical solution provided in this application processes the obtained set of time delay feature values ​​and the set of background noise intensity variation values, and combines them with preset physical environment boundary rules to obtain physical environment constraints, thereby improving the realism and applicability of the simulation. Using pre-stored source domain modulation sample data, a source domain baseband signal is constructed, providing a high-quality, structurally clear foundation signal for subsequent cross-domain migration. By introducing time delay quantization values ​​related to the actual building reflection characteristics, a distorted signal that better conforms to the propagation characteristics of the target domain is simulated. Dynamic noise interference is applied to the distorted signal in conjunction with noise distribution parameters, making the target domain simulated signal closer to the complexity of the actual received signal, thereby improving the classifier's adaptability and recognition robustness in different electromagnetic environments. Furthermore, when reconstructing the baseband signal, this application extracts key parameters such as modulation type, symbol mapping, and pulse shape from the pre-stored source domain modulation sample data, and generates discrete bit sequences and discrete modulation symbol sequences accordingly. Further, through pulse shaping filtering and carrier frequency offset compensation, a source domain baseband signal with physical consistency and modulation characteristics is constructed. This solves the problem of limited migration effects caused by inaccurate signal modeling in existing solutions.

[0058] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1A flowchart of a radar signal modulation recognition method driven by cross-domain transfer learning provided in this application;

[0061] Figure 2 A schematic diagram of the structure of a radar signal modulation recognition system driven by cross-domain transfer learning provided in this application;

[0062] Figure 3 A schematic diagram of the structure of a computing device provided in this application. Detailed Implementation

[0063] 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.

[0064] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0065] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] This application addresses the common problems of multipath interference and dynamic noise changes in radar signal modulation identification tasks in general wireless communication environments, proposing a cross-domain transfer learning-driven identification method. Existing technologies mainly rely on joint modeling of source domain data augmentation and deep neural networks. While they improve model adaptability by expanding the diversity of source domain data when target domain data is scarce, their data augmentation process lacks modeling of the physical propagation characteristics of electromagnetic waves in complex environments, such as reflection and refraction. This leads to discrepancies between the generated simulated signal and the actual received signal, making it difficult to effectively reflect the complexity and non-uniformity of the target domain environment, thus limiting the model's generalization ability and identification stability. Therefore, this application introduces a physical environment constraint mechanism. By collecting the time delay characteristics of building reflection paths and the changes in background noise intensity, it constructs physically consistent multipath propagation parameters and noise distribution parameters. Based on these, it generates a target domain simulated signal that closely approximates real propagation conditions. Combined with cross-domain transfer learning, it achieves effective transfer of source domain knowledge to the target domain, thereby improving adaptability and accuracy in complex electromagnetic environments without relying on a large amount of labeled target domain data.

[0067] Figure 1 A flowchart of a radar signal modulation recognition method driven by cross-domain transfer learning is provided in this application embodiment, as follows: Figure 1 As shown, the method includes:

[0068] Step 101: Obtain the set of time delay feature values ​​of the building reflection path and the set of background noise intensity change values.

[0069] In this step, the building reflection path refers to the propagation path of radar signals reflected from building surfaces in an urban environment. The time delay characteristic value set refers to the time delay measurements containing multiple reflection paths, reflecting the multipath effect of building clusters. The background noise intensity variation value set refers to the dynamic sampling sequence of environmental noise intensity, characterizing electromagnetic interference fluctuations.

[0070] In this embodiment of the application, the propagation delay of the reflected signal of the urban building complex is measured to obtain a set of delay characteristic values ​​of the building reflection path; at the same time, dynamic change data of the background noise intensity of the wireless communication environment is collected to obtain a set of background noise intensity change values.

[0071] Step 102: Based on the set of time delay feature values ​​and the set of background noise intensity change values, generate multipath propagation parameters and noise distribution parameters, and perform boundary processing on the multipath propagation parameters and noise distribution parameters according to preset physical environment boundary rules to construct physical environment constraints, including time delay boundary thresholds and noise boundary thresholds.

[0072] In this step, the preset physical environment boundary rules refer to predefined transformation rules used to constrain the range of physical effects. The physical environment constraints refer to a combination of boundary sets including time delay boundary thresholds and noise boundary thresholds, used to limit characteristic distribution offsets. The time delay boundary threshold refers to the maximum permissible time delay offset calculated based on building reflection characteristics, used to control multipath distortion amplitude. The noise boundary threshold refers to the extreme value of amplitude fluctuation calculated based on environmental noise, used to constrain the range of noise pollution.

[0073] In this embodiment, a discretization and quantization operation is performed on the set of time delay feature values ​​to obtain multipath propagation parameters. The dynamic range of the set of background noise intensity change values ​​is calculated to obtain noise distribution parameters. Based on preset physical environment boundary rules, boundary threshold extraction is performed on the multipath propagation parameters and noise distribution parameters to generate physical environment constraints containing time delay boundary thresholds and noise boundary thresholds.

[0074] Step 103: Based on the pre-stored source domain modulation sample data, reconstruct the baseband signal to obtain the source domain baseband signal.

[0075] In this step, the pre-stored source domain modulation sample data refers to the historical radar signal dataset. The baseband signal refers to the raw signal waveform without carrier modulation. The source domain baseband signal refers to the baseband signal reconstructed from the source domain modulation sample data.

[0076] In this embodiment, modulation type parameters, symbol mapping parameters, and pulse shape parameters are extracted from pre-stored source domain modulation sample data to generate a discrete bit sequence. A baseband waveform is generated using pulse shaping filtering, and the source domain baseband signal is output after carrier frequency compensation.

[0077] Step 104: Calculate the corresponding reflection path delay based on the time delay quantization value in the multipath propagation parameters, and superimpose the reflection path delay onto the source domain baseband signal to obtain a distorted signal carrying the building reflection characteristics.

[0078] In this step, the time delay quantization value refers to the discrete time delay value in the multipath propagation parameters, representing the delay of a single path. The reflection path delay refers to the processing parameter obtained by scaling the time delay quantization value according to the building material attenuation factor. The distorted signal refers to the output signal after the source domain baseband signal is superimposed with the building reflection effect, exhibiting multipath interference characteristics.

[0079] In this embodiment, the time delay quantization value is extracted from the multipath propagation parameters and the reflection path delay is calculated. A time axis offset parameter sequence is generated, the source domain baseband signal is copied and segmented time shift is performed, and the amplitude is superimposed to obtain a distorted signal carrying the building reflection characteristics.

[0080] Step 105: Add random noise to the distorted signal according to the noise distribution parameters to obtain the target domain analog signal.

[0081] In this step, the target domain analog signal refers to the synthesized signal after the distorted signal is superimposed with environmental noise, which simulates the reception effect of the real electromagnetic environment.

[0082] In this embodiment, a noise amplitude fluctuation range is calculated based on the noise distribution parameters through linear scaling and floating range; a random noise sequence is generated by performing a uniform random sampling operation within the fluctuation range; the length of the random noise sequence is adjusted to match the number of sampling points of the distorted signal; a timestamp matching operation is performed between the length-aligned random noise sequence and the distorted signal to generate a synchronous noise signal; a point-by-point amplitude addition operation is performed between the synchronous noise signal and the distorted signal to output the target domain analog signal.

[0083] Step 106: Using the physical environment constraints, process the source domain baseband signal and the target domain analog signal to obtain the modulation recognition result of the target domain analog signal.

[0084] In this step, the modulation identification result refers to the output of the modulation category determination of the target domain analog signal.

[0085] In this embodiment, a segmented feature extraction operation is performed on the source domain baseband signal and the target domain analog signal to generate a feature vector set. The maximum allowable distribution offset is calculated based on the time delay boundary threshold in the physical environment constraints. The feature value fluctuation tolerance range is generated according to the maximum allowable distribution offset and the preset proportional coefficient. Within the tolerance range, a boundary constraint operation is performed on the target domain feature vector set to generate an adjusted feature vector set. The nearest neighbor modulation mode matching operation is performed on the adjusted feature vector set, and the modulation identification result of the target domain analog signal is output.

[0086] This application embodiment uses a physical constraint-driven cross-domain transfer learning mechanism to convert the physical measurements of building reflection delay and background noise into feature distribution constraint boundaries. When reconstructing the baseband signal, multipath distortion and noise pollution effects are implanted to generate a high-fidelity target domain signal. Finally, feature distribution adaptation from the source domain to the target domain is achieved under physical boundary constraints, solving the domain adaptation problem of radar signal modulation recognition in complex urban electromagnetic environments.

[0087] This application provides a specific embodiment. Step 103 involves reconstructing the baseband signal based on pre-stored source domain modulation sample data to obtain the source domain baseband signal. This specifically includes the following steps:

[0088] Step 301: Extract the modulation type parameter, symbol mapping parameter, and pulse shape parameter from the pre-stored source domain modulation sample data.

[0089] In this step, the modulation type parameter refers to the modulation identifier, which determines the spatial dimension of the signal. The symbol mapping parameter refers to the mapping rule from bits to complex numbers. The pulse shape parameter defines the eigenvalues ​​of the waveform filter and controls the bandwidth.

[0090] In this embodiment, modulation type parameters, symbol mapping parameters, and pulse shape parameters are separated from source domain modulation sample data using data parsing techniques.

[0091] Step 302: Determine the modulation order based on the modulation type parameter, and determine the bit length corresponding to each modulation symbol based on the modulation order.

[0092] In this step, the modulation order refers to the number of bits carried by a single symbol, which determines the spectral efficiency. A modulation symbol refers to a discrete point on the complex plane that carries modulation information. The bit length refers to the number of bits corresponding to each symbol.

[0093] In this embodiment, based on the modulation type parameter, a preset mapping table is queried to determine the modulation order, and the logarithm of the modulation order to the base 2 is taken to obtain the bit length corresponding to each modulation symbol, such as log24 = 2 bits.

[0094] Step 303: Based on the bit length, generate the original bit sequence and divide the original bit sequence into multiple bit groups to generate a discrete bit sequence.

[0095] In this step, the original bit sequence refers to the binary stream to be modulated. A bit group refers to a segment grouped by bit length. A discrete bit sequence refers to a set of structured bit groups.

[0096] In this embodiment of the application, a random binary original bit sequence is generated based on the bit length. The calculation formula is: the length of the original bit sequence = the number of symbols × the bit length, such as 011001. The bits are grouped according to the bit length to obtain bit groups, such as 01,10,01, and the discrete bit sequence is output.

[0097] Step 304: According to the symbol mapping parameters, assign a corresponding discrete modulation symbol to each bit group to generate a discrete modulation symbol sequence.

[0098] In this step, a discrete modulation symbol refers to a complex-domain information unit, generated by bit group mapping. A discrete modulation symbol sequence refers to a set of timing symbols.

[0099] In this embodiment of the application, each bit group is mapped to complex plane coordinates according to the symbol mapping parameters, such as 01 being mapped to (-1+j), to generate a discrete modulation symbol sequence, such as (-1+j, 1-j, -1+j).

[0100] Step 305: Perform pulse shaping filtering on the discrete modulation symbol sequence according to the pulse shape parameters to obtain the baseband waveform.

[0101] In this step, the baseband waveform refers to a continuous signal in the time domain, which is output after pulse shaping and filtering.

[0102] In this embodiment, the discrete modulation symbol sequence and the filter coefficients defined by the pulse shape parameters are convolved to output a time-domain continuous baseband waveform.

[0103] Step 306: Calculate the carrier frequency offset of the baseband waveform, and compensate and correct the baseband waveform according to the carrier frequency offset to obtain the compensated baseband signal, and use the compensated baseband signal as the source domain baseband signal.

[0104] In this step, the carrier frequency offset refers to the algebraic difference between the actual carrier frequency and the theoretical carrier frequency, used for compensation. The compensated baseband signal refers to the time-domain waveform after frequency offset correction, satisfying the condition for distortion-free transmission.

[0105] In this embodiment, a Fourier transform is performed on the baseband waveform to extract the main lobe center frequency as the actual carrier frequency. The actual carrier frequency is subtracted from the theoretical carrier frequency of the pulse shape parameter to obtain the carrier frequency offset. A complex exponential compensation factor is constructed based on the offset. The complex exponential compensation factor is multiplied point by point with the baseband waveform to generate the compensated baseband signal and output as the source domain baseband signal.

[0106] This application embodiment transforms modulation type parameters into symbol mapping rules through a parameterized signal reconstruction chain, generates discrete symbol sequences based on bit length, and achieves bandwidth-controlled baseband waveform generation by combining pulse shape parameters. Finally, it eliminates reconstruction errors through carrier frequency offset compensation, providing a high-fidelity source domain signal foundation for cross-domain transfer learning.

[0107] This application provides a specific embodiment. Step 306 involves calculating the carrier frequency offset of the baseband waveform, compensating and correcting the baseband waveform based on the carrier frequency offset to obtain a compensated baseband signal, and using the compensated baseband signal as the source domain baseband signal. The specific steps include:

[0108] Step 311: Calculate the center frequency of the spectrum based on the pulse shape parameters, and use the center frequency of the spectrum as the reference carrier frequency value.

[0109] In this step, the spectral center frequency refers to the frequency value corresponding to the center point of the main lobe of the pulse shaping filter spectrum. The reference carrier frequency refers to the theoretically designed carrier center frequency, which serves as a reference for frequency offset compensation.

[0110] In this embodiment of the application, the center frequency of the spectrum is calculated based on the roll-off coefficient and symbol period of the pulse shape parameters. The calculation formula is: center frequency of spectrum = (1 + roll-off coefficient) / (2 × symbol period), and the center frequency of the spectrum is used as the reference carrier frequency value.

[0111] Step 312: Extract the carrier frequency value from the baseband waveform, and calculate the frequency compensation factor by combining it with the reference carrier frequency value.

[0112] In this step, the carrier frequency value refers to the actual radio frequency center frequency carried by the baseband waveform, which is measured through spectrum analysis. The frequency compensation factor is the negative reciprocal of the carrier frequency offset, used to construct the phase rotation complex exponential factor.

[0113] In this embodiment, a fast Fourier transform operation is performed on the baseband waveform to extract the carrier frequency value corresponding to the peak position of the main lobe of the spectrum. The difference between the carrier frequency value and the reference carrier frequency value is obtained, and the negative reciprocal of the difference is taken to obtain the frequency compensation factor.

[0114] Step 313: Based on the frequency compensation factor and the sampling time series of the baseband waveform, calculate the phase rotation angle at each sampling moment in the baseband waveform, and multiply each complex sample value of the baseband waveform with the corresponding phase rotation angle to obtain the frequency offset correction signal.

[0115] In this step, the phase rotation angle refers to the phase value that needs to be compensated at each sampling point. The complex sample value refers to the complex value of the baseband waveform at the sampling time. The frequency offset correction signal refers to the time-domain waveform generated by multiplying the complex sample value by a complex exponential factor, which eliminates carrier frequency deviation.

[0116] In this embodiment, the sampling time sequence of the baseband waveform is obtained, and the phase rotation angle is calculated. The calculation formula is: frequency compensation factor × 2π × each sampling time. The phase rotation angle is converted into a complex exponential factor. Each complex sample value of the baseband waveform is multiplied by the complex exponential factor of the corresponding sampling time to perform a complex multiplication operation and output the frequency offset correction signal.

[0117] Step 314: Compensate and correct the frequency offset correction signal to generate a compensated baseband signal, and use the compensated baseband signal as the source domain baseband signal.

[0118] In this embodiment, a band-limited constraint verification operation is performed on the frequency offset correction signal to verify whether the highest frequency of the frequency offset correction signal is less than half of the sampling rate of the frequency offset correction signal. If the verification passes, the compensated baseband signal is directly output and used as the source domain baseband signal.

[0119] This application's embodiments calculate a compensation factor based on the difference between the theoretical carrier frequency and the actual carrier frequency, generate a phase rotation angle based on the sampling time series, and use complex exponential multiplication to achieve time-domain phase rotation, effectively eliminating waveform distortion caused by hardware carrier frequency offset, and providing a source domain baseband signal with precise frequency alignment for cross-domain transfer learning.

[0120] This application provides a specific embodiment. Step 104 involves calculating the corresponding reflection path delay based on the time delay quantization value in the multipath propagation parameters, and superimposing the reflection path delay onto the source domain baseband signal to obtain a distorted signal carrying the building reflection characteristics. The specific steps include:

[0121] Step 401: Obtain the material type of the building environment and query the preset building material reflection coefficient database to obtain the corresponding reflection attenuation factor.

[0122] In this step, the environmental building material type refers to the physical category of the building surface material, which determines the electromagnetic wave reflection characteristics. The preset building material reflection coefficient database refers to a mapping table storing material types and attenuation factors, used to quantify reflection loss. The reflection attenuation factor refers to the material-related reflection efficiency coefficient.

[0123] In this embodiment of the application, the material type of the current building in the environment is obtained by image recognition or geographic information system, a preset building material reflectance coefficient database is queried, and the corresponding reflectance attenuation factor is output, such as 0.8 for concrete.

[0124] Step 402: The time delay quantization value corresponding to each reflection path in the multipath propagation parameters is weighted and calculated with the reflection attenuation factor to obtain the corresponding reflection path delay.

[0125] In this embodiment of the application, the time delay quantization value of each reflection path is extracted from the multipath propagation parameters, such as 20ns, and a weighted multiplication operation is performed on it with the reflection attenuation factor. The calculation formula is: reflection path delay = time delay quantization value × reflection attenuation factor. The reflection path delay of each path is output, such as 20ns × 0.8 = 16ns.

[0126] Step 403: Based on the preset multipath reflection order rule, the reflection path delay is converted into a time axis offset parameter sequence.

[0127] In this step, the preset multipath reflection order rule refers to the rule that defines the number of building reflections and controls the depth of the multipath reflection simulation. The time axis offset parameter sequence refers to the set of time delays of the multipath reflections, which is used to segment and delay signal replicas.

[0128] In this embodiment of the application, based on the preset multipath reflection order rule, the reflection path delay is converted into a time axis offset parameter sequence, such as (16ns, 32ns, 48ns), which represents the time delay of multiple reflections.

[0129] Step 404: Perform full waveform copying on the source domain baseband signal to obtain a lossless signal copy.

[0130] In this step, the lossless signal copy refers to a complete replica of the source domain baseband signal, preserving the original waveform and modulation information.

[0131] In this embodiment, a deep copy operation is performed on the source domain baseband signal to generate a lossless signal copy with completely identical waveform data.

[0132] Step 405: Based on the time axis offset parameter sequence, perform a time delay operation on the lossless signal copy to obtain a time offset signal.

[0133] In this step, the time offset signal refers to the output after applying a segmented delay to the signal copy, simulating the multiple reflection effect of the building complex.

[0134] In this embodiment, a segmented time-shift operation is performed on the lossless signal copy according to the time axis offset parameter sequence, and a time-shifted signal carrying multiple reflection effects is output.

[0135] Step 406: Superimpose the time offset signal and the source domain baseband signal to obtain a distorted signal carrying the building reflection characteristics.

[0136] In this embodiment, the amplitude of the time offset signal and the source domain baseband signal at the same time point are added together to output a synthesized distorted signal carrying the building reflection characteristics.

[0137] This application's embodiments calculate accurate path delay using material-aware reflection attenuation factors, dynamically generate multiple reflection sequences based on building height, and achieve physical-level modeling of building reflection effects through segmented time shifts and signal superposition, providing distorted signal samples in real urban environments for cross-domain transfer learning.

[0138] This application provides a specific embodiment. Step 105 involves adding random noise to the distorted signal according to the noise distribution parameters to obtain a target domain analog signal. This specifically includes the following steps:

[0139] Step 501: Based on the preset intensity amplitude conversion coefficient, linearly scale the noise distribution parameters to obtain the noise amplitude reference value.

[0140] In this step, the preset intensity amplitude conversion coefficient defines the proportional relationship between noise intensity in decibels and voltage amplitude, used for physical quantity conversion. The noise amplitude reference value refers to the median of the voltage amplitude generated by linear scaling of the noise distribution parameters, serving as the reference for the center of noise fluctuation.

[0141] In this embodiment of the application, a linear scaling operation is performed on the noise distribution parameter using a preset intensity amplitude conversion coefficient. The calculation formula is: amplitude reference value = noise distribution parameter × preset intensity amplitude conversion coefficient, to obtain the noise amplitude reference value; for example, if the preset intensity amplitude conversion coefficient is 0.05V / dB and the reference noise intensity value is 10dB, the noise amplitude reference value = 0.05V / dB × 10dB = 0.5V.

[0142] Step 502: Calculate the floating boundary value of the noise amplitude reference value according to the preset floating ratio parameter, and perform a randomization generation operation of the noise amplitude value within the range of the floating boundary value to generate a random noise sequence.

[0143] In this step, the preset floating ratio parameter refers to the percentage parameter that controls the noise fluctuation range, determining the offset of the boundary value relative to the reference value. The floating boundary value refers to the interval formed by the minimum and maximum allowable noise amplitude, constraining the range of random noise generation. The random noise sequence refers to the set of discrete amplitudes generated by uniform sampling within the floating boundary value, characterizing the instantaneous characteristics of environmental noise.

[0144] In this embodiment, based on a preset floating ratio parameter, positive and negative boundary calculations are performed on the noise amplitude reference value to output floating boundary values. Within this range, a uniformly distributed random sampling operation is performed to generate a random noise sequence. For example, if the preset floating ratio parameter is ±20%, the noise amplitude reference value is 0.5V, the lower limit of the noise amplitude reference value is 0.5V×(1-0.2), and the upper limit of the noise amplitude reference value is 0.5V×(1+0.2), the floating boundary value is (0.4V, 0.6V). Within this range, a uniformly distributed random sampling operation is performed to generate a random noise sequence (0.42V, 0.58V, 0.39V).

[0145] Step 503: Adjust the length of the random noise sequence according to the number of sampling points of the distorted signal to obtain a length-aligned random noise sequence.

[0146] In this step, the number of sampling points refers to the total number of sampling points of the distorted signal in the time domain, which determines the target for adjusting the length of the noise sequence.

[0147] In this embodiment, the number of sampling points of the distorted signal is obtained, a linear interpolation operation is performed on the random noise sequence to make its length consistent with the number of sampling points, and a length-aligned random noise sequence is output.

[0148] Step 504: Time-align the length-aligned random noise sequence and the distorted signal to obtain a synchronization noise signal.

[0149] In this step, the synchronization noise signal refers to the noise waveform that is strictly aligned with the timestamp of the distortion signal to ensure that the noise changes in sync with the signal.

[0150] In this embodiment, each data point of the length-aligned random noise sequence is bound to the sampling time corresponding to the distorted signal. For example, the 5th noise value is matched with the 5th signal timestamp to obtain a time-synchronized noise signal.

[0151] Step 505: Superimpose the synchronization noise signal and the distortion signal to obtain the target domain analog signal.

[0152] In this embodiment, the amplitude values ​​of the synchronous noise signal and the distortion signal at the same time are arithmetically added. The calculation formula is: target signal amplitude = distortion signal amplitude + noise signal amplitude, and the superimposed target domain analog signal is output.

[0153] This application's embodiments map noise intensity to a voltage amplitude reference through physical dimension conversion, generate dynamic noise boundaries based on floating ratios, and use a dual alignment mechanism to achieve spatiotemporal synchronization between noise sequences and signals. Finally, through amplitude superposition, it accurately simulates the signal pollution effect in a real electromagnetic environment.

[0154] This application provides a specific embodiment. Step 106 involves processing the source domain baseband signal and the target domain analog signal using the physical environment constraints to obtain the modulation recognition result of the target domain analog signal. This specifically includes the following steps:

[0155] Step 601: Extract multiple source domain feature values ​​from the segmented data of the source domain baseband signal, extract multiple target domain feature values ​​from the segmented data of the target domain analog signal, combine all source domain feature values ​​to obtain a set of source domain feature vectors, and combine all target domain feature values ​​to obtain a set of target domain feature vectors.

[0156] In this step, multiple source domain feature values ​​refer to the physical layer parameters extracted from segmented baseband signals in the source domain, reflecting the time-frequency domain characteristics of the signal. Multiple target domain feature values ​​refer to the distortion features extracted from segmented analog signals in the target domain, characterizing the signal state under environmental interference. The source domain feature vector set refers to the structured feature group storing the source domain signal features. The target domain feature vector set refers to the structured storage of the target domain signal features, including environmental distortion information.

[0157] In this embodiment, a piecewise short-time Fourier transform operation is performed on the source domain baseband signal and the target domain analog signal to extract the spectral centroid, bandwidth and zero-crossing rate of each signal segment as feature values. The source domain feature values ​​are combined into a source domain feature vector set, and the target domain feature values ​​are combined into a target domain feature vector set.

[0158] Step 602: Based on the time delay boundary threshold in the physical environment constraints, calculate the maximum allowable distribution offset between the source domain feature vector set and the target domain feature vector set.

[0159] In this step, the maximum permissible distribution offset refers to the upper limit of the feature space distance derived from the time delay boundary threshold.

[0160] In this embodiment of the application, the maximum allowable distribution offset is obtained by multiplying the time delay boundary threshold in the physical environment constraints by a preset environment adaptation coefficient.

[0161] Step 603: Using the maximum allowable distribution offset and the preset scaling factor, calculate the characteristic value fluctuation tolerance range of the target domain analog signal. Within the characteristic value fluctuation tolerance range, adjust the target domain feature vector set to obtain the adjusted target domain feature vector set.

[0162] In this step, the preset scaling factor refers to the fluctuation range scaling factor. The eigenvalue fluctuation tolerance range refers to the allowable value range for each feature dimension. The adjusted target domain feature vector set refers to the feature vector group after boundary truncation to eliminate outliers caused by environmental interference.

[0163] In this embodiment, the maximum allowable distribution offset is multiplied by a preset proportional coefficient to generate the feature value fluctuation tolerance radius, the statistical mean of each dimension of the target domain feature vector set is calculated, the feature value fluctuation tolerance range is constructed with the mean as the center, the feature value fluctuation tolerance range is truncated for feature values ​​that exceed the feature value fluctuation tolerance range, and the adjusted target domain feature vector set is output.

[0164] Step 604: Perform modulation pattern matching on the adjusted target domain feature vector set to obtain the modulation recognition result of the target domain analog signal.

[0165] In this embodiment, the cosine similarity between the adjusted target domain feature vector set and the pre-stored modulation template features is calculated, and the modulation mode corresponding to the highest similarity is taken as the modulation recognition result of the target domain analog signal.

[0166] This application embodiment uses a physical constraint-driven feature distribution alignment mechanism to quantize the time delay boundary threshold into feature offset constraints, corrects the abnormal values ​​of target domain features within the fluctuation tolerance range, and finally achieves robust modulation recognition of radar signals in complex electromagnetic environments through template matching.

[0167] This application provides a specific embodiment. Step 603 involves calculating the eigenvalue fluctuation tolerance range of the target domain analog signal using the maximum permissible distribution offset and a preset scaling factor. Within the eigenvalue fluctuation tolerance range, the target domain feature vector set is adjusted to obtain the adjusted target domain feature vector set. This specifically includes the following steps:

[0168] Step 611: Multiply the maximum allowable distribution offset and the preset proportional coefficient to obtain the characteristic fluctuation tolerance radius of the target domain.

[0169] In this step, the characteristic fluctuation tolerance radius refers to a scalar value generated by multiplying the maximum allowable distribution offset by a scaling factor, which controls the radius of the characteristic value fluctuation range.

[0170] In this embodiment, the maximum allowable distribution offset is multiplied by a preset proportional coefficient. The calculation formula is: tolerance radius = maximum allowable distribution offset × preset proportional coefficient, to obtain the characteristic fluctuation tolerance radius of the target domain.

[0171] Step 612: Perform feature dimension decomposition on the target domain feature vector set to obtain numerical sequences of different feature dimensions, and calculate the average value of each numerical sequence, and use the average value as the center value of the corresponding feature dimension.

[0172] In this step, the numerical sequences of different feature dimensions refer to data sets with a single feature dimension. The central value is the arithmetic mean of the index value sequences, representing the central location of the feature distribution.

[0173] In this embodiment of the application, a feature dimension decomposition operation is performed on the target domain feature vector set to separate the numerical sequence of each feature dimension; an average value is calculated for the numerical sequence of each dimension, and the average value is used as the center value of the corresponding feature dimension.

[0174] Step 613: Based on the characteristic fluctuation tolerance radius, calculate the upper and lower tolerance limits of the center value of each characteristic dimension, and combine the upper and lower tolerance limits of all characteristic dimensions to obtain the characteristic value fluctuation tolerance range of the target domain analog signal.

[0175] In this step, the upper tolerance limit refers to the boundary between the center value and the tolerance radius, allowing for the maximum characteristic fluctuation. The lower tolerance limit refers to the boundary between the center value and the tolerance radius, allowing for the minimum characteristic fluctuation.

[0176] In this embodiment, positive and negative boundary calculation operations are performed on the center values ​​of each dimension based on the characteristic fluctuation tolerance radius. The calculation formula is: upper limit value = center value + tolerance radius, lower limit value = center value - tolerance radius. The upper and lower limit values ​​of all dimensions are combined into a structured boundary group to obtain the characteristic value fluctuation tolerance range of the target domain analog signal.

[0177] Step 614: Within the tolerance range of the eigenvalue fluctuation, perform boundary constraint processing on each numerical sequence to obtain multiple boundary-constrained numerical sequences.

[0178] In this step, the numerical sequence after boundary constraints refers to the data sequence that has been truncated by upper and lower limits.

[0179] In this embodiment of the application, within the tolerance range of the feature value fluctuation, for the numerical sequence of each feature dimension, the data points of the numerical sequence are compared with the upper and lower tolerance limits of the corresponding dimension. If the data points are lower than the lower limit, they are replaced with the lower limit value; if the data points are higher than the upper limit, they are replaced with the upper limit value; if the data points are within the range, the original values ​​are retained, thus obtaining a numerical sequence with multiple boundary constraints.

[0180] Step 615: Reconstruct all the numerical sequences after boundary constraints into vectors to generate an adjusted set of target domain feature vectors.

[0181] In this embodiment of the application, all numerical sequences after boundary constraints are reorganized according to the structure of the feature dimension to obtain the adjusted target domain feature vector set.

[0182] This application embodiment uses a dimensional boundary constraint mechanism to transform the tolerance radius derived from the physical environment into the dynamic fluctuation range of each feature dimension. Based on the center value, abnormal features are truncated and corrected, and finally a set of feature vectors that conforms to physical laws is reconstructed, thereby improving the environmental robustness of cross-domain recognition.

[0183] Figure 2 This application provides a schematic diagram of the structure of a radar signal modulation recognition system driven by cross-domain transfer learning, as shown in the embodiment of the present application. Figure 2 As shown, the system includes:

[0184] The acquisition module 21 is used to acquire the set of time delay characteristic values ​​of the building reflection path and the set of background noise intensity change values.

[0185] The construction module 22 is used to generate multipath propagation parameters and noise distribution parameters based on the set of time delay feature values ​​and the set of background noise intensity change values, and to perform boundary processing on the multipath propagation parameters and the noise distribution parameters according to preset physical environment boundary rules to construct physical environment constraints, the physical environment constraints including time delay boundary thresholds and noise boundary thresholds.

[0186] The reconstruction module 23 is used to reconstruct the baseband signal based on the pre-stored source domain modulation sample data to obtain the source domain baseband signal.

[0187] The calculation module 24 is used to calculate the corresponding reflection path delay based on the time delay quantization value in the multipath propagation parameters, and to superimpose the reflection path delay onto the source domain baseband signal to obtain a distorted signal carrying the building reflection characteristics.

[0188] Adding module 25 is used to add random noise to the distorted signal according to the noise distribution parameters to obtain the target domain analog signal.

[0189] The processing module 26 uses the physical environment constraints to process the source domain baseband signal and the target domain analog signal to obtain the modulation recognition result of the target domain analog signal.

[0190] Figure 2 The aforementioned cross-domain transfer learning-driven radar signal modulation recognition system can perform... Figure 1 The implementation principle and technical effects of the radar signal modulation recognition method driven by cross-domain transfer learning described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit performs its operations in the radar signal modulation recognition system driven by cross-domain transfer learning in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0191] In one possible design, Figure 2 The radar signal modulation recognition system driven by cross-domain transfer learning in the illustrated embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32.

[0192] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0193] The processing component 32 is used for the above Figure 1 The embodiment describes a radar signal modulation recognition method driven by cross-domain transfer learning.

[0194] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0195] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0196] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0197] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0198] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0199] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0200] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a radar signal modulation recognition method driven by cross-domain transfer learning.

[0201] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0202] The device embodiments described above are merely illustrative. 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 network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0203] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0204] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A radar signal modulation recognition method driven by cross-domain transfer learning, characterized in that, The method comprises the following steps: acquiring a set of time delay characteristic values of a building reflection path and a set of background noise intensity variation values; based on the set of time delay characteristic values and the set of background noise intensity variation values, generating multipath propagation parameters and noise distribution parameters, and performing boundary processing on the multipath propagation parameters and the noise distribution parameters according to a preset physical environment boundary rule to construct a physical environment constraint condition, wherein the physical environment constraint condition comprises a time delay boundary threshold and a noise boundary threshold; based on pre-stored source domain modulation sample data, reconstructing a baseband signal to obtain a source domain baseband signal; according to a time delay quantization value in the multipath propagation parameters, calculating a corresponding reflection path delay, and superimposing the reflection path delay on the source domain baseband signal to obtain a distortion signal carrying building reflection characteristics; according to the noise distribution parameters, adding random noise to the distortion signal to obtain a target domain simulation signal; using the physical environment constraint condition, processing the source domain baseband signal and the target domain simulation signal to obtain a modulation recognition result of the target domain simulation signal.

2. The method of claim 1, wherein, Based on pre-stored source domain modulation sample data, reconstructing a baseband signal to obtain a source domain baseband signal, comprising: extracting modulation type parameters, symbol mapping parameters and pulse shape parameters from the pre-stored source domain modulation sample data; determining the modulation order according to the modulation type parameters, to determine the bit length corresponding to each modulation symbol based on the modulation order; based on the bit length, generating an original bit sequence and dividing the original bit sequence into multiple bit groups to generate a discrete bit sequence; according to the symbol mapping parameters, assigning a corresponding discrete modulation symbol to each bit group to generate a discrete modulation symbol sequence; according to the pulse shape parameters, pulse shaping filtering the discrete modulation symbol sequence to obtain a baseband waveform; calculating the carrier frequency offset of the baseband waveform to compensate and correct the baseband waveform according to the carrier frequency offset to obtain a compensated baseband signal, and taking the compensated baseband signal as the source domain baseband signal.

3. The method of claim 2, wherein, Calculating the carrier frequency offset of the baseband waveform to compensate and correct the baseband waveform according to the carrier frequency offset to obtain a compensated baseband signal, and taking the compensated baseband signal as the source domain baseband signal, comprising: based on the pulse shape parameters, calculating the spectral center frequency and taking the spectral center frequency as the reference carrier frequency value; extracting the carrier frequency value from the baseband waveform, combining the reference carrier frequency value to calculate the frequency compensation factor; based on the frequency compensation factor and the sampling time sequence of the baseband waveform, calculating the phase rotation angle of each sampling time in the baseband waveform to multiply each complex sampling value of the baseband waveform by the corresponding phase rotation angle to obtain a frequency offset correction signal; compensating and correcting the frequency offset correction signal to generate a compensated baseband signal, and taking the compensated baseband signal as the source domain baseband signal.

4. The method of claim 1, wherein, According to the delay quantization value in the multipath propagation parameter, a corresponding reflection path delay is calculated, the reflection path delay is superimposed on the source domain baseband signal to obtain a distortion signal carrying building reflection characteristics, including: Obtain the environmental building material type, and query the preset building material reflection coefficient database to obtain the corresponding reflection attenuation factor; The time delay quantization value corresponding to each reflection path in the multipath propagation parameter is weighted and calculated with the reflection attenuation factor respectively to obtain the corresponding reflection path delay; Based on the preset multipath reflection order rule, the reflection path delay is converted into a time axis offset parameter sequence; The source domain baseband signal is full waveform copied to obtain a lossless signal copy; Based on the time axis offset parameter sequence, a time delay operation is performed on the lossless signal copy to obtain a time offset signal; The time offset signal and the source domain baseband signal are superimposed to obtain a distortion signal carrying building reflection characteristics.

5. The method of claim 1, wherein, According to the noise distribution parameter, random noise is added to the distortion signal to obtain a target domain simulation signal, including: According to the preset intensity amplitude conversion coefficient, the noise distribution parameter is linearly scaled to obtain a noise amplitude reference value; According to the preset floating proportion parameter, a floating boundary value of the noise amplitude reference value is calculated, and within the range of the floating boundary value, a noise amplitude value randomization generation operation is performed to generate a random noise sequence; According to the number of sampling points of the distortion signal, the length of the random noise sequence is adjusted to obtain a length-aligned random noise sequence; The length-aligned random noise sequence and the distortion signal are time-aligned to obtain a synchronous noise signal; The synchronous noise signal and the distortion signal are superimposed to obtain a target domain simulation signal.

6. The method of claim 1, wherein, The source domain baseband signal and the target domain simulation signal are processed using the physical environment constraint condition to obtain a modulation recognition result of the target domain simulation signal, including: Extract a plurality of source domain characteristic values from the segmented data of the source domain baseband signal, extract a plurality of target domain characteristic values from the segmented data of the target domain simulation signal, and combine all source domain characteristic values to obtain a source domain characteristic vector set, and combine all target domain characteristic values to obtain a target domain characteristic vector set; Based on the time delay boundary threshold in the physical environment constraint condition, calculate the maximum allowed distribution offset between the source domain characteristic vector set and the target domain characteristic vector set; Using the maximum allowed distribution offset and a preset proportion coefficient, calculate the feature value fluctuation tolerance range of the target domain simulation signal, adjust the target domain characteristic vector set within the feature value fluctuation tolerance range to obtain an adjusted target domain characteristic vector set; Perform modulation mode matching on the adjusted target domain characteristic vector set to obtain the modulation recognition result of the target domain simulation signal.

7. The method of claim 6, wherein, The maximum allowed distribution offset and a preset proportion coefficient are used to calculate a feature value fluctuation tolerance range of the target domain simulation signal. Within the feature value fluctuation tolerance range, the target domain feature vector set is adjusted to obtain an adjusted target domain feature vector set, including: The maximum allowed distribution offset and a preset proportion coefficient are multiplied to obtain a feature fluctuation tolerance radius of the target domain; The target domain feature vector set is subjected to feature dimension decomposition to obtain numerical sequences of different feature dimensions, and the average value of each numerical sequence is calculated as the center value of the corresponding feature dimension; According to the feature fluctuation tolerance radius, the tolerance upper limit value and the tolerance lower limit value of the center value of each feature dimension are calculated, and the tolerance upper limit values and the tolerance lower limit values of all feature dimensions are combined to obtain the feature value fluctuation tolerance range of the target domain simulation signal; Within the feature value fluctuation tolerance range, each numerical sequence is subjected to boundary constraint processing to obtain a plurality of boundary-constrained numerical sequences; All boundary-constrained numerical sequences are subjected to vector reconstruction to generate an adjusted target domain feature vector set. 8.A radar signal modulation recognition system driven by cross-domain transfer learning, characterized in that, Comprising: The acquisition module is configured to acquire a time delay feature value set of a building reflection path and a background noise intensity change value set; The construction module is configured to generate multi-path propagation parameters and noise distribution parameters based on the time delay feature value set and the background noise intensity change value set, and perform boundary processing on the multi-path propagation parameters and the noise distribution parameters according to a preset physical environment boundary rule to construct a physical environment constraint condition, wherein the physical environment constraint condition includes a time delay boundary threshold and a noise boundary threshold; The reconstruction module is configured to reconstruct a baseband signal based on pre-stored source domain modulation sample data to obtain a source domain baseband signal; The calculation module is configured to calculate a corresponding reflection path delay based on a time delay quantization value in the multi-path propagation parameters, and add the reflection path delay to the source domain baseband signal to obtain a distortion signal carrying building reflection characteristics; The addition module is configured to add random noise to the distortion signal according to the noise distribution parameters to obtain a target domain simulation signal; The processing module is configured to process the source domain baseband signal and the target domain simulation signal based on the physical environment constraint condition to obtain a modulation recognition result of the target domain simulation signal.

9. A computing device, comprising: The processing component and the storage component are included; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component, realizing the radar signal modulation recognition method driven by cross-domain transfer learning as claimed in any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer program is stored in the computer, and when the computer program is executed by the computer, the radar signal modulation recognition method driven by cross-domain transfer learning as claimed in any one of claims 1-7 is realized.

Images