A radar main lobe composite jamming suppression method, device and electronic equipment based on multi-displacement features

Abstract

The application provides a radar main lobe receiving signal interference suppression method and device based on multiple displacement features and an electronic device. The method comprises the following steps: sequencing a radar main lobe receiving signal into a complex receiving signal sequence, and processing a real part and an imaginary part of the complex receiving signal sequence into an initial complex feature after channel dimension splicing; constructing a position coding complex feature according to a sequence position of the complex receiving signal sequence, constructing a position enhanced complex feature based on the position coding complex feature and the initial complex feature; performing down-sampling on the position enhanced complex feature, setting multiple reference displacements, respectively shifting the low-resolution complex feature according to the reference displacements, generating multiple reference displacement features, performing weighted fusion, constructing a complex enhanced feature based on a fusion result and the low-resolution complex feature, and inputting the complex enhanced feature into a complex multi-scale U-Net network to obtain a multi-scale complex feature; performing output mapping and scale adjustment on the multi-scale complex feature to generate a radar main lobe complex echo result after interference suppression.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and specifically to a method, apparatus, and electronic device for suppressing radar main lobe composite interference based on multi-displacement characteristics. Background Technology

[0002] Radar mainlobe composite interference suppression is a key technology for radar detection and target localization. It aims to suppress two or more active interference signals simultaneously present in the received signal from the radar main lobe, while preserving as much of the amplitude and phase information of the true target echo signal as possible. For radar complex signals represented in I / Q form, the real and imaginary parts jointly carry the amplitude and phase information of the echo, and there is a clear complex coupling relationship between them. Therefore, in the process of interference suppression, how to preserve the complex signal information while suppressing interference is a crucial factor affecting subsequent functions such as signal detection, range estimation, and target tracking.

[0003] Traditional radar active jamming suppression methods typically rely on filtering, frequency domain suppression, parameter estimation, threshold decision, or signal processing strategies based on prior models. These methods, on the one hand, only consider the suppression of a single type of jamming, and on the other hand, are only applicable under conditions where the jamming model is well-defined and the echo environment is relatively stable. However, in real-world complex electromagnetic environments, the radar main lobe may be subjected to compound jamming from multiple active jamming sources (such as the superposition of suppressive and deceptive jamming), and the compounding forms are rapidly changing. In these complex jamming environments, fixed rules or single prior models often struggle to simultaneously achieve jamming suppression and the preservation of the true target.

[0004] In recent years, deep learning methods have been increasingly used for radar real target echo enhancement and interference suppression tasks. Models based on convolutional neural networks, residual networks, or attention mechanisms can learn the mapping relationship between interference-containing signals and real target echoes from training data, reducing reliance on manual rules and precise interference parameter estimation. Compared to traditional methods, deep models have stronger feature representation capabilities and can automatically model local textures, peak shapes, and multi-scale structures in complex received signals.

[0005] However, existing deep learning-based radar jamming suppression methods still have several shortcomings. For example, some methods simply treat the real and imaginary parts of the radar complex signal as ordinary real-valued channels and splice them together, without explicitly modeling the coupling between the real and imaginary parts, which weakens the amplitude-phase consistency representation. Traditional global self-attention structures have high computational and memory overhead on one-dimensional long sequences, which is not conducive to engineering deployment, while relying solely on ordinary convolutions is insufficient to fully detect broadband jamming and delayed forwarding spoofing responses. Summary of the Invention

[0006] In view of this, the present invention proposes a radar main lobe composite interference suppression method, device, and electronic equipment based on multi-displacement characteristics to address the problem of effectively extracting real target signals in radar composite interference scenarios, and to improve the radar's effective suppression of interference and effective extraction of real target echoes under additive composite interference conditions. Specifically, the present invention is achieved through the following technical solution:

[0007] According to a first aspect of the embodiments of this specification, a radar main lobe composite interference suppression method based on multi-displacement characteristics is provided, the method comprising the following steps:

[0008] Step S1: Serialize the radar main lobe received signal to be processed into a complex signal sequence, and then concatenate the real and imaginary parts of the complex signal sequence according to the channel dimension to obtain the initial complex feature.

[0009] Step S2: Construct normalized position variables based on the sequence positions of the complex signal sequence; generate position codes based on the normalized position variables; set the position codes to zero as real and imaginary parts to form position code complex features; concatenate the position code complex features with the initial complex features in the channel dimension; perform mapping processing and complex feature interaction on the concatenated complex features to obtain position-enhanced complex features.

[0010] Step S3: Downsample the position-enhanced complex features to obtain low-resolution complex features; set multiple reference displacements based on the distribution range of the real target signal and interference signal in the radar received signal, and translate the low-resolution complex features according to each reference displacement to generate multiple reference displacement features; perform weighted fusion on the reference displacement features, and perform residual-gated fusion with the weighted fused multi-displacement complex features and the low-resolution complex features to obtain enhanced low-resolution features; upsample the enhanced low-resolution features to the scale of the position-enhanced complex features, and superimpose and fuse them with the position-enhanced complex features to obtain complex enhanced features;

[0011] Step S4: Input the complex enhanced features into the complex multi-scale U-Net backbone network for multi-scale encoding, bottleneck layer feature extraction and decoding recovery to obtain multi-scale complex features;

[0012] Step S5: Perform output mapping and scale adjustment on the multi-scale complex representation to generate a complex signal after radar main lobe interference suppression.

[0013] According to a second aspect of the embodiments of this specification, a radar main lobe composite interference suppression device based on multi-displacement characteristics is provided, the device comprising:

[0014] The serialization unit is used to serialize the radar main lobe received signal to be processed into a complex signal sequence, and to process the real part and imaginary part of the complex signal sequence into an initial complex feature by concatenating them along the channel dimension.

[0015] The position enhancement unit is used to construct a normalized position variable based on the sequence position of the complex signal sequence, generate a position code based on the normalized position variable, set the position code as the real part and the imaginary part to zero to form a position code complex feature, concatenate the position code complex feature with the initial complex feature in the channel dimension, and perform mapping processing and interaction on the concatenated complex feature to obtain the position-enhanced complex feature.

[0016] A displacement fusion unit is used to downsample the position-enhanced complex features to obtain low-resolution complex features; set multiple reference displacements according to the distribution range of the true target signal and the interference signal in the radar received signal, and translate the low-resolution complex features according to each reference displacement to generate multiple reference displacement features; perform weighted fusion on the reference displacement features, and perform residual-gated fusion on the weighted fused multi-displacement complex features and the low-resolution complex features to obtain enhanced low-resolution features; upsample the enhanced low-resolution features to the scale of the position-enhanced complex features, and superimpose and fuse them with the position-enhanced complex features to obtain complex enhanced features;

[0017] The interference suppression unit is used to input the complex enhanced features into the complex multi-scale U-Net backbone network for multi-scale encoding, bottleneck layer feature extraction and decoding recovery to obtain multi-scale complex features;

[0018] The true target signal recovery unit is used to perform output mapping and scale adjustment on the multi-scale complex representation to generate a complex signal after radar main lobe interference suppression.

[0019] According to a third aspect of the embodiments of this specification, an electronic device is provided, comprising: a processor; and a computer-readable storage medium storing computer program instructions that, when executed by the processor, cause the processor to perform the method as described in the first aspect.

[0020] The embodiments of the present invention have at least the following technical effects:

[0021] (1) The present invention uses complex convolution to extract features from the radar main lobe received signal, so that the real part and the imaginary part are coupled and modeled according to the complex multiplication relationship. Compared with the method of simply treating the real part and the imaginary part as ordinary real value channels, the present invention can better preserve the amplitude and phase structure information in the radar received signal.

[0022] (2) The present invention introduces the linear position, sine position and cosine position as complex position encoding channels with the imaginary part set to zero into the complex feature representation, so that the sampling position relationship can be explicitly used when the radar main lobe received signal is serialized, which helps to enhance the perception of target position, local offset structure and interference response pattern.

[0023] (3) The present invention scores and weights the reference displacement relationship in a lower resolution space, and can model structural relationships such as local misalignment, delayed replication, multi-peak response and wide-peak interference with a small computational overhead. The multi-displacement relationship information is introduced into the original resolution features through upsampling and backfeeding. Attached Figure Description

[0024] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Some specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings in an exemplary and non-limiting manner. The same reference numerals in the drawings indicate the same or similar parts or components. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0025] Figure 1 This is a flowchart illustrating an exemplary embodiment of the present invention of a radar main lobe composite interference suppression method based on multi-displacement features;

[0026] Figure 2 This is a schematic diagram illustrating a process for constructing position-enhanced complex features according to an exemplary embodiment of the present invention;

[0027] Figure 3 This is a schematic diagram illustrating a process for constructing multi-displacement relationship enhancement features according to an exemplary embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram illustrating the structure of a complex gated temporal attention unit according to an exemplary embodiment of the present invention;

[0029] Figure 5 This is a schematic diagram of the structure of a four-layer complex multi-scale U-Net backbone network shown in an exemplary embodiment of the present invention;

[0030] Figure 6 This is a block diagram illustrating an electronic device according to an exemplary embodiment of the present invention;

[0031] Figure 7 This is a block diagram illustrating a radar main lobe composite interference suppression device based on multi-displacement characteristics, as shown in an exemplary embodiment of the present invention. Detailed Implementation

[0032] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0033] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0034] The following detailed explanation of each step is provided in conjunction with the accompanying drawings.

[0035] Figure 1 This is a flowchart illustrating an exemplary embodiment of the present invention of a radar main lobe composite interference suppression method based on multi-displacement features, as shown below. Figure 1 As shown, the radar main lobe composite interference suppression method includes the following steps:

[0036] Step S1: Serialize the radar main lobe received signal to be processed into a complex signal sequence, and then concatenate the real and imaginary parts of the complex signal sequence according to the channel dimension to obtain the initial complex feature.

[0037] Step S2: Construct normalized position variables based on the sequence positions of the complex signal sequence; generate position codes based on the normalized position variables; set the position codes to zero as real and imaginary parts to form position code complex features; concatenate the position code complex features with the initial complex features in the channel dimension; perform mapping processing and complex feature interaction on the concatenated complex features to obtain position-enhanced complex features.

[0038] Step S3: Downsample the position-enhanced complex features to obtain low-resolution complex features; set multiple reference displacements based on the distribution range of the real target signal and interference signal in the radar received signal, and translate the low-resolution complex features according to each reference displacement to generate multiple reference displacement features; perform weighted fusion on the reference displacement features, and perform residual-gated fusion with the weighted fused multi-displacement complex features and the low-resolution complex features to obtain enhanced low-resolution features; upsample the enhanced low-resolution features to the scale of the position-enhanced complex features, and superimpose and fuse them with the position-enhanced complex features to obtain complex enhanced features;

[0039] Step S4: Input the complex enhanced features into the complex multi-scale U-Net backbone network for multi-scale encoding, bottleneck modeling and decoding recovery to obtain multi-scale complex features;

[0040] Step S5: Perform output mapping and scale adjustment on the multi-scale complex representation to generate a complex signal after radar main lobe interference suppression.

[0041] Figure 1 The radar main lobe composite interference suppression method shown above, through the above steps S1 to S5, sequentially performs complex feature extraction, sequence position information embedding, multi-reference displacement feature fusion, multi-scale complex feature encoding and decoding, and complex received signal reconstruction on the radar main lobe complex received signal. It can effectively suppress interference components in the composite scenario of suppression interference and deception interference, while maintaining the amplitude and phase structure and peak position of the real target echo, providing high-fidelity complex echo input for subsequent target detection, range estimation, and Doppler analysis.

[0042] Next, combined Figures 2 to 5 The radar main lobe composite interference suppression scheme of the present invention is described in detail.

[0043] The radar main lobe composite interference suppression method provided by this invention can be deployed in one or more signal processing devices within a radar system. These signal processing devices include, but are not limited to, general-purpose computers, embedded signal processing platforms, digital signal processors (DSPs), field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), or combinations thereof. Each signal processing device includes at least a processor and a memory, the memory storing computer program instructions. When the processor executes these computer program instructions, the signal processing device performs the method described in this invention.

[0044] refer to Figure 2 The radar main lobe composite interference suppression process of this system includes: the feature construction process of initial complex features, position-enhanced complex features, and complex enhancement features, as well as the feature extraction process based on the complex multi-scale U-Net backbone network.

[0045] Regarding the construction of initial complex features:

[0046] In some embodiments, the initial complex features are constructed through the following steps:

[0047] The real and imaginary parts of the complex received signal sequence are used as two input channels, and they are spliced ​​along the channel dimension to form an input tensor;

[0048] The input tensor is normalized, and the normalized input tensor is then subjected to numerical safety processing, replacing non-numerical, positive infinity, and negative infinity values ​​with finite values ​​within a preset numerical range to obtain the safety-processed features.

[0049] The security-processed features are mapped to multi-channel complex features through complex one-dimensional convolution, and the multi-channel complex features are normalized and non-linearly activated to obtain initial complex features. The complex one-dimensional convolution is implemented by combining real part convolution kernels and imaginary part convolution kernels, so that the real part and imaginary part of the input features interact according to the complex multiplication relationship.

[0050] The complex signal sequence in this embodiment is:

[0051] (1)

[0052] in, Indicates the first The complex signal value at each sampling point Represents the real part of the complex signal. Represents the imaginary part of a complex signal. Represents the imaginary unit. Indicates the sampling point number.

[0053] The real and imaginary parts of the complex signal sequence are used as two input channels, and concatenated along the channel dimension to form an input tensor:

[0054] (2)

[0055] in, Indicates the input tensor. This represents the input channel composed of the real parts of the received complex signal. This represents the input channel composed of the imaginary part of the received complex signal. This indicates the batch size, and 2 represents the two channels: the real part and the imaginary part. Indicates the one-dimensional sampling length. Represents the real number field.

[0056] The input tensor is normalized or scaled to ensure that the input amplitudes of different samples are within a suitable range for network processing. The amplitude normalization can be expressed as:

[0057] (3)

[0058] in, This represents the normalized input tensor. This represents a stability constant used to avoid a denominator of zero.

[0059] In this embodiment, the normalized input tensor is used as the input to the subsequent radar main lobe composite interference suppression network, so that the network can simultaneously utilize the real part information, imaginary part information, and amplitude-phase relationship between the radar complex received signal.

[0060] In practical applications, the normalized input tensor may contain non-numerical values, positive infinity, and negative infinity. This embodiment further inputs the normalized input tensor to a numerical safety processing unit, performing finite value replacement on the non-numerical, positive, and negative infinity values ​​in the input data, and restricting the replaced data within a preset numerical range to obtain stable input characteristics. The numerical safety processing can be expressed as:

[0061] (4)

[0062] in, This represents the stable input characteristics after numerical security processing. This indicates the substitution operation for non-numeric and infinite values. This indicates a limit operation. Indicates a preset lower bound. This indicates a preset upper bound.

[0063] The stable input features are fed into the complex convolution initial mapping module, where a single complex input is mapped to multi-channel complex features via a one-dimensional complex convolution. The one-dimensional complex convolution is implemented using a combination of real and imaginary convolution kernels, allowing the real and imaginary parts of the input features to interact according to a complex multiplication relationship.

[0064] Let the input complex features of the complex convolution initial mapping module be... The complex convolution kernel is Then the complex convolution output satisfies:

[0065] (5)

[0066] in, and These represent the real and imaginary parts of the input complex number features, respectively. and Let these represent the real and imaginary parts of the complex convolution kernel, respectively. and represents the real and imaginary parts of the output complex features, respectively, and * represents a one-dimensional convolution operation.

[0067] The complex convolution output is then subjected to normalization and nonlinear activation processing to obtain the initial complex features:

[0068] (6)

[0069] in, Indicates the initial complex characteristics, This represents a one-dimensional convolution operation on complex numbers. This indicates normalization processing. This represents a non-linear activation function.

[0070] Regarding the construction of location-enhanced complex features:

[0071] After constructing the initial complex features, the initial complex features are input into the complex coordinate position encoding module to generate position-enhanced complex features.

[0072] Let the initial complex number characteristics be:

[0073] (7)

[0074] in, Indicates the initial complex characteristics, The real part of the initial complex number characteristic is represented by... It represents the imaginary part of the initial complex number characteristic.

[0075] like Figure 2 As shown, the complex coordinate position encoding module operates along the one-dimensional sampling direction of the complex received signal sequence and constructs normalized position variables based on the sequence position of the complex received signal sequence. For a one-dimensional complex received signal sequence of length T, the normalized position variable can be expressed as:

[0076] (8)

[0077] in, Indicates the first The normalized position value corresponding to each sampling point.

[0078] Based on the normalized position variables, construct linear position channels, sinusoidal position channels, and cosine position channels:

[0079] (9)

[0080] in, Represents linear positional encoding. Represents the sine position code. This represents the cosine position code.

[0081] By treating the position-encoded channels as real-part position features and setting their imaginary parts to zero, we obtain the position-encoded complex features. :

[0082] (10)

[0083] in, Represents positional encoding complex features The real part, Represents positional encoding complex features The imaginary part.

[0084] The initial complex features and the position-encoded complex features are concatenated along the complex channel dimension, and the position-enhanced complex features are obtained through complex pointwise convolution, normalization, and nonlinear activation:

[0085] (11)

[0086] in, Indicates position-enhanced complex features, This indicates a channel splicing operation.

[0087] Through the above processing, the complex coordinate position encoding module can introduce one-dimensional sampling position information into the complex feature representation without destroying the complex structure.

[0088] Regarding complex enhancement features:

[0089] The position-enhanced complex features are input into a low-resolution multi-displacement relationship mixing module to generate complex enhanced features containing multiple positional relationships. The low-resolution multi-displacement relationship mixing module is configured as follows:

[0090] The low-resolution complex features are normalized, and complex scores corresponding to each reference displacement are generated using complex score mapping.

[0091] The response score corresponding to each reference displacement is calculated based on the square of the amplitude of each complex score, and the response score corresponding to each reference displacement is normalized to obtain the competitive weight corresponding to each reference displacement.

[0092] Based on the competition weights corresponding to each reference displacement, the features of the multiple reference displacements are weighted and fused to obtain the multi-displacement fused complex features.

[0093] Specifically, the structure of the low-resolution multi-displacement relationship hybrid module is as follows: Figure 3 As shown, the specific processing steps are as follows:

[0094] The location-enhanced complex features are subjected to complex downsampling to obtain low-resolution complex features:

[0095] (12)

[0096] in, Representing low-resolution complex features, This indicates a complex downsampling operation.

[0097] On the one hand, multiple reference displacement features are generated through reference displacement branches.

[0098] Set a reference displacement set:

[0099] (13)

[0100] in, Represents the set of reference displacements. Indicates the first One reference displacement, Indicates the reference displacement quantity.

[0101] The reference displacement can be set according to the input sequence length, target peak width, interference offset range, or sampling resolution.

[0102] Based on the reference displacement set, perform cyclic displacement operations on the low-resolution complex features to generate multiple reference displacement features:

[0103] (14)

[0104] in, Indicates according to the reference displacement Candidate displacement features obtained after cyclic displacement. This indicates a cyclic displacement operation.

[0105] On the other hand, competitive weights corresponding to each reference displacement are generated through scoring branches.

[0106] This embodiment normalizes the low-resolution complex features and uses a complex scoring mapping layer to generate complex scores that correspond one-to-one with each candidate displacement.

[0107] Specifically, normalizing the low-resolution complex features yields:

[0108] (15)

[0109] in, This represents the normalized low-resolution complex features. This indicates a normalization operation.

[0110] If the normalized low-resolution complex features have Each complex channel, with a length of The reference displacement set contains A reference displacement, then through having A complex scoring mapping layer with multiple complex output channels will Each input complex channel is mapped as Plural scoring channels:

[0111] (16)

[0112] in, ; This represents a complex scoring mapping operation, which couples the real and imaginary parts of the input features according to the complex multiplication relationship. Indicates inclusion The scoring characteristics of a complex scoring channel, the first Complex scoring channels and reference displacement correspond.

[0113] In one implementation, the complex scoring mapping layer is implemented using complex pointwise convolution, which, through a complex one-dimensional convolution with a kernel size of 1, ... Each input complex channel is mapped as Multiple scoring channels.

[0114] For parameter displacement The corresponding complex score can be expressed as:

[0115] (17)

[0116] in, Indicates reference displacement The corresponding complex score, For reference displacement The real part of the corresponding complex score, For reference displacement The imaginary part of the corresponding complex score.

[0117] The candidate displacement response score is calculated based on the squared magnitude of the complex score sequence corresponding to each reference displacement. For the reference displacement... In the At each sampling location, the square of the magnitude of the complex score is:

[0118] (18)

[0119] The reference displacement is obtained by averaging the squared magnitudes of the complex score along the one-dimensional sampling direction. Corresponding response score:

[0120] (19)

[0121] in, Indicates reference displacement The corresponding response score The one-dimensional sampling length representing the low-resolution complex features. Indicates the sampling location index. This represents a constant used to improve numerical stability.

[0122] The response scores corresponding to each reference displacement are normalized to obtain the competition weights corresponding to each reference displacement:

[0123] (20)

[0124] in, Indicates reference displacement The corresponding competitive weights, This represents the rating scaling factor. Indicates the first The response score corresponding to each reference displacement.

[0125] Through the above normalization process, the competition weights of each reference displacement satisfy:

[0126] (twenty one)

[0127] After obtaining the competition weights and features of each reference displacement, the features of multiple reference displacements are weighted and fused according to the competition weights to obtain multi-displacement fused complex features. :

[0128] (twenty two)

[0129] The multi-displacement fused complex features are sequentially subjected to complex pointwise convolution, normalization, and nonlinear activation to obtain displacement relationship enhancement features. :

[0130] (twenty three)

[0131] The displacement relationship enhancement feature and the low-resolution complex feature are fused using residual gating to obtain the low-resolution enhancement feature. :

[0132] (twenty four)

[0133] in, The preset residual fusion coefficient.

[0134] The low-resolution enhanced features are upsampled and restored using complex convolution, and the restored features are then fed back into the location-enhanced complex features to obtain the complex enhanced features. :

[0135] (25)

[0136] in, This represents a complex upsampling recovery operation consisting of linear interpolation upsampling and complex convolution. This is the preset recharge coefficient.

[0137] Next, the complex augmented features are input into the complex multi-scale U-Net backbone network for multi-scale encoding, bottleneck feature modeling, decoding recovery, and skip connection fusion to obtain a multi-scale complex representation. The structure of the complex multi-scale U-Net backbone network is as follows: Figure 4 As shown. The specific steps are as follows:

[0138] Complex augmented features are used as input to the complex multi-scale U-Net backbone network:

[0139] (26)

[0140] in, This represents the initial input features of the complex multi-scale U-Net backbone network.

[0141] The input features are extracted and downsampled step by step using a multi-level complex encoder:

[0142] (27)

[0143] in, Indicates the first The encoded features output by the level encoder. Indicates the first Complex features after level downsampling Indicates the first Level complex number encoding unit, Indicates the first Level complex downsampling unit.

[0144] The last-level downsampled features are input into the bottleneck layer for low-resolution, high-channel complex feature modeling:

[0145] (28)

[0146] in, Representing the complex characteristics of the bottleneck, This represents the bottleneck feature processing unit.

[0147] Complex-gated temporal attention units can be set in both the complex encoding unit and the bottleneck feature processing unit, such as... Figure 5 As shown, the input features of the complex-gated temporal attention unit are After mixing complex depthwise dilated convolution and complex pointwise convolution, we get:

[0148] (29)

[0149] in, Indicates the characteristics of the mixed complex number. and These represent the real part and the imaginary part, respectively.

[0150] Gating weights are generated based on complex amplitude features:

[0151] (30)

[0152] in, Indicates the gating weight, and Represents a one-dimensional convolution mapping. This represents the Sigmoid function. This indicates the complex amplitude characteristic.

[0153] When central prior is enabled, normalized coordinates are constructed and central prior weights are generated:

[0154] (31)

[0155] in, Indicates the central prior weight. This represents a non-negative learnable parameter. express The centralized, normalized position value obtained after the centralization transformation ranges from -1 to 1. The final gating weights are:

[0156] (32)

[0157] in, This represents the prior gating weights of the fusion center. This indicates element-wise multiplication.

[0158] Thus, the output of the complex-gated temporal attention unit is:

[0159] (33)

[0160] in, This represents the complex features after attention enhancement. This represents the learnable residual proportion parameter.

[0161] In this embodiment, the complex encoding unit and decoding unit may also include a complex depthwise convolution feedforward shaping unit. The output of the complex depthwise convolution feedforward shaping unit can be expressed as:

[0162] (34)

[0163] in, Represents the complex features after feedforward shaping. Represents complex depthwise convolution. This represents pointwise expansion convolution of complex numbers. This represents pointwise compressive convolution of complex numbers. This represents the learnable residual proportion parameter.

[0164] The bottleneck complex features are upsampled, skipped, fused, and restored through a multi-level complex decoder.

[0165] make The decoding process can then be represented as:

[0166] (35)

[0167] in, Indicates the first Level upsampling features, This represents the complex feature resulting from concatenating the upsampled feature with the encoder skip feature. Indicates the first Level decoding output features, Indicates the first Level complex upsampling unit, Indicates the first Level complex fusion unit, Indicates the first Level complex number decoding unit.

[0168] The output features of the final stage decoding are used as the output of the complex multi-scale U-Net backbone network. , This represents the multi-scale complex features output by the complex multi-scale U-Net backbone network.

[0169] Finally, the multi-scale complex representation is input into the complex output module to generate the radar main lobe complex echo result after interference suppression.

[0170] In some embodiments, the complex output module is configured as follows:

[0171] The multi-scale complex features are normalized to obtain normalized complex features;

[0172] The normalized complex features are mapped to a single complex output feature through complex convolution. The complex convolution is implemented by combining real part convolution kernels and imaginary part convolution kernels, so that the real and imaginary parts of the input features interact according to the complex multiplication relationship.

[0173] The complex output characteristics of the aforementioned channel are scaled to obtain the radar main lobe complex echo result after interference suppression.

[0174] Specifically, the multi-scale complex features are normalized:

[0175] (36)

[0176] in, This represents the characteristics of a complex number after normalization.

[0177] The normalized complex features are input into the complex output convolutional layer, and the multi-channel complex features are mapped into a single complex output feature through complex convolution:

[0178] (37)

[0179] in, This represents a single complex output feature obtained from a complex output convolutional layer.

[0180] Scaling the complex output characteristics of one channel yields the radar complex echo results after interference suppression:

[0181] (38)

[0182] in, This represents the radar complex echo result after interference suppression. This indicates that the output scale parameter can be learned.

[0183] In practical applications, the output results of complex numbers can be limited according to the specific needs of the application:

[0184] (39)

[0185] in, This represents the complex number output result after clipping. Indicates the lower bound of the output. This indicates the upper bound of the output.

[0186] The amplitude-limited complex output result is used as the final interference suppression result. , This represents the real part channel of the radar complex echo after interference suppression. This represents the imaginary part of the radar complex signal after interference suppression.

[0187] Based on the above, the present invention has at least the following technical advantages:

[0188] (1) This invention uses complex convolution to extract features from the one-dimensional complex signal of the radar main lobe, so that the real and imaginary parts are coupled and modeled according to the complex multiplication relationship. Compared with the method of simply treating the real and imaginary parts as ordinary real-value channels, this invention can better preserve the amplitude and phase structure information in the radar echo.

[0189] (2) The present invention sets up a complex coordinate position encoding module, which introduces linear position, sine position and cosine position as pure real complex channels into the complex feature representation, so that the network can explicitly utilize the sampling position relationship when processing one-dimensional radar complex sequences, which helps to enhance the perception of target peak position, local offset structure and interference response pattern.

[0190] (3) The present invention sets up a low-resolution multi-displacement relationship hybrid module to score and weight the candidate displacement relationship in a lower resolution space. It can model structural relationships such as local misalignment, delayed replication, multi-peak response and wide-peak interference with a small computational overhead, and introduce multi-displacement relationship information into the original resolution features through upsampling and backfeeding.

[0191] (4) This invention employs a complex-gated temporal attention module, which constructs a selective enhancement mechanism for the one-dimensional sampling direction using complex depthwise dilated convolution and complex amplitude gating. Compared to the computationally intensive global self-attention structure, this module can expand the receptive range while reducing computational overhead and perform gating enhancement on effective responses.

[0192] (5) This invention employs a complex multi-scale U-Net structure for encoding, bottleneck modeling, decoding recovery, and skip fusion, which can simultaneously utilize shallow local details and deep wide-range structural information. The decoding stage uses a combination of linear interpolation upsampling and complex convolution, which helps to reduce upsampling artifacts and maintain the stability of peak positions.

[0193] (6) The present invention introduces numerical safety processing, output limiting and gradient guardrail mechanism, which can reduce the impact of large dynamic range, abnormal values ​​and abnormal gradients of radar received signals on the training and inference process, thereby improving the stability of the model in complex radar received signal processing scenarios.

[0194] Figure 6 This is a schematic diagram of an electronic device illustrated in this specification according to an exemplary embodiment. Please refer to... Figure 6 At the hardware level, the device includes a processor 110, an internal bus 120, a network interface 130, memory 140, hardware acceleration device 150, and non-volatile memory 160, and may also include other hardware required for its functions. One or more embodiments of the present invention can be implemented in software, for example, the processor 110 reads the corresponding computer program from the non-volatile memory 160 into the memory 140 and then runs it. Of course, in addition to software implementation, one or more embodiments of the present invention do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the above processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0195] Figure 7This is a structural block diagram illustrating an exemplary embodiment of the present invention of a radar main lobe composite interference suppression device based on multi-displacement characteristics. The radar main lobe composite interference suppression device can be applied to, for example... Figure 6 The electronic device shown implements the technical solution of the present invention. The radar main lobe composite interference suppression device includes: a serialization unit 210, a position enhancement unit 220, a displacement fusion unit 230, an interference suppression unit 240, and an echo recovery unit 250, wherein:

[0196] The serialization unit 210 is used to serialize the radar main lobe received signal to be processed into a complex received signal sequence, and to process the real part and imaginary part of the complex received signal sequence into an initial complex feature by concatenating the channel dimensions.

[0197] The position enhancement unit 220 is used to construct a normalized position variable based on the sequence position of the complex received signal sequence, generate a position code based on the normalized position variable, set the position code as the real part and the imaginary part to zero to form a position code complex feature, concatenate the position code complex feature with the initial complex feature in the channel dimension, and perform mapping processing and interaction on the concatenated complex feature to obtain a position-enhanced complex feature;

[0198] The displacement fusion unit 230 is used to downsample the position-enhanced complex features to obtain low-resolution complex features; set multiple reference displacements according to the distribution range of the target signal and interference signal in the radar received signal, and translate the low-resolution complex features according to each reference displacement to generate multiple reference displacement features; perform weighted fusion on the reference displacement features, and perform residual-gated fusion on the weighted fused multi-displacement complex features and the low-resolution complex features to obtain enhanced low-resolution features; upsample the enhanced low-resolution features to the scale of the position-enhanced complex features, and superimpose and fuse them with the position-enhanced complex features to obtain complex enhanced features;

[0199] The interference suppression unit 240 is used to input the complex enhanced features into the complex multi-scale U-Net backbone network for multi-scale encoding, bottleneck modeling and decoding recovery to obtain multi-scale complex features;

[0200] The echo recovery unit 250 is used to perform output mapping and scale adjustment on the multi-scale complex representation to generate the radar main lobe complex echo result after interference suppression.

[0201] In some embodiments, the serialization unit 210 is used to concatenate the real and imaginary parts of the complex received signal sequence as two input channels along the channel dimension to form an input tensor; normalize the input tensor, and perform numerical safety processing on the normalized input tensor, replacing non-numerical, positive infinity, and negative infinity values ​​with finite values ​​within a preset numerical range to obtain safety-processed features; map the safety-processed features to multi-channel complex features through complex one-dimensional convolution, and normalize and nonlinearly activate the multi-channel complex features to obtain initial complex features; wherein the complex one-dimensional convolution is implemented using a combination of real part convolution kernels and imaginary part convolution kernels, so that the real and imaginary parts of the input features interact according to the complex multiplication relationship.

[0202] In some embodiments, the displacement fusion unit 230 is configured to normalize the low-resolution complex features and generate complex scores corresponding to each reference displacement using a complex score mapping; calculate the response score corresponding to each reference displacement based on the square of the magnitude of each complex score, and normalize the response score corresponding to each reference displacement to obtain the competition weight corresponding to each reference displacement; and perform weighted fusion of the multiple reference displacement features based on the competition weight corresponding to each reference displacement to obtain the multi-displacement fused complex features.

[0203] In some embodiments, the echo recovery unit 250 is used to normalize the multi-scale complex features to obtain normalized complex features; to map the normalized complex features into a single complex output feature through complex convolution, wherein the complex convolution is implemented using a combination of real and imaginary convolution kernels, so that the real and imaginary parts of the input features interact according to the complex multiplication relationship; and to scale the single complex output feature to obtain the radar main lobe complex echo result after interference suppression.

[0204] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0205] Accordingly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the above embodiments.

[0206] Accordingly, embodiments of the present invention also provide a computer program product configured to perform the methods described in any of the above embodiments.

[0207] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can take the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email sending and receiving device, game console, tablet computer, wearable device, or any combination of these devices.

[0208] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0209] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0210] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0211] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily intended to describe features of specific embodiments of a particular invention. Certain features described in the various embodiments herein may also be implemented in combination in a single embodiment. Conversely, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation thereof.

[0212] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0213] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0214] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0215] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for suppressing interference in radar main lobe received signals based on multi-displacement characteristics, characterized in that, Includes the following steps: Step S1: The radar main lobe received signal to be processed is serialized into a complex received signal sequence, and the real part and imaginary part of the complex received signal sequence are concatenated by channel dimension and then processed into initial complex features. Step S2: Construct normalized position variables based on the sequence positions of the complex received signal sequence; generate position codes based on the normalized position variables; set the position codes to zero as real and imaginary parts to form position code complex features; concatenate the position code complex features with the initial complex features in the channel dimension; perform mapping processing and interaction on the concatenated complex features to obtain position-enhanced complex features. Step S3: Downsample the position-enhanced complex features to obtain low-resolution complex features; set multiple reference displacements according to the distribution range of the target signal and interference signal in the radar received signal, and translate the low-resolution complex features according to each reference displacement to generate multiple reference displacement features; The reference displacement features are weighted and fused, and the weighted and fused multi-displacement complex features are residual-gated and fused with the low-resolution complex features to obtain the enhanced low-resolution features. The enhanced low-resolution features are upsampled to the scale of the location-enhanced complex features and then superimposed and fused with the location-enhanced complex features to obtain complex enhanced features; Step S4: Input the complex enhanced features into the complex multi-scale U-Net backbone network for multi-scale encoding, bottleneck modeling and decoding recovery to obtain multi-scale complex features; Step S5: Perform output mapping and scale adjustment on the multi-scale complex features to generate the radar main lobe complex echo result after interference suppression.

2. The method according to claim 1, characterized in that, Step S1 includes: The real and imaginary parts of the complex received signal sequence are used as two input channels, and they are spliced ​​along the channel dimension to form an input tensor; The input tensor is normalized, and the normalized input tensor is then subjected to numerical safety processing, replacing non-numerical, positive infinity, and negative infinity values ​​with finite values ​​within a preset numerical range to obtain the safety-processed features. The security-processed features are mapped to multi-channel complex features through complex one-dimensional convolution, and the multi-channel complex features are normalized and non-linearly activated to obtain initial complex features. The complex one-dimensional convolution is implemented by combining real part convolution kernels and imaginary part convolution kernels, so that the real part and imaginary part of the input features interact according to the complex multiplication relationship.

3. The method according to claim 1, characterized in that, The position encoding in step S2 includes linear position encoding, sine position encoding, and cosine position encoding.

4. The method according to claim 1, characterized in that, Step S3 involves weighted fusion of the reference displacement features, including: The low-resolution complex features are normalized, and complex scores corresponding to each reference displacement are generated using complex score mapping. The response score corresponding to each reference displacement is calculated based on the square of the amplitude of each complex score, and the response score corresponding to each reference displacement is normalized to obtain the competitive weight corresponding to each reference displacement. Based on the competition weights corresponding to each reference displacement, the features of the multiple reference displacements are weighted and fused to obtain the multi-displacement fused complex features.

5. The method according to claim 4, characterized in that, The response scores corresponding to each reference displacement are as follows: ； in, Indicates reference displacement The corresponding response score Indicates the length of the low-resolution complex feature. Indicates the sampling point index. This represents a preset constant used to improve numerical stability.

6. The method according to claim 1, characterized in that, Step S5 includes: The multi-scale complex features are normalized to obtain normalized complex features; The normalized complex features are mapped to a single complex output feature through complex convolution. The complex convolution is implemented by combining real part convolution kernels and imaginary part convolution kernels, so that the real and imaginary parts of the input features interact according to the complex multiplication relationship. The complex output characteristics of the aforementioned channel are scaled to obtain the radar main lobe complex echo result after interference suppression.

7. A radar main lobe receiving signal interference suppression device based on multi-displacement characteristics, characterized in that, The device includes: The serialization unit is used to serialize the radar main lobe received signal to be processed into a complex received signal sequence, and to process the real part and imaginary part of the complex received signal sequence by concatenating them along the channel dimension to obtain the initial complex features. The position enhancement unit is used to construct a normalized position variable based on the sequence position of the complex received signal sequence, generate a position code based on the normalized position variable, set the position code as the real part and the imaginary part to zero to form a position code complex feature, concatenate the position code complex feature with the initial complex feature in the channel dimension, and perform mapping processing and interaction on the concatenated complex feature to obtain the position enhanced complex feature; A displacement fusion unit is used to downsample the position-enhanced complex features to obtain low-resolution complex features; set multiple reference displacements according to the distribution range of target signals and interference signals in the radar received signal, and translate the low-resolution complex features according to each reference displacement to generate multiple reference displacement features; perform weighted fusion on the reference displacement features, and perform residual-gated fusion on the weighted fused multi-displacement complex features and the low-resolution complex features to obtain enhanced low-resolution features; upsample the enhanced low-resolution features to the scale of the position-enhanced complex features, and superimpose and fuse them with the position-enhanced complex features to obtain complex enhanced features; The interference suppression unit is used to input the complex enhanced features into the complex multi-scale U-Net backbone network for multi-scale encoding, bottleneck modeling and decoding recovery to obtain multi-scale complex features; The echo recovery unit is used to perform output mapping and scale adjustment on the multi-scale complex features to generate the radar main lobe complex echo result after interference suppression.

8. An electronic device, characterized in that, include: processor; as well as A computer-readable storage medium storing computer program instructions that, when executed by the processor, cause the processor to perform the method as described in any one of claims 1 to 6.

Images