Electromagnetic signal denoising method based on deep denoiser model

Abstract

This invention discloses an electromagnetic signal denoising method based on the DeepDenoiser model, belonging to the field of electromagnetic signal processing technology. It solves the technical problems of poor denoising effect and insufficient adaptive capability in existing electromagnetic signal denoising technologies. This invention effectively captures the feature information of electromagnetic signals at different scales by employing multi-scale feature extraction and adaptive feature fusion mechanisms, thus improving the denoising effect. By introducing an attention mechanism, the model can adaptively focus on important feature channels and spatial locations, enhancing its adaptive capability. By designing a composite loss function, it comprehensively considers the denoising performance in both the time and frequency domains, ensuring that the denoised signal maintains waveform similarity while its spectral characteristics are closer to those of the clean signal. The method of this invention is applicable to various types of electromagnetic signals and has good versatility and practicality.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic signal processing technology, and in particular to an electromagnetic signal denoising method based on the DeepDenoiser model. Background Technology

[0002] Electromagnetic signals are inevitably subject to various noise interferences during transmission, including thermal noise, impulse noise, and narrowband interference. These noises can severely affect signal quality and reduce the reliability and effectiveness of communication systems. Therefore, electromagnetic signal denoising is an important research topic in the field of signal processing.

[0003] Traditional electromagnetic signal denoising methods mainly include filtering, wavelet transform, and spectral subtraction. Filtering suppresses noise by designing specific filters, but its effectiveness is limited for non-stationary signals and complex noisy environments. Wavelet transform uses wavelet basis functions to perform multi-resolution analysis of the signal, enabling simultaneous location of signal features in the time and frequency domains; however, the choice of wavelet basis significantly impacts the denoising effect. Spectral subtraction, based on the statistical properties of noise, estimates the noise spectrum and subtracts it from the noisy signal spectrum to achieve denoising, but it is prone to generating musical noise.

[0004] In recent years, with the rapid development of deep learning technology, signal denoising methods based on deep neural networks have made significant progress. DeepDenoiser is an end-to-end signal denoising model based on deep learning, capable of automatically learning the complex mapping relationship between signals and noise, and showing obvious advantages in denoising performance and generalization ability. However, existing DeepDenoiser models still face the following problems when applied to electromagnetic signal denoising: First, electromagnetic signals have complex time-frequency characteristics, and single-scale feature extraction is difficult to fully capture signal features; second, electromagnetic signals of different frequency bands and different modulation methods vary greatly, and the model's adaptive ability needs to be improved; third, model training requires a large amount of labeled data, while obtaining clean electromagnetic signals is relatively difficult.

[0005] Therefore, an electromagnetic signal denoising method based on the DeepDenoiser model is needed. Summary of the Invention

[0006] To address the problems of poor electromagnetic signal denoising performance and insufficient adaptability in existing technologies, this invention provides an electromagnetic signal denoising method based on the DeepDenoiser model. The specific technical solution is as follows: Electromagnetic signal denoising methods based on the DeepDenoiser model include: Acquire the noisy electromagnetic signal to be processed, and preprocess the noisy electromagnetic signal to obtain standardized signal data; The standardized signal data is input into the trained DeepDenoiser model, and the encoder of the DeepDenoiser model extracts the multi-scale feature representation of the signal. The feature fusion module of the DeepDenoiser model is used to adaptively fuse the multi-scale features to obtain fused features; The fused features are reconstructed by the decoder of the DeepDenoiser model, and the denoised electromagnetic signal is output.

[0007] Preferably, the preprocessing includes: The noisy electromagnetic signal is subjected to sampling rate normalization processing to convert signals with different sampling rates into a preset standard sampling rate. The normalized signal is subjected to amplitude standardization processing to scale the signal amplitude to a preset range; The standardized signal is subjected to frame-by-frame windowing to obtain a fixed-length signal frame sequence.

[0008] Preferably, the encoder of the DeepDenoiser model includes multiple cascaded convolutional blocks, each convolutional block containing: One-dimensional convolutional layers are used to extract local features of a signal; Batch normalization layers are used to accelerate model convergence and improve generalization ability; Activation function layer, used to introduce nonlinear transformations; Downsampling layers are used to reduce feature dimensionality and expand the receptive field.

[0009] Preferably, the feature fusion module employs an attention mechanism, including: The channel attention submodule is used to learn the importance weights of different feature channels; The spatial attention submodule is used to learn the importance weights of different spatial locations in the feature map; The feature weighted fusion unit is used to adaptively weight and fuse multi-scale features according to the importance weights.

[0010] Preferably, the training process of the DeepDenoiser model includes: Construct a training dataset, which includes clean electromagnetic signals and corresponding noisy signals; Design a composite loss function, which includes mean square error loss, sensing loss, and frequency domain loss; An adaptive learning rate strategy is used to iteratively train the model until it converges.

[0011] Preferably, the expression for the composite loss function is: in, For mean square error loss, In order to perceive loss, For frequency domain loss, α, β, and γ are the weighting coefficients for each loss.

[0012] Preferably, the method further includes: The quality of the denoised electromagnetic signal is evaluated, and the signal-to-noise ratio improvement and signal distortion are calculated. The parameters of the DeepDenoiser model were fine-tuned and optimized based on the evaluation results.

[0013] An electromagnetic signal denoising device based on the DeepDenoiser model, comprising: The signal acquisition module is used to acquire the noisy electromagnetic signal to be processed; The preprocessing module is used to preprocess the noisy electromagnetic signal to obtain standardized signal data; The feature extraction module is used to input the standardized signal data into the trained DeepDenoiser model and extract multi-scale feature representations of the signal through the encoder of the DeepDenoiser model. The feature fusion module is used to adaptively fuse the multi-scale features using the feature fusion module of the DeepDenoiser model to obtain fused features. The signal reconstruction module is used to reconstruct the fused features through the decoder of the DeepDenoiser model and output the denoised electromagnetic signal.

[0014] A computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the electromagnetic signal denoising method based on the DeepDenoiser model as described above.

[0015] A processor for running a program, wherein the program, when running, performs an electromagnetic signal denoising method based on the DeepDenoiser model as described above.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention effectively captures the feature information of electromagnetic signals at different scales by employing multi-scale feature extraction and adaptive feature fusion mechanisms, thus improving the denoising effect. By introducing an attention mechanism, the model can adaptively focus on important feature channels and spatial locations, enhancing its adaptability. By designing a composite loss function, the denoising performance in both the time and frequency domains is comprehensively considered, ensuring that the denoised signal maintains waveform similarity while its spectral characteristics are closer to those of the clean signal. The method of this invention is applicable to various types of electromagnetic signals and has good versatility and practicality. Attached Figure Description

[0017] 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. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] Example 1 The method implemented in this embodiment can be implemented in a system, on a server, or on a terminal; no specific limitation is made. The method in this application will be described below from the perspective of system implementation. The method provided in this application embodiment is as follows: Figure 1 As shown, it includes the following steps: 101. Obtain the noisy electromagnetic signal to be processed.

[0024] In this embodiment, the system acquires the noisy electromagnetic signal to be processed through a signal acquisition device. The signal acquisition device can be a software-defined radio (SDR) device, an oscilloscope, or other electromagnetic signal acquisition device. The acquired signal can be a radio frequency signal, a baseband signal, or an intermediate frequency signal, and the sampling rate is determined according to the actual application requirements.

[0025] 102. Preprocess the noisy electromagnetic signals.

[0026] Preprocessing includes three steps: sampling rate normalization, amplitude normalization, and frame windowing. Sampling rate normalization converts signals with different sampling rates into a preset standard sampling rate (e.g., 10MHz); amplitude normalization scales the signal amplitude to the range of [-1, 1]; frame windowing divides the continuous signal into frames of fixed length (e.g., 1024 sampling points per frame) and applies a Hamming window to reduce spectral leakage.

[0027] Furthermore, three key parameters are introduced: signal-to-noise ratio estimation parameter (instantaneous signal-to-noise ratio estimate), impulse interference detection parameter (impulse interference detection coefficient), and bandwidth occupancy parameter (bandwidth occupancy factor). These three parameters are then integrated into an adaptive preprocessing gain coefficient. It is used to dynamically adjust the amplitude normalization range and frame length, as follows: In the formula, This is the instantaneous signal-to-noise ratio estimate. The signal-to-noise ratio threshold. The pulse interference detection coefficient is... This is the frequency band occupancy factor. , , These are the signal-to-noise ratio weighting coefficient, impulse interference weighting coefficient, and bandwidth occupancy weighting coefficient, respectively. It is the hyperbolic tangent function. It is an S-shaped function.

[0028] From the formula, it can be seen that when pulse interference is detected ( >3) when, It activates quickly and tends to 0 when there is no interference.

[0029] Get Then, adaptive preprocessing is performed based on this, as follows: 1. Adjust the original fixed range [−1,1] to a dynamic range. : In the formula, This is the dynamic range adjustment coefficient, a constant (e.g., 0.5).

[0030] This formula shows that when When the noise level is low (significant), increase the dynamic range to preserve details of weak signals; when... When the signal is large (good signal quality), use the standard range.

[0031] 2. The original fixed frame length (1024 points) is adjusted to a dynamic frame length. : In the formula, Based on the base frame length (1024). This is the frame length adjustment coefficient, which is a constant (e.g., 0.2).

[0032] Because electromagnetic signals have complex time-frequency characteristics, for narrowband signals ( →1) A longer frame length is required to ensure frequency resolution; for broadband signals ( →0) A shorter frame length is required to ensure temporal resolution.

[0033] 3. Based on the pulse interference detection coefficient Dynamic selection window function: In the formula, For window functions, For Hanning window, For Kaiser windows, For sampling point index, The pulse detection threshold is set to 3.

[0034] As can be seen from the above formula, in the absence of pulse interference, priority is given to spectral leakage suppression, while in the presence of pulse interference, priority is given to sidelobe attenuation. This is because the "musical noise" problem of spectral subtraction is related to spectral leakage, and the Hanning window balances between a flat top and sidelobe attenuation; the Kaiser window adjusts... The parameter (6 is chosen here because of the sidelobe suppression requirement in the case of impulse interference) can better suppress the spectral spread caused by the impulse.

[0035] For example, the instantaneous signal-to-noise ratio estimate is used to reflect the signal-to-noise ratio level of the current signal frame and is used to dynamically adjust the preprocessing intensity. The calculation process is as follows: In the formula, The total power of the signal frame. This is an estimate of the noise power.

[0036] The total power of the signal frame is obtained by calculating the spectral energy after performing an FFT transform on the signal x(n): In the formula, For FFT points, For frequency point index, This represents the signal spectrum value.

[0037] The process of obtaining the noise power estimate is as follows: using the spectral subtraction principle from the background technique, select frequency points in the signal spectrum without obvious signal components (usually the guard bands at both ends of the spectrum) to calculate the average power: In the formula, To protect the number of frequency points, To protect the frequency index, To protect the frequency band set, To protect the band spectrum values.

[0038] For example, the impulse interference detection coefficient is used to identify whether impulse noise exists in the current signal frame and its severity, and its specific values ​​are as follows: in, The average value of the signal frames. The variance of the signal frame. It is the fourth-order central moment.

[0039] For example, when The presence of impulse interference is determined at that time (the Gaussian distribution has a kurtosis of 3, and the impulse noise has a peak characteristic with a kurtosis much greater than 3).

[0040] For example, the bandwidth occupancy factor reflects the concentration of signal distribution in the frequency domain and is used for adaptive framing strategies. Its calculation process is as follows: in, For the first Normalized power spectral density at each frequency point Spectral entropy measures the uniformity of the spectral distribution. This represents the maximum entropy value.

[0041] The value range is [0,1]. The larger the value, the more concentrated the signal energy (obvious narrowband characteristics), and the smaller the value, the more dispersed the energy (wideband characteristics).

[0042] 103. Input the preprocessed signal into the DeepDenoiser model for feature extraction.

[0043] The DeepDenoiser model's encoder consists of four cascaded convolutional blocks. Each block contains a one-dimensional convolutional layer (kernel size 3, stride 1), a batch normalization layer, a ReLU activation function layer, and a max-pooling downsampling layer (kernel size 2). Through multi-level convolution and downsampling, the encoder is able to extract multi-scale feature representations of the signal, capturing hierarchical features from local details to global structure.

[0044] 104. Adaptive feature fusion is performed using the feature fusion module.

[0045] The feature fusion module employs a mechanism combining channel attention and spatial attention. The channel attention submodule learns the importance weights of different feature channels through global average pooling and fully connected layers; the spatial attention submodule learns the importance weights of different spatial locations in the feature map through convolutional operations. The feature weighted fusion unit adaptively weights and fuses multi-scale features based on the learned weights, highlighting important features and suppressing irrelevant features.

[0046] 105. Reconstruct the denoised electromagnetic signal using a decoder.

[0047] The decoder employs a symmetrical structure to the encoder, comprising four cascaded deconvolutional blocks. Each deconvolutional block contains an upsampling layer, a one-dimensional convolutional layer, a batch normalization layer, and a ReLU activation function layer. The decoder progressively upsamples and reconstructs the fused features, ultimately outputting the denoised electromagnetic signal. In the final layer of the decoder, the Tanh activation function is used to constrain the output to the range [-1, 1].

[0048] 106. Output the denoised electromagnetic signal.

[0049] The system outputs the denoised electromagnetic signal to subsequent processing modules, such as modulation identification, demodulation, or spectrum analysis modules. Simultaneously, the system can evaluate the denoising effect, calculating the signal-to-noise ratio improvement (SNRI) and signal distortion (SD) of the output signal for model performance monitoring and optimization.

[0050] In summary, the process of reconstructing the fused features and outputting the denoised electromagnetic signal through the decoder of the DeepDenoiser model can be understood as follows: using a mask matrix to process the time-frequency domain image under the signal-noise aliasing state, the time-frequency domain images of the separated signal and noise can be obtained separately. Then, the time-frequency domain image is restored to the time domain through inverse short-time Fourier transform to obtain the separated time-domain signal and noise, thereby completing the signal denoising process.

[0051] Example 2 Embodiments of the device of the present invention include the following modules: The signal acquisition module is used to acquire the noisy electromagnetic signal to be processed; The preprocessing module is used to preprocess the noisy electromagnetic signal to obtain standardized signal data; The feature extraction module is used to input the standardized signal data into the trained DeepDenoiser model and extract multi-scale feature representations of the signal through the encoder of the DeepDenoiser model. The feature fusion module is used to adaptively fuse the multi-scale features using the feature fusion module of the DeepDenoiser model to obtain fused features. The signal reconstruction module is used to reconstruct the fused features through the decoder of the DeepDenoiser model and output the denoised electromagnetic signal.

[0052] For specific limitations regarding the device, please refer to the limitations of the method above, which will not be repeated here. Each module in the above device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0053] In summary, this invention effectively captures the feature information of electromagnetic signals at different scales by employing multi-scale feature extraction and adaptive feature fusion mechanisms, thereby improving the denoising effect. By introducing an attention mechanism, the model can adaptively focus on important feature channels and spatial locations, enhancing its adaptability. By designing a composite loss function that comprehensively considers the denoising performance in both the time and frequency domains, the denoised signal maintains waveform similarity while its spectral characteristics are closer to those of the clean signal. The method of this invention is applicable to various types of electromagnetic signals and has good versatility and practicality.

[0054] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0055] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0056] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0057] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for electromagnetic signal denoising based on the DeepDenoiser model, characterized in that, include: S1: Acquire the noisy electromagnetic signal to be processed, and preprocess the noisy electromagnetic signal to obtain standardized signal data. The preprocessing includes: After normalizing and standardizing the signal, the normalization range [−1,1] is adjusted to the dynamic range. The details are as follows: In the formula, This is the dynamic range adjustment coefficient. For adaptive preprocessing gain coefficients; S2: Input the standardized signal data into the trained DeepDenoiser model, and extract the multi-scale feature representation of the signal through the encoder of the DeepDenoiser model; S3: Adaptively fuse the multi-scale features using the feature fusion module of the DeepDenoiser model to obtain fused features; S4: The fused features are reconstructed by the decoder of the DeepDenoiser model, and the denoised electromagnetic signal is output.

2. The electromagnetic signal denoising method based on the DeepDenoiser model according to claim 1, characterized in that, Adaptive preprocessing gain coefficient The possible values ​​are as follows: In the formula, This is the instantaneous signal-to-noise ratio estimate. The signal-to-noise ratio threshold. The pulse interference detection coefficient is... This is the frequency band occupancy factor. , , These are the signal-to-noise ratio weighting coefficient, impulse interference weighting coefficient, and bandwidth occupancy weighting coefficient, respectively. It is the hyperbolic tangent function. It is an S-shaped function.

3. The electromagnetic signal denoising method based on the DeepDenoiser model according to claim 1, characterized in that, The encoder of the DeepDenoiser model comprises multiple cascaded convolutional blocks, each containing: One-dimensional convolutional layers are used to extract local features of a signal; Batch normalization layers are used to accelerate model convergence and improve generalization ability; Activation function layer, used to introduce nonlinear transformations; Downsampling layers are used to reduce feature dimensionality and expand the receptive field.

4. The electromagnetic signal denoising method based on the DeepDenoiser model according to claim 1, characterized in that, The feature fusion module employs an attention mechanism, including: The channel attention submodule is used to learn the importance weights of different feature channels; The spatial attention submodule is used to learn the importance weights of different spatial locations in the feature map; The feature weighted fusion unit is used to adaptively weight and fuse multi-scale features according to the importance weights.

5. The electromagnetic signal denoising method based on the DeepDenoiser model according to claim 1, characterized in that, The training process of the DeepDenoiser model includes: Construct a training dataset, which includes clean electromagnetic signals and corresponding noisy signals; Design a composite loss function, which includes mean square error loss, sensing loss, and frequency domain loss; An adaptive learning rate strategy is used to iteratively train the model until it converges.

6. The electromagnetic signal denoising method based on the DeepDenoiser model according to claim 1, characterized in that, The expression for the composite loss function is: in, For mean square error loss, In order to perceive loss, For frequency domain loss, α, β, and γ are the weighting coefficients for each loss.

7. The electromagnetic signal denoising method based on the DeepDenoiser model according to claim 1, characterized in that, The method further includes: The quality of the denoised electromagnetic signal is evaluated, and the signal-to-noise ratio improvement and signal distortion are calculated. The parameters of the DeepDenoiser model were fine-tuned and optimized based on the evaluation results.

8. An electromagnetic signal denoising device based on the DeepDenoiser model, characterized in that, include: The signal acquisition module is used to acquire the noisy electromagnetic signal to be processed; The preprocessing module is used to preprocess the noisy electromagnetic signal to obtain standardized signal data; The feature extraction module is used to input the standardized signal data into the trained DeepDenoiser model and extract multi-scale feature representations of the signal through the encoder of the DeepDenoiser model. The feature fusion module is used to adaptively fuse the multi-scale features using the feature fusion module of the DeepDenoiser model to obtain fused features. The signal reconstruction module is used to reconstruct the fused features through the decoder of the DeepDenoiser model and output the denoised electromagnetic signal.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the electromagnetic signal denoising method based on the DeepDenoiser model as described in any one of claims 1 to 7.

10. A processor, characterized in that, The processor is used to run a program, wherein the program executes the electromagnetic signal denoising method based on the DeepDenoiser model as described in any one of claims 1 to 7.

Images