Adaptive noise reduction signal modulation class recognition method, device and system

Abstract

The application relates to the field of communication, in particular to a signal modulation category identification method, device and system with adaptive noise reduction and electronic equipment. The method comprises the following steps: acquiring waveform statistical features of a one-dimensional radio time domain signal; determining noise signal strength of noise signals in the one-dimensional radio time domain signal based on the waveform statistical features, and constructing an adaptive filter according to the noise signal strength; filtering the one-dimensional radio time domain signal by using the adaptive filter to obtain a one-dimensional filtered signal; performing short-time discrete Fourier transform on the one-dimensional filtered signal to obtain a two-dimensional frequency spectrum image; extracting and classifying features of the two-dimensional frequency spectrum image by using a convolutional neural network model, and determining a modulation category of the one-dimensional radio time domain signal. The signal modulation category identification method, device and system with adaptive noise reduction and electronic equipment provided by the application realize adaptive noise reduction in the process of wireless communication modulation identification, and improve the technical effects of noise reduction effect and modulation identification accuracy.

Description

Technical Field

[0001] This application relates to the field of communications, specifically to an adaptive noise reduction signal modulation category identification method, apparatus and system, and electronic equipment. Background Technology

[0002] Automatic Modulation Recognition (AMR) is a core intermediate step between signal detection and demodulation in wireless communication systems. In scenarios such as non-cooperative communication, spectrum monitoring, and electromagnetic situational awareness, the receiver cannot know the modulation scheme of the transmitted signal in advance. It needs to rely on automatic recognition technology to determine the signal modulation scheme, providing a core basis for subsequent signal demodulation, interference investigation, and spectrum resource management.

[0003] With the large-scale deployment of wireless services such as 5G, 6G, IoT, satellite communication, and shortwave communication, the space electromagnetic environment is becoming increasingly complex, with problems such as multi-signal aliasing, multipath fading, strong noise interference, and low signal-to-noise ratio becoming commonplace. Traditional manual identification and simple machine learning identification schemes can no longer meet the needs of identifying massive amounts of real-time signals. Convolutional Neural Networks (CNNs), with their advantages such as local receptive fields, weight sharing, and feature downsampling, are being used in conjunction with modulation recognition, which is currently the mainstream research direction in the industry.

[0004] However, existing wireless communication modulation recognition technologies have poor filtering effects on noise signals, and suffer from problems such as easy loss of signal features in frequency domain denoising and poor recognition stability in low signal-to-noise ratio environments, resulting in reduced modulation recognition accuracy. Summary of the Invention

[0005] In view of this, it is necessary to provide a signal modulation category identification method, device, system, and electronic device for adaptive noise reduction, so as to achieve adaptive noise reduction in the process of wireless communication modulation identification, thereby improving the noise reduction effect and modulation identification accuracy.

[0006] To achieve the aforementioned technical effects, firstly, this application provides an adaptive noise reduction signal modulation category identification method, comprising: Acquire one-dimensional radio time-domain signals and obtain the waveform statistical characteristics of the one-dimensional radio time-domain signals; Based on the waveform statistical characteristics, the noise signal intensity of the noise signal in the one-dimensional radio time domain signal is determined, and an adaptive filter is constructed based on the noise signal intensity. The one-dimensional radio time-domain signal is filtered using the adaptive filter to obtain a one-dimensional filtered signal. A short-time discrete Fourier transform is performed on the one-dimensional filtered signal to obtain a two-dimensional spectrum image. A convolutional neural network model is then used to extract features and classify the two-dimensional spectrum image to determine the modulation category of the one-dimensional radio time-domain signal.

[0007] In one possible embodiment, the adaptive filter is a Gaussian filter, and the step of constructing the adaptive filter based on the noise signal intensity includes: Construct a standard Gaussian filter; The standard deviation of the target filter is determined based on the noise signal intensity. Replace the standard deviation of the standard Gaussian filter with the standard deviation of the target filter.

[0008] In one possible embodiment, multiple preset index ranges and preset signal strengths corresponding to each preset index range are provided. The step of determining the noise signal strength of the noise signal in the one-dimensional radio time-domain signal based on the waveform statistical characteristics includes: The waveform statistical features are weighted according to preset weighting coefficients to obtain noise evaluation indicators; The preset index range to which the noise evaluation index belongs is determined as the target index range, and the preset signal strength corresponding to the target index range is determined as the noise signal strength.

[0009] In one possible embodiment, the convolutional neural network model includes multiple convolutional layers, multiple max-pooling layers, densely connected layers, and a softmax layer. The multiple convolutional layers are sequentially connected via the max-pooling layers, and a unique max-pooling layer is set between every two convolutional layers. The step of using the convolutional neural network model to extract features and classify the two-dimensional spectral image includes: The multiple convolutional layers and the multiple max pooling layers are used to extract depth features from the two-dimensional spectral image to obtain two-dimensional spectral features. The densely connected layer is used to perform dimensional mapping on the two-dimensional spectral features to obtain a mapped image; The Softmax layer is used to classify the mapped image, and the probability distribution of each modulation category is output.

[0010] In one possible embodiment, it further includes: The convolutional neural network model is trained using the gradient descent method.

[0011] In one possible embodiment, it further includes: The learning rate of the convolutional neural network model is set to gradually increase as the number of training iterations increases.

[0012] In one possible embodiment, it further includes: The number of filters in the multiple convolutional layers decreases sequentially, and the pooling kernel size of the multiple max pooling layers is the same.

[0013] Secondly, this application provides an adaptive noise reduction signal modulation category identification device, comprising: A signal acquisition module is used to acquire one-dimensional radio time-domain signals and obtain the waveform statistical characteristics of the one-dimensional radio time-domain signals. An adaptive filter construction module is used to determine the noise signal intensity of the noise signal in the one-dimensional radio time-domain signal based on the waveform statistical characteristics, and to construct an adaptive filter based on the noise signal intensity. A signal filtering module is used to filter the one-dimensional radio time-domain signal using the adaptive filter to obtain a one-dimensional filtered signal. The modulation category identification module is used to perform a short-time discrete Fourier transform on the one-dimensional filtered signal to obtain a two-dimensional spectrum image, and to use a convolutional neural network model to extract features and classify the two-dimensional spectrum image to determine the modulation category of the one-dimensional radio time-domain signal.

[0014] Thirdly, this application provides an electronic device, including: an input serial bus, an output serial bus, a memory, and a processor, wherein... The input serial bus is used to input one-dimensional radio time-domain signals; The memory is used to store the program and the one-dimensional radio time-domain signal; The processor, coupled to the memory, is used to execute the program stored in the memory based on the one-dimensional radio time-domain signal to implement the steps in the aforementioned adaptive noise reduction signal modulation category identification method, so as to determine the modulation category of the one-dimensional radio time-domain signal. The output serial bus is used to output the modulation category of the one-dimensional radio time-domain signal to the result display device.

[0015] Fourthly, this application provides an adaptive noise reduction signal modulation category recognition system, comprising: A signal acquisition device, a signal modulation category identification device for adaptive noise reduction as described above, and a result display device connected to the signal modulation category identification device for adaptive noise reduction; The signal acquisition device is used to acquire one-dimensional radio time-domain signals; The adaptive noise reduction signal modulation category identification device is used to perform the steps in the aforementioned adaptive noise reduction signal modulation category identification method based on the one-dimensional radio time domain signal to determine the modulation category of the one-dimensional radio time domain signal; The result display device is used to display the modulation category.

[0016] The beneficial effects of this application are as follows: Compared with related technologies, the adaptive noise reduction signal modulation category identification method, apparatus, system, and electronic device provided in this application, after acquiring the one-dimensional radio time-domain signal for which signal modulation category identification is required, directly obtains the waveform statistical characteristics of the one-dimensional radio time-domain signal, then determines the noise signal intensity of the noise signal included in the one-dimensional radio time-domain signal based on the waveform statistical characteristics, adaptively constructs an adaptive filter based on the noise signal intensity, uses the adaptive filter to filter the one-dimensional radio time-domain signal to obtain a one-dimensional filtered signal, and performs a short-time discrete Fourier transform on the one-dimensional filtered signal. The process involves obtaining a two-dimensional spectrum image, using a convolutional neural network model to extract and classify features from the two-dimensional spectrum image, and determining the modulation category of the one-dimensional radio time-domain signal. Since this application constructs an adaptive filter corresponding to the waveform statistical characteristics of each one-dimensional radio time-domain signal, the adaptive filter has a better filtering effect on the one-dimensional radio time-domain signal, resulting in a more accurate one-dimensional filtered signal with less residual noise. Consequently, the accuracy of the modulation category determined based on the one-dimensional filtered signal is also higher. This achieves the technical effect of adaptive noise reduction in the process of wireless communication modulation identification, improving the noise reduction effect and modulation identification accuracy. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the adaptive noise reduction signal modulation category identification method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the convolutional neural network model in the adaptive noise reduction signal modulation category recognition method provided in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the change in the recognition accuracy of the convolutional neural network with the number of training iterations during the model training process in the adaptive noise reduction signal modulation category recognition method provided in this application embodiment. Figure 4This is a schematic diagram of the structure of the adaptive noise reduction signal modulation category identification device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this application; Figure 6 This is a schematic diagram of the adaptive noise reduction signal modulation category recognition system provided in the embodiments of this application; Figure 7 This diagram illustrates the relationship between the recognition accuracy of SCNN, SCNN2, CNNR-IQ, and CNNR-IQFOC and the signal-to-noise ratio (SNR) of a one-dimensional radio time-domain signal. Figure 8 This diagram illustrates the relationship between the recognition accuracy of SCNN2, AF-CNN, and SCF-CNN and the signal-to-noise ratio (SNR) of a one-dimensional radio time-domain signal. Figure 9 and Figure 10 This diagram illustrates the computational complexity of SCNN2, CNNR-IQ, and CNNR-IQFOC. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The terms "first," "second," etc., used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] This application provides a method, apparatus, system, and electronic device for adaptive noise reduction signal modulation category identification, which will be described below.

[0024] Please refer to Figure 1 The adaptive noise reduction signal modulation category identification method provided in this application includes: Step S101: Acquire one-dimensional radio time-domain signals and obtain the waveform statistical characteristics of the one-dimensional radio time-domain signals.

[0025] In this step, the one-dimensional radio time-domain signal refers to the raw data of a radio signal that varies continuously in the time dimension. This signal typically contains useful modulation information and various noise components, serving as the initial input for signal modulation category identification. Taking a simple communication system consisting of a transmitter, channel, and receiver as an example, let s(t) represent the transmitted symbol to be sent to the receiver. The transmitted symbol s(t) is first converted into a transmitted signal using a modulation function F, and then the signal is transmitted to the receiver through the communication channel h(t). Let y(t) represent the observed signal of the transmitted symbol s(t) at the receiver, i.e., the one-dimensional radio time-domain signal in this step. y(t) can also be expressed as... ,in The transmitted symbol s(t) is the signal generated by changes during transmission, while v(t) is the noise signal. Based on this, for a one-dimensional radio time-domain signal y(t), modulation identification is to predict the modulation function F based on the one-dimensional radio time-domain signal y(t), thereby providing modulation information so as to demodulate the transmitted symbol s(t) from the observed signal y(t).

[0026] Furthermore, waveform statistical characteristics are a quantitative description of a one-dimensional radio time-domain signal at the statistical level. Specifically, they include at least some of various characteristics such as instantaneous amplitude, instantaneous phase, instantaneous frequency, and higher-order statistics. In this step, obtaining the waveform statistical characteristics of a one-dimensional radio time-domain signal involves performing statistical analysis on the signal, calculating its instantaneous amplitude, instantaneous phase, instantaneous frequency, and higher-order statistics to obtain waveform statistical characteristics that characterize the signal.

[0027] In this embodiment, the one-dimensional radio signal includes mainstream wireless communication modulation signals such as BPSK, QPSK, 8PSK, 16QAM, 64QAM, BFSK, CPFSK, PAM4, WB-FM, AM-SSB, and AM-DSB, and the specific signal sampling format is IQ quadrature sampling format.

[0028] Step S102: Determine the noise signal intensity of the noise signal in the one-dimensional radio time domain signal based on the waveform statistical characteristics, and construct an adaptive filter based on the noise signal intensity.

[0029] In this step, multiple preset index ranges and preset signal strengths corresponding to each preset index range are pre-set. Based on this, the waveform statistical features are weighted according to preset weight coefficients to obtain noise evaluation indexes. The preset index range to which the noise evaluation index belongs is determined as the target index range, and the preset signal strength corresponding to the target index range is determined as the noise signal strength.

[0030] The preset index range divides the possible value space of the noise evaluation index into several continuous or discrete intervals, each interval being a preset index range, and each preset index range is associated with a preset signal strength. In this embodiment, the correspondence between the preset index range and the preset signal strength is specifically determined through experimental analysis and statistical modeling of a large number of radio signals with different noise levels. For example, in a laboratory environment, by injecting noise signals of known strength and recording their waveform statistical characteristics and corresponding noise evaluation indexes, the correspondence between the index range and the noise signal strength is established. Alternatively, in some other embodiments of this application, the setting can be based on expert experience or domain knowledge, such as manually dividing the index range and specifying the corresponding noise signal strength according to common noise levels in different communication scenarios.

[0031] The preset weight coefficients are optimized during the training phase using machine learning methods to maximize the correlation between noise evaluation metrics and actual noise intensity, or they can be set based on theoretical knowledge or rules of thumb in the field of signal processing.

[0032] In this embodiment, the waveform statistical features include instantaneous amplitude variance. instantaneous phase change variance Instantaneous frequency fluctuation and fourth-order cumulative quantity Each corresponds to a preset weighting coefficient. , , , Based on the formula Noise evaluation index was calculated For example, Meanwhile, the system pre-sets several preset index ranges and their corresponding preset signal strengths: when the noise evaluation index is in the range of [0, 0.3), the corresponding preset signal strength is 0.1 (indicating low noise); when the noise evaluation index is in the range of [0.3, 0.7), the corresponding preset signal strength is 0.5 (indicating medium noise); and when the noise evaluation index is in the range of [0.7, 1.0], the corresponding preset signal strength is 0.9 (indicating high noise). Because the calculated noise evaluation index... Since it falls within the preset index range of [0.3, 0.7), the target index range is determined to be [0.3, 0.7), and the corresponding preset signal strength of 0.5 is determined as the noise signal strength of the current one-dimensional radio time domain signal.

[0033] Furthermore, after determining the noise signal strength, an adaptive filter is constructed based on the noise signal strength. In this embodiment, a Gaussian filter is used as an example to construct the adaptive filter based on the noise signal strength, including: constructing a standard Gaussian filter, that is, a Gaussian filter with a standard deviation of 1, expressed by the formula: Then, the standard deviation σ of the target filter is determined based on the noise signal strength. Specifically, when the preset signal strength is high, for example, greater than a set signal strength threshold, it is determined that the current signal is subject to strong noise interference, and the standard deviation σ of the Gaussian filter is increased to improve the smoothness; when the preset signal strength is low, for example, less than the set signal strength threshold, the standard deviation σ is decreased to reduce the filtering strength, so as to avoid over-filtering of effective modulation features. Finally, by replacing the standard deviation of the standard Gaussian filter with the standard deviation of the target filter, the adaptive filter can be expressed as: .

[0034] Step S103: Use an adaptive filter to filter the one-dimensional radio time-domain signal to obtain a one-dimensional filtered signal.

[0035] Step S104: Perform a short-time discrete Fourier transform on the one-dimensional filtered signal to obtain a two-dimensional spectrum image. Use a convolutional neural network model to extract features and classify the two-dimensional spectrum image to determine the modulation category of the one-dimensional radio time-domain signal.

[0036] In this step, during the short-time discrete Fourier transform of the one-dimensional filtered signal, let w(·) represent a window function of length J, and let K be the window shift. The short-time discrete Fourier transform of the one-dimensional filtered signal can then be expressed as:

[0037] Where Y(f, m) are the DFT coefficients at the frequency domain index f and the discrete-time frame index m. for The sampled signal at time period n, It is the discrete frequency variable at point f. Then, a two-dimensional spectrum image is given. . yes The mixed time-frequency representation, i.e., the two-dimensional spectrum image.

[0038] In this step, the Hann window is used as the window function in the process of performing short-time discrete Fourier transform on the one-dimensional filtered signal. The frame length is set to 40 sampling points and the frame overlap rate is set to 90%. The original spectrum image is generated with a resolution of 200×200×3, and then scaled to the standard size of 100×100×3 by the nearest neighbor interpolation algorithm.

[0039] For further details, please refer to Figure 2 In this embodiment, the convolutional neural network model includes multiple convolutional layers 201, multiple max-pooling layers 202, densely connected layers 203, and a softmax layer 204. The multiple convolutional layers 201 are sequentially connected via max-pooling layers 202, and a unique max-pooling layer 202 is set between every two convolutional layers 201. The convolutional neural network model based on this structure performs feature extraction and classification of two-dimensional spectral images specifically by: using multiple convolutional layers and multiple max-pooling layers to extract deep features from the two-dimensional spectral image to obtain two-dimensional spectral features; using densely connected layers to perform dimensionality mapping on the two-dimensional spectral features to obtain a mapped image; and using a softmax layer to perform image classification on the mapped image, outputting the probability distribution of each modulation category.

[0040] The multiple convolutional layers scan and map features from the input data using convolutional kernels, extracting local features at different levels. Each convolutional layer includes a set of learnable kernels that perform convolution operations with the input data through a sliding window to generate a feature map. Multiple max-pooling layers downsample the feature maps output by the convolutional layers, reducing feature dimensionality. Within each pooling window, the max-pooling layer selects the maximum value as the output. The densely connected layers integrate the local features extracted by the preceding convolutional and pooling layers, performing high-level abstraction and non-linear combination to prepare for the final classification task. Linear transformations are performed using weights and biases, followed by non-linear mapping using activation functions (such as ReLU or Sigmoid). The softmax layer transforms the output of the densely connected layers into a probability distribution representing each class, ensuring that the sum of the probabilities of all classes is 1, thus achieving multi-class classification and obtaining the predicted probability for each class.

[0041] In this embodiment, the number of filters in the multiple convolutional layers 201 decreases sequentially along the data processing direction, and the pooling kernel size of the multiple max pooling layers is the same. For example, the number of filters used in the four convolutional layers is set to 64, 32, 12 and 8 respectively, and the pooling kernel size of the three max pooling layers is set to 2×2.

[0042] Furthermore, for the convolutional neural network model, this embodiment specifically trains the model using gradient descent. Gradient descent is an iterative optimization algorithm that adjusts parameters along the direction of fastest gradient descent of the loss function of the convolutional neural network model to gradually approach the local minimum of the function. The formula is expressed as: ,in, To the number of training samples, For correct labeling, For predicting labels, These are the model parameters for the convolutional neural network model. The parameters of the trained model. This is the loss function.

[0043] Furthermore, during the training process of the convolutional neural network model, the learning rate of the model is set to gradually increase with the increase of the number of training iterations. For example... Figure 3 The figure shows the change in the recognition accuracy of the convolutional neural network with the number of training iterations during the model training process of this application. The initial learning rate during the model training phase is set to 0.0005. After every 100 iterations of training, the learning rate is reduced to 1 / 10 of the original value. When the validation set loss value does not decrease for 15 consecutive iterations, the training is automatically terminated and the optimal model parameters are saved.

[0044] Compared with related technologies, the adaptive noise reduction signal modulation category identification method provided in this application directly acquires the waveform statistical features of the one-dimensional radio time-domain signal after acquiring it for modulation category identification. Then, based on these features, the noise signal intensity of the noise signal included in the one-dimensional radio time-domain signal is determined. An adaptive filter is adaptively constructed based on the noise signal intensity, and this filter is used to filter the one-dimensional radio time-domain signal to obtain a one-dimensional filtered signal. A short-time discrete Fourier transform is performed on the filtered signal to obtain a two-dimensional spectrum image. A convolutional neural network model is used to extract features and classify the two-dimensional spectrum image to determine the modulation category of the one-dimensional radio time-domain signal. Because this application constructs an adaptive filter corresponding to the waveform statistical features of each one-dimensional radio time-domain signal, the adaptive filter has a better filtering effect on the signal, resulting in a more accurate filtered signal with less residual noise. Consequently, the accuracy of the modulation category determined based on the filtered signal is also higher. This achieves adaptive noise reduction during wireless communication modulation identification, improving both the noise reduction effect and the modulation identification accuracy.

[0045] To better implement the adaptive noise reduction signal modulation category identification method in the embodiments of this application, based on the adaptive noise reduction signal modulation category identification method, correspondingly, as follows: Figure 4 As shown in the figure, this application embodiment also provides an adaptive noise reduction signal modulation category identification device, which includes: The signal acquisition module 401 is used to acquire one-dimensional radio time-domain signals and obtain the waveform statistical characteristics of the one-dimensional radio time-domain signals. The adaptive filter construction module 402 is used to determine the noise signal intensity of the noise signal in the one-dimensional radio time domain signal based on the waveform statistical characteristics, and to construct an adaptive filter based on the noise signal intensity. The signal filtering module 403 is used to filter a one-dimensional radio time-domain signal using an adaptive filter to obtain a one-dimensional filtered signal. The modulation category identification module 404 is used to perform short-time discrete Fourier transform on the one-dimensional filtered signal to obtain a two-dimensional spectrum image. The convolutional neural network model is used to extract features and classify the two-dimensional spectrum image to determine the modulation category of the one-dimensional radio time-domain signal.

[0046] The adaptive noise reduction signal modulation category identification device provided in the above embodiments can realize the technical solutions described in the above adaptive noise reduction signal modulation category identification method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above adaptive noise reduction signal modulation category identification method embodiments, which will not be repeated here.

[0047] Please refer to Figure 5 This application also provides an electronic device 500. The electronic device 500 includes: an input serial bus 501, an output serial bus 502, a memory 503, and a processor 504.

[0048] The system includes an input serial bus 501 for inputting a one-dimensional radio time-domain signal. A memory 503 stores the adaptive noise reduction signal modulation category identification program and the one-dimensional radio time-domain signal input via the serial bus 501. A processor 504, coupled to the memory 503, executes the adaptive noise reduction signal modulation category identification program stored in the memory 503 to implement the steps in the adaptive noise reduction signal modulation category identification method provided in the aforementioned method embodiments, determining the modulation category of the one-dimensional radio time-domain signal. An output serial bus 502 outputs the modulation category of the one-dimensional radio time-domain signal to a result display device.

[0049] In some embodiments, processor 504 may be a central processing unit (CPU), microprocessor, or other data processing chip for running program code stored in memory 503 or processing data. In some embodiments, processor 504 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 504 may be local or remote. In some embodiments, processor 504 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, inter-cloud, multi-cloud, or any combination thereof.

[0050] In one embodiment, when the processor 504 executes the adaptive noise reduction signal modulation category identification program in the memory 503, the following steps can be implemented: Acquire one-dimensional radio time-domain signals and obtain the waveform statistical characteristics of the one-dimensional radio time-domain signals; The noise signal intensity in a one-dimensional radio time-domain signal is determined based on waveform statistical characteristics, and an adaptive filter is constructed based on the noise signal intensity. One-dimensional filtered signals are obtained by using an adaptive filter to filter one-dimensional radio time-domain signals; A short-time discrete Fourier transform is performed on the one-dimensional filtered signal to obtain a two-dimensional spectrum image. A convolutional neural network model is then used to extract features from and classify the two-dimensional spectrum image to determine the modulation category of the one-dimensional radio time-domain signal.

[0051] Please refer to Figure 6 This application also provides an adaptive noise reduction signal modulation category identification system, including: a signal acquisition device 601, an adaptive noise reduction signal modulation category identification device 602 connected to the signal acquisition device 601 as provided in the foregoing embodiments, and a result display device 603 connected to the adaptive noise reduction signal modulation category identification device 602. Signal acquisition device 601 is used to acquire one-dimensional radio time-domain signals; The adaptive noise reduction signal modulation category identification device 602 is used to perform the steps in the adaptive noise reduction signal modulation category identification method provided in the foregoing embodiments based on a one-dimensional radio time-domain signal, so as to determine the modulation category of the one-dimensional radio time-domain signal; The result display device 603 is used to display the modulation category.

[0052] Please refer to Figure 7 , Figure 7This paper presents the relationship between the recognition accuracy of SCNN (Spatial CNN), SCNN2 (the adaptive noise reduction signal modulation category recognition method provided in this application), CNNR-IQ (CNN-based Recognition with In-phase and Quadrature), and CNNR-IQFOC (CNN-based Recognition with In-phase, Quadrature, and Fourth-order Cumulants) and the signal-to-noise ratio (SNR) of a one-dimensional radio time-domain signal. Figure 7 The results show that when the SNR is below -4dB, SCNN2's recognition accuracy is about 2% higher than the SCNN method. When the SNR is between -4dB and 16dB, SCNN2's recognition accuracy is about 10% higher. Furthermore, both SCNN2 and CNNR-IQFOC have better recognition accuracy than CNNR-IQ. When the SNR is below -4dB, SCNN2's recognition performance is about 5% higher than CNNR-IQ; when the SNR is above -2dB, SCNN2's recognition performance is about 15% to 20% higher than CNNR-IQ. When the SNR is between 2dB and 18dB, SCNN2's accuracy is about 8% higher than CNNR-IQFOC. At 2dB, its performance is better than CNNR-IQFOC.

[0053] Please refer to Figure 8 , Figure 8 This paper describes the relationship between the recognition accuracy of SCNN2, AF-CNN (Attention-based Convolutional Neural Network), and SCF-CNN (Spectral Correlation Function-Aided Convolutional Neural Network) and the signal-to-noise ratio (SNR) of a one-dimensional radio time-domain signal. From... Figure 8 This shows that SCNN2 outperforms AF-CNN and SCF-CNN. Specifically, SCNN2's accuracy is about 15% higher than SCF-CNN. When SNR is... When the SNR is above 2dB, the accuracy is about 20% higher than AF-CNN. When the SNR is below At 8dB, the recognition performance of SCNN2 and SCF-CNN is very similar.

[0054] Please refer to Figure 9 , Figure 10 , Figure 9 and Figure 10 This diagram illustrates the computational complexity of SCNN2, CNNR-IQ, and CNNR-IQFOC. Specifically, this application evaluates the computational complexity of SCNN2, CNNR-IQ, and CNNR-IQFOC by comparing their memory consumption, number of learning parameters, and average training time per signal. Figure 9 , Figure 10 As shown, SCNN2 has fewer learning parameters than the reference method because it uses pooling access, resulting in fewer filters per convolutional layer. In contrast, the CNNR-IQ network architecture does not include pooling layers, and CNNR-IQFOC uses more filters per convolutional layer. Furthermore, SCNN2 has a longer training time than CNNR-IQ but a shorter time than CNNR-IQFOC.

[0055] The above provides a detailed description of the adaptive noise reduction signal modulation category identification method, apparatus, system, and electronic device provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for adaptive noise reduction and signal modulation category identification, characterized in that, include: Acquire one-dimensional radio time-domain signals and obtain the waveform statistical characteristics of the one-dimensional radio time-domain signals; Based on the waveform statistical characteristics, the noise signal intensity of the noise signal in the one-dimensional radio time domain signal is determined, and an adaptive filter is constructed based on the noise signal intensity. The one-dimensional radio time-domain signal is filtered using the adaptive filter to obtain a one-dimensional filtered signal. A short-time discrete Fourier transform is performed on the one-dimensional filtered signal to obtain a two-dimensional spectrum image. A convolutional neural network model is then used to extract features and classify the two-dimensional spectrum image to determine the modulation category of the one-dimensional radio time-domain signal.

2. The adaptive noise reduction signal modulation category identification method according to claim 1, characterized in that, The adaptive filter is a Gaussian filter; the process of constructing the adaptive filter based on the noise signal intensity includes: Construct a standard Gaussian filter; The standard deviation of the target filter is determined based on the noise signal intensity. Replace the standard deviation of the standard Gaussian filter with the standard deviation of the target filter.

3. The adaptive noise reduction signal modulation category identification method according to claim 1, characterized in that, Also includes: It provides multiple preset index ranges and preset signal strengths corresponding to each preset index range; Determining the noise signal intensity of the one-dimensional radio time-domain signal based on the waveform statistical characteristics includes: The waveform statistical features are weighted according to preset weighting coefficients to obtain noise evaluation indicators; The preset index range to which the noise evaluation index belongs is determined as the target index range, and the preset signal strength corresponding to the target index range is determined as the noise signal strength.

4. The adaptive noise reduction signal modulation category identification method according to claim 1, characterized in that, The convolutional neural network model includes multiple convolutional layers, multiple max pooling layers, densely connected layers, and softmax layers. The multiple convolutional layers are connected sequentially through the max pooling layers, and a unique max pooling layer is set between every two convolutional layers. The step of using a convolutional neural network model to extract features and classify the two-dimensional spectral image includes: The multiple convolutional layers and the multiple max pooling layers are used to extract depth features from the two-dimensional spectral image to obtain two-dimensional spectral features. The densely connected layer is used to perform dimensional mapping on the two-dimensional spectral features to obtain a mapped image; The Softmax layer is used to classify the mapped image, and the probability distribution of each modulation category is output.

5. The adaptive noise reduction signal modulation category identification method according to claim 4, characterized in that, Also includes: The convolutional neural network model is trained using the gradient descent method.

6. The adaptive noise reduction signal modulation category identification method according to claim 5, characterized in that, Also includes: The learning rate of the convolutional neural network model is set to gradually increase as the number of training iterations increases.

7. The adaptive noise reduction signal modulation category identification method according to claim 4, characterized in that, Also includes: The number of filters in the multiple convolutional layers decreases sequentially, and the pooling kernel size of the multiple max pooling layers is the same.

8. A signal modulation category identification device for adaptive noise reduction, characterized in that, include: A signal acquisition module is used to acquire one-dimensional radio time-domain signals and obtain the waveform statistical characteristics of the one-dimensional radio time-domain signals. An adaptive filter construction module is used to determine the noise signal intensity of the noise signal in the one-dimensional radio time-domain signal based on the waveform statistical characteristics, and to construct an adaptive filter based on the noise signal intensity. A signal filtering module is used to filter the one-dimensional radio time-domain signal using the adaptive filter to obtain a one-dimensional filtered signal. The modulation category identification module is used to perform a short-time discrete Fourier transform on the one-dimensional filtered signal to obtain a two-dimensional spectrum image, and to use a convolutional neural network model to extract features and classify the two-dimensional spectrum image to determine the modulation category of the one-dimensional radio time-domain signal.

9. An electronic device, characterized in that, include: The system consists of an input serial bus, an output serial bus, a memory, and a processor. The input serial bus is used to input one-dimensional radio time-domain signals; The memory is used to store the program and the one-dimensional radio time-domain signal; The processor, coupled to the memory, is used to execute the program stored in the memory based on the one-dimensional radio time-domain signal to implement the steps in the adaptive noise reduction signal modulation category identification method according to any one of claims 1 to 7, so as to determine the modulation category of the one-dimensional radio time-domain signal. The output serial bus is used to output the modulation category of the one-dimensional radio time-domain signal to the result display device.

10. An adaptive noise reduction signal modulation category recognition system, characterized in that, include: A signal acquisition device, a signal modulation category identification device for adaptive noise reduction as described in claim 8, and a result display device connected to the signal modulation category identification device for adaptive noise reduction. The signal acquisition device is used to acquire one-dimensional radio time-domain signals; The adaptive noise reduction signal modulation category identification device is used to perform the steps of the adaptive noise reduction signal modulation category identification method as described in any one of claims 1 to 7 based on the one-dimensional radio time-domain signal, so as to determine the modulation category of the one-dimensional radio time-domain signal. The result display device is used to display the modulation category.

Images