Complex environment radio communication interference noise reduction method based on recurrent neural network

Through the recurrent neural network combined with Fourier transform technology, the radio communication system's low recognition accuracy and poor real-time performance of signal interference processing in complex environments is solved, efficient noise suppression and signal integrity preservation are achieved, and signal noise is significantly improved and communication quality.

CN118944782APending Publication Date: 2024-11-12MICROPHASE TECHNOLOGY (SHANGHAI) CO LTD
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410993852.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

When existing radio communication systems face signal interference and noise processing in complex environments, they have low recognition accuracy and poor real-time performance, and traditional methods often lead to signal distortion and spectrum loss.

Method used

Using a recurrent neural network-based method, noise estimation and noise reduction are performed through short-term Fourier transform and inverse short-term Fourier transform, combining gated cyclic units and long-term memory networks, which significantly improves the signal-to-noise ratio and retains it. signal integrity.

Benefits of technology

It significantly improves the signal-to-noise ratio, reduces signal distortion, improves the recognition accuracy and real-time processing capabilities of complex signals, and is suitable for a variety of radio communication scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118944782A_ABST
    Figure CN118944782A_ABST
Patent Text Reader

Abstract

The invention relates to a complex environment radio communication interference noise reduction method based on a recurrent neural network, and the method comprises the steps: obtaining a radio communication signal containing interference, carrying out the short-time Fourier transform, obtaining an amplitude spectrum and a phase spectrum of the signal, and carrying out the noise estimation and noise reduction of the amplitude spectrum through a recurrent neural network model. Then, inverse short-time Fourier transform is used to reconstruct the denoised time domain signal. By introducing a multi-layer gating circulation unit and a multi-gating structure, the method can effectively capture time sequence characteristics of signals, reduce noise interference in a complex environment and improve communication signal quality. Experimental results show that the signal-to-noise ratio of signals can be remarkably increased in a complex environment, and the method has a wide application prospect.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The method for performing noise reduction processing on a radio signal containing interference under a complex environment of the present invention is applied to the technical field of radio communication signals. Background Art

[0002] Existing radio communication systems often face strong signal interference and noise problems in complex environments. Although traditional filter technology and linear prediction algorithms can suppress noise to a certain extent, they have limitations in dealing with complex interference and maintaining signal integrity. Especially when processing high-frequency signals, traditional methods often lead to signal distortion or spectrum loss, which limits their application in real-time communication and spectrum management.

[0003] The accuracy and efficiency of traditional methods in identifying and processing burst signals, non-burst signals, and calibration signals in complex environments also need to be improved. Therefore, the limitations of existing technologies have prompted researchers to seek a new solution that can more effectively deal with signal interference and noise problems in complex environments, while improving the accuracy and real-time performance of signal processing. Summary of the invention

[0004] To solve the above technical problems, the present invention provides a complex environment radio communication interference noise reduction method based on recurrent neural network, which uses recurrent neural network to more effectively capture the timing characteristics of the signal, and more accurately estimate and suppress noise and interference, thereby significantly improving the signal-to-noise ratio. The present invention effectively solves the problems of low interference identification accuracy and poor real-time performance faced by traditional radio communications in complex environments.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for reducing interference noise in complex environment radio communication based on recurrent neural network, comprising the following steps:

[0007] (1) Collecting radio signal data containing interference;

[0008] (2) preprocessing the radio signal data containing interference collected in step (1) using a short-time Fourier transform method to obtain short-time Fourier transform spectrum data;

[0009] (3) Inputting the short-time Fourier transform spectrum data obtained in step (2) into the recurrent neural network model, Perform noise estimation and noise reduction processing to obtain the denoised signal spectrum;

[0010] (4) Using an inverse short-time Fourier transform method, the denoised signal spectrum obtained in step (3) is subjected to an inverse short-time Fourier transform to obtain a denoised radio signal.

[0011] Preferably, in the step (2), when preprocessing the collected radio signal data, the following steps are included:

[0012] (2-1) Divide the signal into frames;

[0013] (2-2) Perform short-time Fourier transform on each frame signal to obtain short-time Fourier transform spectrum data.

[0014] Preferably, in the step (2-1), the frame division adopts 256-point short-time Fourier transform.

[0015] Preferably, in the step (2-2), the short-time Fourier transform spectrum data obtained includes an amplitude spectrum and a phase spectrum of the radio signal.

[0016] Preferably, in step (3), the recurrent neural network model adopts a gated recurrent unit or a long short-term memory network structure.

[0017] Preferably, in step (3), the recurrent neural network model includes three parts: a noise tracking part, a signal estimation part and a filtering part, and each part is updated through a gating mechanism.

[0018] Preferably, in step (3), the training data of the recurrent neural network model includes pairs of clean signals and signals with added noise.

[0019] Preferably, the updating formula of the noise tracking part is:

[0020]

[0021] Where N(t) is the noise spectrum estimate at time t, N(t-1) is the noise spectrum estimate at time t-1, X(t) is the received signal spectrum at time t, g1(t) and g2(t) are gating functions, is the convolution symbol.

[0022] Preferably, the signal-to-noise ratio (SNR) of the training data ranges from -5dB to 15dB, with an interval of 5dB.

[0023] Preferably, the gating function is implemented by a recurrent neural network and optimized using a complex spectrum to perform noise reduction on the spectrum data to improve noise reduction performance and eliminate data noise interference.

[0024] Preferably, the complex spectrum optimization adopts the L2-norm loss function:

[0025]

[0026] in: is the estimated clean signal spectrum at time t, and Y(t) is the actual clean signal spectrum at time t.

[0027] Preferably, the dimensions of the input layer and output layer of the recurrent neural network model are 516 and 129 respectively.

[0028] Preferably, the recurrent neural network model has a depth of 3 to 4 layers and a width of 32 to 258 neuron units.

[0029] Preferably, the activation function of the recurrent neural network model is a logistic function, and the learning rate is 0.03.

[0030] Preferably, in step (4), when performing inverse short-time Fourier transform on the denoised spectrum, the following steps are included:

[0031] (4-1) Perform inverse Fourier transform on each frame of denoised spectrum data to obtain a time domain signal;

[0032] (4-2) Overlap and add the time domain signals to restore the complete radio signal.

[0033] Preferably, in step (4), after performing an inverse short-time Fourier transform on the denoised spectrum to obtain a denoised radio signal, the signal distortion ratio is used as an evaluation index to evaluate the noise reduction performance of the complex environment radio communication interference noise reduction method.

[0034] Compared with the prior art, the present invention has the following obvious outstanding substantive features and significant advantages:

[0035] 1. Existing radio communication interference noise reduction methods, such as traditional filter technology and linear prediction algorithm, can suppress noise to a certain extent, but the effect is limited when facing strong interference in complex environments; the method of the present invention adopts recurrent neural network, especially long short-term memory network and gated recurrent unit, which can more effectively capture the timing characteristics of the signal, estimate and suppress noise and interference more accurately, thereby significantly improving the signal-to-noise ratio;

[0036] 2. Traditional noise reduction methods often cause distortion of the target signal while filtering out noise, especially serious loss of high-frequency components; however, the method of the present invention can better preserve the details and integrity of the signal by introducing a loss function of complex spectrum optimization. Experimental results show that after using the method of the present invention for noise reduction processing, the spectral structure of the signal is better preserved and the distortion is significantly reduced;

[0037] 3. The multi-head self-attention mechanism and the self-attention calculation within and between windows adopted by the method of the present invention enable the model to better capture the time-frequency characteristics of spectrum signals when processing spectrum data, and improve the recognition accuracy of burst signals, non-burst signals and calibration signals. This advantage is particularly important in real-time spectrum monitoring and dynamic spectrum management;

[0038] 4. The method of the present invention utilizes the gating mechanism and multi-layer neural network structure of the recurrent neural network, and has high computational efficiency and real-time performance when processing complex signals; compared with the high computational complexity and latency of traditional methods, the method of the present invention can complete the processing of large-scale data in a shorter time, meeting the real-time requirements in actual radio communication systems;

[0039] 5. The recurrent neural network model of the method of the present invention is trained by a large amount of labeled data and has strong generalization ability. force, can adapt to different types of noise and interference environments; the method of the present invention is not only applicable to vehicle-mounted communications, mobile communications Xindengchang The method of the present invention can be applied to common radio communication scenarios and more complex special environments such as military communication and emergency communication, and has broad application prospects. The method of the present invention can be easily integrated with the existing radio communication system and used as an independent noise reduction module. At the same time, due to the modular design, the whole system has good scalability. In the future, the structure and parameters of the neural network can be adjusted and optimized as needed to cope with the ever-changing noise and interference challenges. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The present invention is a flowchart of a method for reducing radio communication interference noise according to a preferred embodiment of the present invention.

[0041] Figure 2 This is a topological diagram of the neural network used in the radio communication interference noise reduction method in the preferred embodiment of the present invention.

[0042] Figure 3 This is a spectrum diagram before noise reduction in the radio communication interference noise reduction method according to a preferred embodiment of the present invention.

[0043] Figure 4 This is a spectrum diagram after noise reduction in the radio communication interference noise reduction method according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0044] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0045] Embodiment 1:

[0046] See also Figure 1 , a complex environment radio communication interference denoising method based on recurrent neural network, characterized in that it includes the following steps:

[0047] (1) Collecting radio signal data containing interference;

[0048] (2) preprocessing the radio signal data containing interference collected in step (1) using a short-time Fourier transform method to obtain short-time Fourier transform spectrum data;

[0049] (3) inputting the short-time Fourier transform spectrum data obtained in step (2) into a recurrent neural network model to perform noise estimation and noise reduction processing to obtain a denoised signal spectrum;

[0050] (4) Using an inverse short-time Fourier transform method, the denoised signal spectrum obtained in step (3) is subjected to an inverse short-time Fourier transform to obtain a denoised radio signal.

[0051] The present embodiment is a complex environment radio communication interference denoising method based on a recurrent neural network, which uses a recurrent neural network to more effectively capture the timing characteristics of the signal, and to more accurately estimate and suppress noise and interference, thereby significantly improving the signal-to-noise ratio. The present invention effectively solves the problems of low interference identification accuracy and poor real-time performance faced by traditional radio communications in complex environments.

[0052] Embodiment 2:

[0053] This embodiment is basically the same as the first embodiment, with the following special features:

[0054] like Figure 1 and Figure 2 As shown, in the step (2), when the collected radio signal data is preprocessed, the following steps are included:

[0055] (2-1) Divide the signal into frames;

[0056] (2-2) Perform short-time Fourier transform on each frame signal to obtain short-time Fourier transform spectrum data.

[0057] In the step (2-1), the frame division adopts a 256-point short-time Fourier transform;

[0058] In the step (2-2), the short-time Fourier transform spectrum data obtained includes the amplitude spectrum and phase spectrum of the radio signal.

[0059] In step (3), the recurrent neural network model adopts a gated recurrent unit or a long short-term memory network structure.

[0060] In step (3), the recurrent neural network model includes three parts: a noise tracking part, a signal estimation part and a filtering part, and each part is updated through a gating mechanism.

[0061] In step (3), the training data of the recurrent neural network model includes pairs of clean signals and signals with added noise.

[0062] The update formula of the noise tracking part is:

[0063]

[0064] Where N(t) is the noise spectrum estimate at time t, N(t-1) is the noise spectrum estimate at time t-1, X(t) is the received signal spectrum at time t, g1(t) and g2(t) are gating functions, is the convolution symbol.

[0065] The signal-to-noise ratio (SNR) of the training data ranges from -5dB to 15dB, with an interval of 5dB.

[0066] The gating function is implemented by a recurrent neural network and optimized using a complex spectrum to perform noise reduction on the spectrum data.

[0067] The complex spectrum optimization adopts the L2-norm loss function:

[0068]

[0069] in: is the estimated clean signal spectrum at time t, and Y(t) is the actual clean signal spectrum at time t.

[0070] The dimensions of the input layer and output layer of the recurrent neural network model are 516 and 129 respectively.

[0071] The recurrent neural network model has a depth of 3 to 4 layers and a width of 32 to 258 neuron units.

[0072] The activation function of the recurrent neural network model is a logistic function, and the learning rate is 0.03.

[0073] In the step (4), when the denoised spectrum is subjected to inverse short-time Fourier transform, the following steps are included:

[0074] (4-1) Perform inverse Fourier transform on each frame of denoised spectrum data to obtain a time domain signal;

[0075] (4-2) Overlap and add the time domain signals to restore the complete radio signal.

[0076] In the step (4), after performing an inverse short-time Fourier transform on the denoised spectrum to obtain a denoised radio signal, the signal distortion ratio is used as an evaluation index to evaluate the denoising performance of the complex environment radio communication interference denoising method.

[0077] This embodiment is based on a complex environment radio communication interference denoising method of a recurrent neural network. By acquiring a radio communication signal containing interference, a short-time Fourier transform is performed to obtain the amplitude spectrum and phase spectrum of the signal, and a recurrent neural network model is used to perform noise estimation and denoising on the amplitude spectrum. Subsequently, an inverse short-time Fourier transform is used to reconstruct the denoised time domain signal. By introducing a multi-layer gated cyclic unit and a multiple gated structure, this method can effectively capture the timing characteristics of the signal, reduce noise interference in a complex environment, and improve the quality of the communication signal. Experimental results show that the present invention can significantly improve the signal-to-noise ratio of the signal in a complex environment and has broad application prospects.

[0078] Embodiment three:

[0079] This embodiment is basically the same as the above embodiment, with the following special features:

[0080] The complex environment radio communication interference denoising method based on recurrent neural network in this embodiment aims to solve the problem that the existing radio communication systems and methods are interfered in complex noise environments, resulting in reduced communication quality. By introducing recurrent neural network technology and taking advantage of its advantages in processing time series data and noise estimation, effective noise reduction of communication signals is achieved. The specific implementation steps of the method and its corresponding technical details are described in detail below:

[0081] like Figure 1 As shown, it shows a flowchart of the radio communication interference noise reduction method described in this embodiment, and the method includes the following steps: collecting radio signal data; preprocessing the collected radio signal data to obtain a short-time Fourier transform spectrum; inputting the spectrum data into a recurrent neural network model for noise reduction processing to obtain a denoised spectrum; performing an inverse short-time Fourier transform on the denoised spectrum to obtain a denoised radio signal.

[0082] like Figure 2 As shown in FIG. 1 , it shows the topology of the neural network used in the radio communication interference denoising method of the present invention. The long short-term memory network uses its internal gating mechanism to process time series data, that is, continuous signal samples. In radio communication, these samples may contain useful signals and various interferences, such as noise and interference from other radio equipment. The long short-term memory network receives the input signal X at each time step. t , through a series of linear transformations and nonlinear activations, such as ReLu layers, and further processing by network layers, to learn and identify interference patterns in the signal. Through training, the network learns how to distinguish between signals and interference, and gradually adjusts its internal state, and finally outputs Y tThe effect of interference is reduced or eliminated, thereby providing a clearer communication signal. This approach is particularly suitable for dynamically changing interference environments because it can adapt and learn the characteristics of interference that change over time.

[0083] The radio communication interference noise reduction method specifically comprises the following steps:

[0084] Get a radio communication signal that contains interference:

[0085] First, a radio communication signal containing noise interference is acquired through a radio receiver device. When acquiring the signal, it is necessary to ensure that the sampling frequency is appropriate to meet the signal bandwidth requirements. After the signal is acquired, it is divided into several frames, each containing a fixed number of sampling points, such as 256 points. This frame division method helps to preserve the time and frequency characteristics of the signal in the subsequent short-time Fourier transform process.

[0086] Perform a short-time Fourier transform on each frame of signal to obtain the spectrum data of the signal. The short-time Fourier transform can analyze the signal simultaneously in time and frequency by performing a piecewise Fourier transform on the signal. In the specific implementation, the Hamming window function is used to perform windowing processing on each frame of signal, and then a Fourier transform is performed. The window function length is the same as the frame length, usually 256 points, and the frame shift is 128 points to ensure a 50% overlap between frames and enhance the smoothness of the spectrum. This process can be easily implemented through the short-time Fourier transform function in the numpy library.

[0087] The preprocessed spectrum data is prepared as the input of the recurrent neural network. After short-time Fourier transform processing, the amplitude spectrum and phase spectrum of the signal are obtained. The amplitude spectrum contains the energy information of the signal, while the phase spectrum records the phase change of the signal. Since the phase spectrum is very important in reconstructing the signal, when performing subsequent noise reduction processing, usually only the amplitude spectrum is processed, and the original phase spectrum is saved for inverse transformation. Finally, the amplitude spectrum is used as the input data of the recurrent neural network for noise estimation and noise reduction processing.

[0088] The relevant code is as follows:

[0089] import numpy as np

[0090] from scipy.signal import stft

[0091] defpreprocess_signal(signal,fs=16000,nperseg=256,noverlap=128):

[0092] """

[0093] Perform short-time Fourier transform on the radio signal to obtain spectrum data.

[0094] :param signal: input time domain signal

[0095] :param fs: sampling frequency

[0096] :param nperseg: number of points per frame

[0097] :param noverlap: The number of points of the frame overlap

[0098] :return:Amplitude spectrum and phase spectrum

[0099] """

[0100] _,_,Zxx=stft(signal,fs,nperseg=nperseg,noverlap=noverlap)

[0101] magnitude = np.abs(Zxx)

[0102] phase = np.angle(Zxx)

[0103] return magnitude,phase

[0104] Design a recurrent neural network with multiple layers of gated recurrent units. First, define the input layer of the network, whose dimensions are consistent with the dimensions of the amplitude spectrum after the short-time Fourier transform. Next, add three layers of gated recurrent unit layers, each containing 128 neurons. The gated recurrent unit layer has high efficiency and performance when processing time series data, and can effectively capture the time series characteristics and noise patterns in the signal. The output of the gated recurrent unit layer is mapped to the target output space, that is, the denoised amplitude spectrum, through a fully connected layer.

[0105] Select appropriate activation function and loss function. The activation function determines the output form of the neuron, usually using sigmoid or relu function. In order to improve the noise reduction effect, the loss function selects the mean square error to minimize the difference between the network output and the amplitude spectrum of the clean signal. In the model compilation stage, select the Adam optimizer to optimize the network parameters. The Adam optimizer combines the advantages of momentum and adaptive learning rate adjustment to accelerate the convergence process.

[0106] Define the input and output dimensions and training strategy of the model. The dimension of the input layer is (number of frames, number of frequency points), and the dimension of the output layer is the same as the input layer. During the training process, the input amplitude spectrum and the target clean amplitude spectrum are used as training data pairs, and the network parameters are continuously adjusted through the back propagation algorithm to minimize the training error. The training data set uses the RadioML data set, in which the signal-to-noise ratio ranges from -5dB to 15dB. By adding various types of noise, the generalization ability of the model is enhanced. Noise types include white noise and environmental noise.

[0107] The relevant code is as follows:

[0108] import tensorflow as tf

[0109] from tensorflow.keras import layers,models

[0110] defbuild_rnn_model(input_shape):

[0111] """

[0112] Construct a recurrent neural network model for signal noise reduction.

[0113] :param input_shape: The shape of the input data

[0114] :return: Constructed model

[0115] """

[0116] model = models.Sequential()

[0117] model.add(layers.InputLayer(input_shape=input_shape))

[0118] model.add(layers.GRU(128,activation='relu',return_sequences=True))

[0119] model.add(layers.GRU(128,activation='relu',return_sequences=True))

[0120] model.add(layers.GRU(128,activation='relu'))

[0121] model.add(layers.Dense(input_shape[-1],activation='linear'))

[0122] model.compile(optimizer='adam',loss='mean_squared_error')

[0123] return model

[0124] input_shape = (None, 129) # Assume the number of frequency points is 129

[0125] model=build_rnn_model(input_shape)

[0126] Dataset preparation and data enhancement. The RadioML dataset is used for model training. A portion of the data is selected as the training set and the other portion as the validation set. In order to improve the generalization ability of the model, noise of various types and intensities are added to the clean signals in the training set to generate noisy signals. These noises can include white noise, pink noise, environmental noise, etc. The signal-to-noise ratio range is set to -5dB to 15dB to simulate radio communication scenarios in different noise environments.

[0127] Model training process. The preprocessed noisy amplitude spectrum is used as input, and the clean amplitude spectrum is used as the target output. The network weights are adjusted through the back-propagation algorithm. During the training process, the loss changes on the validation set are monitored to prevent overfitting. Usually, the training process requires dozens of rounds. In each round of training, the batch gradient descent method is used to batch the data, and the batch size is set to 32 to ensure the stability and efficiency of the training process. The convergence speed and final performance of the model are further optimized through the learning rate scheduling strategy, such as learning rate decay.

[0128] Model evaluation and preservation. During the training process, the model performance is evaluated using indicators such as signal distortion ratio and mean square error to ensure that the model can maintain good noise reduction effects in different noise environments. When the performance indicators on the validation set are no longer significantly improved, the training is terminated and the trained model parameters are saved for inference and noise reduction in actual applications.

[0129] The relevant code is as follows:

[0130] deftrain_rnn_model(model,X_train,y_train,X_val,y_val,epochs=50,batch_size=32):

[0131] """

[0132] Train a recurrent neural network model.

[0133] :param model: The constructed model

[0134] :param X_train: training data

[0135] :param y_train: training label

[0136] :param X_val: verification data

[0137] :param y_val: validation label

[0138] :param epochs: number of training rounds

[0139] :parambatch_size: batch size

[0140] :return: The trained model

[0141] """

[0142] model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),loss='mean_squared_error')

[0143] history=model.fit(X_train,y_train,epochs=epochs,batch_size=batch_size,validation_data=(X_val,y_val))

[0144] return model

[0145] #Assume X_train, y_train, X_val, y_val are ready

[0146] trained_model=train_rnn_model(model,X_train,y_train,X_val,y_val)

[0147] The preprocessed spectrum data is input into the trained recurrent neural network model for noise reduction. First, the preprocessed amplitude spectrum and phase spectrum are read, and the amplitude spectrum is used as the model input for forward propagation to obtain the noise-reduced amplitude spectrum output. In this process, the parameters learned by the model during training are used for noise estimation and noise reduction. The output of the model is a clean amplitude spectrum after processing.

[0148] Perform an inverse short-time Fourier transform on the denoised spectrum data to restore the time domain signal. Combine the denoised amplitude spectrum with the original phase spectrum and perform an inverse short-time Fourier transform. The inverse transform process is the opposite of the short-time Fourier transform, which converts the frequency domain signal back to the time domain signal. To ensure the accuracy of the reconstructed signal, the same window function and frame shift settings as the short-time Fourier transform are used in the inverse transform process. Finally, the reconstructed time domain signal is the radio communication signal after denoising.

[0149] Figure 3 This is the spectrum diagram of the radio communication interference noise reduction method of this embodiment before noise reduction. In the spectrum diagram before noise reduction, the noise and interference contained in the original signal can be observed. These noises are presented at different frequencies, resulting in multiple peaks in the spectrum diagram or an increase in the background noise level. These interferences may come from electronic devices, environmental noise, or other radio signal sources. If the spectrum of the communication signal overlaps with these interferences, they will reduce the identifiability of the signal, thereby affecting the communication quality.

[0150] Figure 4 This is the spectrum diagram after noise reduction of the radio communication interference noise reduction method of this embodiment. The spectrum diagram after noise reduction should show the impact of the signal processing algorithm on the original signal. It can be seen that the noise reduction algorithm greatly reduces the background noise and interference peaks, making the useful communication signal clearer and more prominent in the spectrum. The improvement of the signal-to-noise ratio is an important indicator of the noise reduction effect. In the spectrum diagram after noise reduction, the main components of the communication signal are more obvious without noise interference, while the peaks of other non-target signals are effectively suppressed.

[0151] The radio communication noise reduction technology based on recurrent neural network in this embodiment shows great potential in various wireless communication scenarios with its advanced algorithm and flexible adaptability. This technology is not only applicable to traditional mobile communication networks, but also can effectively serve the rapidly developing field of Internet of Things and smart home systems to ensure the clarity and stability of signal transmission. Its design is highly modular and scalable, and can be easily integrated into the existing communication architecture to provide seamless noise reduction services.

[0152] The noise reduction method based on the recurrent neural network in this embodiment can also adapt to the changing spectrum environment, intelligently identify and suppress various interferences, thereby significantly improving the communication quality. It achieves rapid response to complex signal environments through the self-learning and self-adjustment capabilities of the deep learning model, providing users with an intelligent and efficient communication experience. In addition, the method can also perform self-optimization based on real-time feedback to ensure the best noise reduction effect under changing communication conditions, and has significant application prospects and important practical value.

[0153] The above describes the embodiments of the present invention in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Various changes can be made according to the purpose of the invention. Any changes, modifications, substitutions, combinations or simplifications made according to the spirit and principle of the technical solution of the present invention should be equivalent replacement methods. As long as they meet the purpose of the invention and do not deviate from the technical principles and inventive concepts of the present invention, they belong to the protection scope of the present invention.

Claims

1. A method for reducing interference in complex environment radio communication based on recurrent neural network, characterized in that: The following steps are involved: (1) Collecting radio signal data containing interference; (2) preprocessing the radio signal data containing interference collected in step (1) using a short-time Fourier transform method to obtain short-time Fourier transform spectrum data; (3) inputting the short-time Fourier transform spectrum data obtained in step (2) into a recurrent neural network model to perform noise estimation and noise reduction processing to obtain a denoised signal spectrum; (4) Using an inverse short-time Fourier transform method, the denoised signal spectrum obtained in step (3) is subjected to an inverse short-time Fourier transform to obtain a denoised radio signal.

2. The method for reducing interference of radio communication in complex environments based on recurrent neural network according to claim 1, characterized in that: In the step (2), when the collected radio signal data is preprocessed, the following steps are included: (2-1) Divide the signal into frames; (2-2) Perform short-time Fourier transform on each frame signal to obtain short-time Fourier transform spectrum data.

3. The method for reducing interference of radio communication in complex environments based on recurrent neural network according to claim 2 is characterized in that: In the step (2-1), the frame division adopts a 256-point short-time Fourier transform; Alternatively, in step (2-2), the short-time Fourier transform spectrum data obtained includes the amplitude spectrum and phase spectrum of the radio signal.

4. The method for reducing interference of radio communication in complex environments based on recurrent neural network according to claim 1, characterized in that: In the step (3), the recurrent neural network model adopts a gated recurrent unit or a long short-term memory network structure; Alternatively, in step (3), the recurrent neural network model includes three parts: a noise tracking part, a signal estimation part and a filtering part, and each part is updated through a gating mechanism. Alternatively, in step (3), the training data of the recurrent neural network model includes pairs of clean signals and signals with added noise.

5. The method for reducing interference of radio communication in complex environments based on recurrent neural network according to claim 4 is characterized in that: The update formula of the noise tracking part is: Where N(t) is the noise spectrum estimate at time t, N(t-1) is the noise spectrum estimate at time t-1, X(t) is the received signal spectrum at time t, g1(t) and g2(t) are gating functions, is the convolution symbol; Alternatively, the signal-to-noise ratio (SNR) of the training data ranges from -5 dB to 15 dB, with an interval of 5 dB.

6. The method for reducing interference noise in complex environment radio communication based on recurrent neural network according to claim 5, characterized in that: The gating function is implemented by a recurrent neural network and optimized using a complex spectrum to perform noise reduction on the spectrum data.

7. The method for reducing interference in complex environment radio communication based on recurrent neural network according to claim 6, characterized in that: The complex spectrum optimization adopts the L2-norm loss function: in: is the estimated clean signal spectrum at time t, and Y(t) is the actual clean signal spectrum at time t.

8. The method for reducing interference of radio communication in complex environments based on recurrent neural network according to claim 7, characterized in that: The dimensions of the input layer and output layer of the recurrent neural network model are 516 and 129 respectively; Alternatively, the recurrent neural network model has a depth of 3 to 4 layers and a width of 32 to 258 neuron units; Alternatively, the activation function of the recurrent neural network model is a logistic function, and the learning rate is 0.

03.

9. The method for reducing interference noise in complex environment radio communication based on recurrent neural network according to claim 1, characterized in that: In the step (4), when the denoised spectrum is subjected to inverse short-time Fourier transform, the following steps are included: (4-1) Perform inverse Fourier transform on each frame of denoised spectrum data to obtain a time domain signal; (4-2) Overlap and add the time domain signals to restore the complete radio signal.

10. The complex environment radio communication interference denoising method based on recurrent neural network according to claim 1, characterized in that: In the step (4), after performing an inverse short-time Fourier transform on the denoised spectrum to obtain a denoised radio signal, the signal distortion ratio is used as an evaluation index to evaluate the denoising performance of the complex environment radio communication interference denoising method.

Citation Information

Cited By

  • Weak component extraction method of spectrum aliasing wireless electromagnetic signal

    CN121030310A

  • Neural network magnetic anomaly signal denoising method and system combined with weak supervised learning

    CN121069508A

  • Neural network magnetic anomaly signal denoising method and system combined with weak supervision learning

    CN121069508B