Wireless signal interference detection and separation method based on autoencoder signal reconstruction

Abstract

The present invention discloses a wireless signal interference detection and separation method based on autoencoder signal reconstruction, comprising the following steps: S1: constructing a signal reconstruction model based on an autoencoder; S2: using a non-interference signal as a training data set, inputting the signal reconstruction model based on the autoencoder for training, and obtaining a first reconstructed signal output by the signal reconstruction model and a trained signal reconstruction model under the condition that the loss function is less than 0.1%; S3: using the difference between the non-interference signal and the first reconstructed signal as a positive sample to train a binary classifier, and the binary classifier will be used for subsequent interference detection and judgment; S4: inputting an interference signal into the trained signal reconstruction model, outputting a second reconstructed signal, using the difference between the interference signal and the second reconstructed signal as a negative sample, and inputting the binary classifier obtained in step S3 for testing; S5: if the test achieves a preset effect, the training is completed, and the trained signal reconstruction model and classifier are used to perform interference detection; otherwise, the execution returns to step S2 to continue.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and more specifically, to a wireless signal interference detection and separation method based on autoencoder signal reconstruction. Background Art

[0002] Wireless communication solutions present a variety of challenges, and interference signals are a common and critical issue. When transmitting data through a channel, signals carrying the same frequency can easily be superimposed by interference signals. Directly decoding the received signal in this situation can lead to numerous errors, severely impacting communication. To ensure the quality of wireless communication, interference detection is essential.

[0003] In traditional communications, interference signals are typically detected using time-frequency transforms to extract features or by performing cyclostationary analysis on the signal. However, in real-world communication environments, some interference signals have low power and the same frequency as the original signal, making them difficult to detect. Furthermore, these methods only detect the interference but fail to isolate it for further analysis and processing. Furthermore, due to the complexity and variability of interference signals, feature extraction-based interference detection methods are not universally applicable. Therefore, it is crucial to develop interference detection algorithms that are applicable to universal communication scenarios. Summary of the Invention

[0004] In order to solve the above problems in the prior art, the present invention provides a wireless signal interference detection and separation method based on autoencoder signal reconstruction, which can not only detect the presence of interference signals, but also realize the separation of interference signals.

[0005] In order to achieve the above-mentioned purpose of the present invention, the technical solutions adopted are as follows:

[0006] A method for detecting and separating wireless signal interference based on autoencoder signal reconstruction, the method comprising the following steps:

[0007] S1: Build a signal reconstruction model based on autoencoder;

[0008] S2: Use the interference-free signal as a training data set and input it into the signal reconstruction model based on the autoencoder for training. Under the condition that the loss function is less than 0.1%, obtain the first reconstructed signal output by the signal reconstruction model and the trained signal reconstruction model;

[0009] S3: Using the difference between the interference-free signal and the first reconstructed signal as a positive sample to train a binary classifier, which will be used for subsequent interference detection and judgment;

[0010] S4: Input the interference signal into the trained signal reconstruction model, output a second reconstructed signal, and use the difference between the interference signal and the second reconstructed signal as a negative sample, which is input into the binary classifier obtained in step S3 for testing;

[0011] S5: If the test achieves the preset effect, the training is completed, and the trained signal reconstruction model and classifier are used to perform interference detection; otherwise, the process returns to step S2 to continue.

[0012] Preferably, the signal reconstruction model includes an encoder and a decoder;

[0013] The encoder is constructed by stacking N layers of first LSTM networks. From the input end to the output end of the encoder, the output dimension of each layer of the first LSTM network gradually decreases, so that the signal is compressed into a low-dimensional signal feature by the encoder;

[0014] The decoder is constructed by stacking N layers of second LSTM network and 1 layer of Dense layer;

[0015] The output dimension of the second LSTM network gradually increases from the input end to the output end of the decoder, and the output dimension of the Dense layer is equal to the length of the input signal.

[0016] Furthermore, the first LSTM network has three layers, and the number of neurons thereof is 256, 250, and 128 respectively; correspondingly, the second LSTM network has three layers, and the number of neurons thereof is 128, 250, and 256 respectively.

[0017] Furthermore, the activation function in the signal reconstruction model adopts the tanh activation function, the optimization algorithm adopts the Adam algorithm, and the loss function adopts the mse function.

[0018] Furthermore, the interference-free signal includes one of an undisturbed QPSK signal, an undisturbed 16QAM signal, and an undisturbed FM signal.

[0019] Furthermore, in step S2, the interference-free signal is used as a label for training the signal reconstruction model, and the signal reconstruction model is trained. By calculating the loss function, the network parameters are continuously adjusted to continuously reduce the difference between the first reconstructed signal and the interference-free signal, so that the first reconstructed signal is close to the interference-free signal.

[0020] Furthermore, in step S3, a binary classifier is trained using multivariate features. Specifically, the difference between the interference-free signal and the first reconstructed signal is calculated, and a sliding time window method is adopted for the difference. The time window is slid with a step value of 1 to extract features of a length of 10 as a frame of data, which is input into the binary classifier for training.

[0021] Furthermore, the classifier adopts SVDD algorithm.

[0022] A computer system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0023] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above method.

[0024] The beneficial effects of the present invention are as follows:

[0025] The present invention can adapt to various modulated wireless signals and does not change due to changes in the interference type or the type of transmitted signal. It does not require artificial feature extraction for calculation, has strong adaptive learning capabilities, and has good effects in special cases, such as when the frequency of the interfering signal is the same as that of the interfered signal. Moreover, the present invention can not only detect the existence of an interfering signal, but also determine the specific location of the interfering signal, and can even separate the interfering signal and support its analysis and identification. The present invention provides a new approach to interference detection and separation, and has achieved excellent performance in a variety of signal and interference combination scenarios.

[0026] This paper designs a signal reconstruction model and training method for interference detection and isolation within wireless signals. As part of the interference detection model, the signal reconstruction model effectively reconstructs the original signal in the presence of noise and calculates the difference between the reconstructed signal and the input signal as a feature for interference detection.

[0027] The present invention can construct a signal reconstruction model for the original non-interference signal in a wireless signal scenario. Compared with traditional prediction models, it can more accurately reconstruct the interfered signal from the received signal, making the presence characteristics of the interference signal more obvious and achieving better detection effects under low signal-to-noise ratio conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of the method described in Example 1.

[0029] Figure 2 This is a structural diagram of the signal reconstruction model based on the autoencoder described in Example 1.

[0030] Figure 3 3 is a comparison diagram of the reconstructed signal and the original signal in Example 1.

[0031] Figure 4 This is a comparison diagram of the original transmitted signal and the reconstructed signal constellation in Example 1.

[0032] Figure 5 This is a reconstructed representation of the interference signal in Example 1.

[0033] Figure 6 This is a diagram showing the interference detection of the QAM-QPSK signal in Example 1. DETAILED DESCRIPTION

[0034] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] Example 1

[0036] like Figure 1 As shown, a wireless signal interference detection and separation method based on autoencoder signal reconstruction is provided, wherein the method comprises the following steps:

[0037] S1: Build a signal reconstruction model based on autoencoder;

[0038] S2: Use the interference-free signal as a training data set and input it into the signal reconstruction model based on the autoencoder for training. Under the condition that the loss function is less than 0.1%, obtain the first reconstructed signal output by the signal reconstruction model and the trained signal reconstruction model;

[0039] S3: Using the difference between the interference-free signal and the first reconstructed signal as a positive sample to train a binary classifier, which will be used for subsequent interference detection and judgment;

[0040] S4: Input the interference signal into the trained signal reconstruction model, output a second reconstructed signal, and use the difference between the interference signal and the second reconstructed signal as a negative sample, which is input into the binary classifier obtained in step S3 for testing;

[0041] S5: If the test achieves the preset effect, the training is completed, and the trained signal reconstruction model and classifier are used to perform interference detection; otherwise, the process returns to step S2 to continue.

[0042] In a specific embodiment, Figure 2 As shown, the signal reconstruction model includes an encoder and a decoder;

[0043] The encoder is constructed by stacking N layers of first LSTM networks. From the input end to the output end of the encoder, the output dimension of each layer of the first LSTM network gradually decreases, so that the signal is compressed into a low-dimensional signal feature by the encoder;

[0044] The decoder is constructed by stacking N layers of second LSTM network and 1 layer of Dense layer;

[0045] The output dimension of the second LSTM network gradually increases from the input end to the output end of the decoder, and the output dimension of the Dense layer is equal to the length of the input signal.

[0046] The first LSTM network has three layers, and the number of neurons is 256, 250, and 128 respectively; correspondingly, the second LSTM network has three layers, and the number of neurons is 128, 250, and 256 respectively.

[0047] In the signal reconstruction model described in this embodiment, the number of LSTM neurons in the encoder must gradually decrease, while the number of LSTM neurons in the decoder must gradually increase. This can achieve dimensionality reduction and then dimensionality increase of the data, thereby preventing the network from "memorizing" the original signal.

[0048] The activation function in the signal reconstruction model adopts the tanh activation function, the optimization algorithm adopts the Adam algorithm, and in order to reduce the reconstruction error, the loss function adopts the mse function.

[0049] This example uses an LSTM network because wireless signals are periodic time series. LSTM networks have memory and can reconstruct input signals based on the correlation between the previous and next signal sampling points.

[0050] In a specific embodiment, the training signal in the training data set can be a signal of any modulation mode, but it must be ensured that the training signal is not affected by interference. Therefore, the interference-free signal can be an undisturbed QPSK signal, an undisturbed 16QAM signal, or an undisturbed FM signal.

[0051] In a specific embodiment, in step S2, interference-free signals are input into the reconstruction model for training. Each input interference-free signal is compressed into low-dimensional signal features by an encoder, and then the compressed low-dimensional signal features are decoded to generate a reconstructed signal. Because the training dataset is randomly generated based on prior information such as the modulation method of the transmitted signal at the receiving end, noise-free signals can be obtained as labels for model training, while the training dataset itself is superimposed with noise.

[0052] When training the model, since the signals selected from the training set data are all interference-free signals, and the interference-free signals are selected as labels for signal reconstruction model training, the purpose of selecting interference-free signals as labels here is to allow the signal reconstruction model to ignore the influence of noise on the signal in the training set, so that the signal reconstruction model can focus on the reconstruction of the interference-free signal itself, thereby overcoming the interference of noise. By calculating the loss function and continuously adjusting the network parameters, the difference between the first reconstructed signal and the interference-free signal is continuously reduced, so that the first reconstructed signal is close to the interference-free signal. The first reconstructed signal obtained by training in this way can not only reconstruct the original interference-free transmitted signal, but also a large error will occur when reconstructing the interference signal. On the other hand, due to the adoption of a noise-free training method, a certain noise reduction effect is achieved. When the signal-to-noise ratio is relatively low, the influence of noise can be ignored, and the focus is on the reconstruction of the regular signal itself. At the same time, the reconstructed signal itself also has a high signal-to-noise ratio.

[0053] In a specific embodiment, step S3 uses multivariate features to train a binary classifier. Specifically, the difference between the interference-free signal and the first reconstructed signal is calculated, and a sliding time window method is adopted for the difference. The time window is slid with a step value of 1 to extract features of a length of 10 as a frame of data, which is input into the binary classifier for training.

[0054] The signal reconstruction model trained in this way not only utilizes the amplitude of the difference feature but also takes into account the correlation between sequences. This multivariate feature-based model can overcome the influence of random errors. Once the signal reconstruction model is trained, the difference feature between the interfering signal and its reconstructed signal can be calculated to test the effectiveness of the classifier. This classifier uses the SVDD algorithm.

[0055] After the signal reconstruction model is trained, its parameters are fixed and it becomes an offline network. At this time, the signal reconstruction model can be tested with an interference signal. For the test data, the difference between the signal before the noise is superimposed and the reconstructed signal is calculated to observe whether the difference features are significantly different between the interference part and the non-interference part.

[0056] In this embodiment, an interference signal is used to evaluate the signal reconstruction model and the SVDD algorithm. Figure 3 The output results of the signal reconstruction model are given. The figure compares the simulation results when SNR = 6db. The first figure is the input of the model, that is, the actual received signal, which contains Gaussian white noise; the second figure is the original transmitted signal corresponding to the received signal. This signal has not been transmitted through the channel, so it does not contain Gaussian white noise. Figure 3 The third figure in FIG is the output of the signal reconstruction model described in this embodiment, that is, the noise reduction signal obtained after the original signal is reconstructed by the signal reconstruction model. Figure 3Comparing the three figures in the figure shows that the signal reconstruction model reconstructs the original signal very close to the original noise-free signal, indicating that the signal reconstruction model has excellent signal noise reduction capabilities. This also demonstrates that the signal reconstruction model designed in this embodiment can better ignore the influence of Gaussian white noise, thereby accurately extracting the characteristics of the transmitted signal itself.

[0057] In order to further verify the accuracy of the reconstructed signal, we give the constellation diagram of the 16QAM modulated received signal after noise reduction and the constellation diagram of the reconstructed signal as follows: Figure 4 As shown. The left figure is the original transmission signal constellation diagram, and the right figure is the constellation diagram of the reconstructed signal obtained by reconstructing the received signal with a signal-to-noise ratio of 12db using the signal reconstruction model. Figure 4 It can be seen that the constellation diagram of the reconstructed signal is similar to the original noise-free signal, which shows that the information carried by the signal can be basically preserved completely after signal reconstruction, thus indirectly proving the feasibility of using the separated interference signal for detection.

[0058] like Figure 5 Figure 2 shows the difference between the reconstructed signal and the original signal in the presence of interference. In this embodiment, an interference signal with the same frequency as the original signal is superimposed at sampling points 100-200, 1100-1200, 2500-2500, and 4300-4400 (here the Signal to Interference plus Noise Ratio (SINR) is 4 dB). Figure 5In the figure, the first figure shows the input signal of the signal reconstruction model. This signal is the received signal, which is superimposed with noise and interference. Because the power of the interference signal is set relatively low, the location of the interference signal is not significantly different from the other components, making it difficult to determine the specific location of these interference signals by human observation alone. The second figure shows the reconstructed output signal of the signal reconstruction model for the input containing interference. Because the signal reconstruction model is trained entirely on normal signals, the reconstructed signal output by the signal reconstruction model is also the original interference-free signal. As can be seen from the figure, the reconstructed signal obtained after the signal reconstruction model is an ideal noise-reduced signal compared to the input signal. Therefore, from another perspective, the difference between the model's input and output can be regarded as a mixture of separated noise and interference signals. The third figure calculates the difference between the first two figures, successfully separating the original signal from the interference signal and noise. In the separated interference signal and noise, it can be seen that the interference signal is already quite obvious in the time domain alone. Therefore, considering that interference can be detected using anomaly detection methods and the location of the interference signal can be more clearly determined under such conditions, this figure demonstrates the feasibility of using extracted interference and noise features for interference detection.

[0059] Figure 6 This is the result of interference detection using the SVDD algorithm. It can be seen that the algorithm can better detect the specific location of the interference signal, verifying the feasibility of the algorithm in wireless signal interference detection.

[0060] Example 2

[0061] A computer system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following method steps are implemented:

[0062] S1: Build a signal reconstruction model based on autoencoder;

[0063] S2: Use the interference-free signal as a training data set and input it into the signal reconstruction model based on the autoencoder for training. Under the condition that the loss function is less than 0.1%, obtain the first reconstructed signal output by the signal reconstruction model and the trained signal reconstruction model;

[0064] S3: Using the difference between the interference-free signal and the first reconstructed signal as a positive sample to train a binary classifier, which will be used for subsequent interference detection and judgment;

[0065] S4: Input the interference signal into the trained signal reconstruction model, output a second reconstructed signal, and use the difference between the interference signal and the second reconstructed signal as a negative sample, which is input into the binary classifier obtained in step S3 for testing;

[0066] S5: If the test achieves the preset effect, the training is completed, and the trained signal reconstruction model and classifier are used to perform interference detection; otherwise, the process returns to step S2 to continue.

[0067] Example 3

[0068] A computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the following method steps are implemented:

[0069] S1: Build a signal reconstruction model based on autoencoder;

[0070] S2: Use the interference-free signal as a training data set and input it into the signal reconstruction model based on the autoencoder for training. Under the condition that the loss function is less than 0.1%, obtain the first reconstructed signal output by the signal reconstruction model and the trained signal reconstruction model;

[0071] S3: Using the difference between the interference-free signal and the first reconstructed signal as a positive sample to train a binary classifier, which will be used for subsequent interference detection and judgment;

[0072] S4: Input the interference signal into the trained signal reconstruction model, output a second reconstructed signal, and use the difference between the interference signal and the second reconstructed signal as a negative sample, which is input into the binary classifier obtained in step S3 for testing;

[0073] S5: If the test achieves the preset effect, the training is completed, and the trained signal reconstruction model and classifier are used to perform interference detection; otherwise, the process returns to step S2 to continue.

[0074] The various embodiments of the present invention can be combined arbitrarily to achieve different technical effects.

[0075] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in this application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive).

[0076] Those skilled in the art will appreciate that all or part of the process steps in the above-described method embodiments can be implemented by a computer program instructing the relevant hardware. The program can be stored in a computer-readable storage medium, and when executed, the program can include the process steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A wireless signal interference detection and separation method based on autoencoder signal reconstruction, characterized by: The method comprises the following steps: S1: Build a signal reconstruction model based on autoencoder; S2: Use the interference-free signal as a training data set and input it into the signal reconstruction model based on the autoencoder for training. Under the condition that the loss function is less than 0.1%, obtain the first reconstructed signal output by the signal reconstruction model and the trained signal reconstruction model; S3: Using the difference between the interference-free signal and the first reconstructed signal as a positive sample to train a binary classifier, which will be used for subsequent interference detection and judgment; S4: Input the interference signal into the trained signal reconstruction model, output a second reconstructed signal, and use the difference between the interference signal and the second reconstructed signal as a negative sample, which is input into the binary classifier obtained in step S3 for testing; S5: If the test achieves the preset effect, the training is completed and the trained signal reconstruction model and classifier are used to perform interference detection; otherwise, the process returns to step S2 to continue. Wherein, the signal reconstruction model includes an encoder and a decoder; The encoder is constructed by stacking N layers of first LSTM networks. From the input end to the output end of the encoder, the output dimension of each layer of the first LSTM network gradually decreases, so that the signal is compressed into a low-dimensional signal feature by the encoder; The decoder is constructed by stacking N layers of second LSTM network and 1 layer of Dense layer; The output dimension of the second LSTM network gradually increases from the input end to the output end of the decoder, and the output dimension of the Dense layer is equal to the length of the input signal.

2. The wireless signal interference detection and separation method based on autoencoder signal reconstruction according to claim 1, characterized in that: The first LSTM network has three layers, and the number of neurons is 256, 250, and 128 respectively; correspondingly, the second LSTM network has three layers, and the number of neurons is 128, 250, and 256 respectively.

3. The wireless signal interference detection and separation method based on autoencoder signal reconstruction according to claim 1, characterized in that: The activation function in the signal reconstruction model adopts the tanh activation function, the optimization algorithm adopts the Adam algorithm, and the loss function adopts the mse function.

4. The wireless signal interference detection and separation method based on autoencoder signal reconstruction according to any one of claims 1 to 3, characterized in that: The non-interference signal includes one of an undisturbed QPSK signal, an undisturbed 16QAM signal, and an undisturbed FM signal.

5. The wireless signal interference detection and separation method based on autoencoder signal reconstruction according to claim 4 is characterized in that: In step S2, the interference-free signal is used as a label for training the signal reconstruction model, and the signal reconstruction model is trained. By calculating the loss function, the network parameters are continuously adjusted to continuously reduce the difference between the first reconstructed signal and the interference-free signal, so that the first reconstructed signal is close to the interference-free signal.

6. The wireless signal interference detection and separation method based on autoencoder signal reconstruction according to claim 5, characterized in that: Step S3, use multivariate features to train a binary classifier. Specifically, calculate the difference between the interference-free signal and the first reconstructed signal, and use a sliding time window method for the difference. Slide the time window with a step value of 1 to extract features of a length of 10 as a frame of data, and input it into the binary classifier for training.

7. The wireless signal interference detection and separation method based on autoencoder signal reconstruction according to claim 6, characterized in that: The classifier adopts SVDD algorithm.

8. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images