A wireless communication signal analysis method and system

Abstract

The present invention relates to the technical field of communication signal recognition, and specifically to a method and system for analyzing wireless communication signals. The analysis method comprises the following steps: S1: receiving a wireless communication signal and preprocessing the signal; S2: modulating the wireless communication signal and constructing a UAD signal data set; S3: constructing a signal recognition model based on a CTNN neural network; S4: modulating the preprocessed wireless communication signal and inputting it into the signal recognition model, and outputting the corresponding analysis result. The present invention adopts an improved CTNN model, combines a convolutional layer with an encoder, improves the model's acquisition of local information, and improves the model's recognition performance. The modulation recognition method based on Transformer is more efficient, more in line with parallel distributed GPU computing, has higher recognition accuracy, and the convolutional neural network with an attention mechanism is more adaptive and scalable.

Description

Technical Field

[0001] The present invention relates to the technical field of communication signal recognition, and in particular to a method for analyzing wireless communication signals and a system for analyzing wireless communication signals. Background Art

[0002] The specific steps of the analysis method of wireless communication signals usually include the following steps: 1. Signal acquisition: obtain signal samples from wireless communication devices. 2. Preprocessing: perform denoising, filtering and other processing on signal samples for subsequent analysis. 3. Time-frequency domain analysis: use time-frequency analysis methods such as short-time Fourier transform or wavelet transform to decompose the signal to obtain the time-frequency distribution characteristics of the signal. 4. Feature extraction: extract parameters that affect communication quality from the time-frequency distribution characteristics, such as power spectrum, phase delay, etc. 5. Communication quality evaluation: evaluate the quality of wireless communication based on the extracted feature parameters.

[0003] The existing wireless communication signal analysis methods have the following defects: 1. There are relatively few image data sets in the public data set that can meet the signal detection requirements of the convolutional neural network training requirements, or the data volume is insufficient. 2. In complex environments or under strong interference, the analysis results are not accurate enough. 3. Lack of adaptability: The existing wireless communication signal analysis methods may be applicable to certain specific scenarios, but cannot be adaptively adjusted according to different scenarios. Summary of the invention

[0004] Based on this, it is necessary to propose a method and system for analyzing wireless communication signals in order to address the problems of low recognition accuracy and poor adaptability in existing analysis methods for wireless communication signals.

[0005] The present invention is implemented by the following technical solution: A method for analyzing wireless communication signals comprises the following steps:

[0006] S1: Receive wireless communication signals and pre-process the wireless communication signals.

[0007] S2: Collect wireless communication signal data sets, modulate the wireless communication signals in the data sets, and construct UAD signal data sets. The specific method is as follows:

[0008] S21: Using a modulator to modulate the wireless communication signal in the data set to obtain a modulated signal.

[0009] S22: Use a multiplier to multiply the modulated signal and the carrier signal to obtain a frequency hopping signal.

[0010] S23: Collect the frequency hopping signal through the Rayleigh channel and map it into a time domain signal image.

[0011] S3: Construct a signal recognition model based on the CTNN neural network, which is a convolutional transform neural network. The specific method is as follows:

[0012] S31: Get the encoder of the Transformer neural network.

[0013] S32: Add the positional encoding to the input embedding of the encoder to obtain an improved encoder.

[0014] S33: The time domain signal image is segmented into patches of fixed size and the linear embedding sequence of the patches is used as the input of the encoder.

[0015] S34: Add the improved encoder to the CTNN network model.

[0016] S4: Use the UAD signal data set to train the signal recognition model, retain the model parameters that meet the test accuracy, modulate the preprocessed wireless communication signal and input it into the signal recognition model, and output the corresponding analysis results.

[0017] The above analysis method adopts an improved CTNN model, combining the convolutional layer with the encoder to improve the model's acquisition of local information and the recognition performance of the model. The Transformer-based modulation recognition method is more efficient, more suitable for parallel distributed GPU computing, and has higher recognition accuracy. The convolutional neural network with an attention mechanism is more adaptive and scalable.

[0018] Furthermore, in step S1, the signal preprocessing method is specifically as follows:

[0019] S11: Signal sampling.

[0020] S12: Time domain filtering. Using FIR filter for time domain filtering, the output is b n It is expressed as:

[0021]

[0022] Where h(q) is the coefficient of the filter, a(np) is the signal after a(n) is delayed by p cycles, and P is the number of zero points in the plane, p = 0, 1, ..., P-1.

[0023] S13: Frequency domain filtering. The signal is converted to the frequency domain by Fourier transform and frequency domain operation is performed. The signal is converted back to the time domain by inverse Fourier transform. The formula for converting the signal to the frequency domain is expressed as:

[0024]

[0025] The formula to convert the signal back to the time domain is:

[0026]

[0027] Wherein, a and b are spatial variables, a=0, 1,…, N-1, b=0, 1,…, M-1, u and v are frequency variables, u=0, 1,…, N-1, v=0, 1,…, M-1, and j is an imaginary number.

[0028] Further, in step S22, the calculation formula of the frequency hopping signal is expressed as:

[0029] S(t)=d(t)FH(t)

[0030] Where FH(t) is the frequency hopping baseband signal, expressed as:

[0031]

[0032] In the formula, t is the time variable, A represents the amplitude, α is the number of frequency points, ω T is a rectangular window, T h is the period of the frequency hopping signal, which is also the width of the rectangular window, f 0 、f 1 ,…,f k represents the frequency hopping frequency set of the FM signal, k+1 is the number of frequency hopping frequencies, represents the initial phase, n=0, 1, 2,…, α-1.

[0033] Further, in step S23, the formula of the wireless communication signal received through the Rayleigh channel is expressed as:

[0034] r(t)=X(t)cosω c tY(t)sinω c t

[0035]

[0036] In the formula, X(t) and Y(t) are independent random variables. is the phase, a i (t) represents the amplitude of the signal propagating through the path numbered i, ω c is the carrier frequency.

[0037] Furthermore, in step S32, the position coding formula is expressed as:

[0038]

[0039] Where PE(pos, 2j) is the encoder position code, PE(pos, 2j+1) is the decoder position code, pos is the position in the sequence, j is the dimension index in the position code vector, and d m is the embedding dimension.

[0040] Furthermore, in step S34, the improved CTNN network model includes 4 convolutional layer groups, 1 encoder module and 1 fully connected layer group.

[0041] The first convolutional layer group contains a convolutional layer with a kernel size of 2×8, a batch normalization layer, a ReLU activation layer, and a maximum pooling layer. The second convolutional layer group contains a point-wise convolutional layer. The encoder module contains γ improved encoders. The third convolutional layer group contains a point-wise convolutional layer. The fourth convolutional layer group contains a convolutional layer with a kernel size of 1×4. The fully connected layer group contains an average pooling layer, a flattening layer, a fully connected layer, and a Softmax activation function.

[0042] The present invention also provides an analysis system for wireless communication signals. The analysis system includes a signal receiving module, a preprocessing module, a data set building module, and a signal recognition model.

[0043] The signal receiving module is used to receive wireless communication signals.

[0044] The preprocessing module is used to perform time domain filtering and frequency domain filtering on the wireless communication signal.

[0045] The dataset construction module is used to modulate the wireless communication signal and use the multiplier to calculate the frequency hopping signal. The frequency hopping signal is collected through the Rayleigh channel and mapped into a time domain signal image to obtain the UAD signal dataset.

[0046] The signal recognition model includes a CTNN network model composed of a Transformer encoder and a convolutional neural network. The CTNN network model includes 4 convolutional layer groups, 1 encoder module and 1 fully connected layer group. Among them, the first convolutional layer group includes a convolutional layer with a convolution kernel size of 2×8, a batch normalization layer, a ReLU activation layer, and a maximum pooling layer. The second convolutional layer group includes a point-by-point convolution layer. The encoder module includes γ improved encoders. The third convolutional layer group includes a point-by-point convolution layer. The fourth convolutional layer group includes a convolutional layer with a convolution kernel size of 1×4. The fully connected layer group includes an average pooling layer, a flattening layer, a fully connected layer and a Softmax activation function.

[0047] The signal recognition model is trained using the UAD signal data set, and the model parameters that meet the test accuracy are retained. After the signal recognition model is trained, it takes the preprocessed wireless communication signal as input and outputs the signal analysis results.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention adopts an improved CTNN model, combines the convolutional layer with the encoder, improves the model's acquisition of local information, and improves the model's recognition performance. The Transformer-based modulation recognition method is more efficient and more suitable for parallel distributed GPU computing. The convolutional neural network with an attention mechanism is more adaptive and scalable.

[0050] The present invention constructs a UAD signal data set, adds Gaussian white noise with a certain signal-to-noise ratio to the frequency hopping signal to simulate the frequency hopping signal more realistically, and makes the frequency hopping signal pass through the Rayleigh channel. The receiving end collects the frequency hopping signal and maps it into a time domain signal image. The present invention simulates and generates signals by itself, adds annotations to construct a data set, expands the signal data set sufficient for neural network testing, and thus meets the testing requirements of the neural network and improves the accuracy of neural network recognition analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a step diagram of a method for analyzing wireless communication signals in Embodiment 1 of the present invention;

[0052] Figure 2 for Figure 1 Flowchart of the method for constructing the UAD signal dataset;

[0053] Figure 3 for Figure 1 Schematic diagram of the structure of the CTNN network model. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] It should be noted that when a component is referred to as being "mounted on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component at the same time. When a component is considered to be "fixed to" another component, it may be directly fixed on the other component or there may be a central component at the same time.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" used herein includes any and all combinations of one or more of the related listed items.

[0057] Example 1: Please refer to Figure 1 This embodiment provides a method for analyzing a wireless communication signal, comprising the following steps:

[0058] S1: Receive wireless communication signals and pre-process the wireless communication signals. The specific method is as follows:

[0059] S11: Signal sampling. Assume the sampling frequency is f s , the sampling time interval is T s , then the sampled signal a(n) can be expressed as:

[0060] a(n)=a(t)|(t=nT s )

[0061] Where n is the number of sampling times.

[0062] S12: Time domain filtering. Time domain filtering can use FIR or IIR filters. This embodiment uses FIR filters to output b n It can be expressed as:

[0063]

[0064] Where h(q) is the coefficient of the filter, a(np) is the signal after a(n) is delayed by p cycles, and P is the number of zero points in the plane, p = 0, 1, ..., P-1.

[0065] S13: Frequency domain filtering. Frequency domain filtering can convert the signal to the frequency domain through Fourier transform, then perform frequency domain operations, and finally convert the signal back to the time domain through inverse Fourier transform.

[0066] Among them, the formula for transforming the signal into the frequency domain is expressed as:

[0067]

[0068] Wherein, a and b are spatial variables, a=0, 1,…, N-1, b=0, 1,…, M-1, u and v are frequency variables, u=0, 1,…, N-1, v=0, 1,…, M-1, and j is an imaginary number.

[0069] The formula to convert the signal back to the time domain is:

[0070]

[0071] Fourier transform has a wide range of applications in signal processing, communication system analysis and other fields, and can help understand the spectral characteristics and frequency domain behavior of signals.

[0072] S2: Collect wireless communication signal data sets, modulate the wireless communication signals in the data sets, and construct a UAD signal data set.

[0073] See also Figure 2 ,The specific method of constructing the UAD signal dataset is as follows:

[0074] S21: Use a modulator to modulate the wireless communication signal in the data set to obtain a modulated signal. In this embodiment, the wireless communication signal is assumed to be b(t), which is modulated by the modulator to generate a modulated signal d(t).

[0075] S22: Use a multiplier to multiply the modulated signal and the required carrier signal to obtain a frequency hopping signal. The calculation formula of the frequency hopping signal is expressed as:

[0076] S(t)=d(t)FH(t)

[0077] Where FH(t) is the frequency hopping baseband signal, expressed as:

[0078]

[0079] In the formula, t is the time variable, A represents the amplitude, α is the number of frequency points, ω T is a rectangular window, T h is the period of the frequency hopping signal, which is also the width of the rectangular window, f 0 、f 1 ,…,f k represents the frequency hopping frequency set of the FM signal, k+1 is the number of frequency hopping frequencies, represents the initial phase, n=0, 1, 2,…, α-1.

[0080] S23: collect the frequency hopping signal through the Rayleigh channel and map it into a time domain signal image. The specific process is as follows:

[0081] Assuming that the frequency hopping signal is a single frequency sine wave, ignoring the direct path, and the multipath channel has n different paths, each path has a different transmission time extension and path loss, and there is no correlation between the different paths reaching the receiving end, the signal synthesis wave reaching the receiving end can be formulated as follows:

[0082]

[0083] In the formula, a i(t) represents the amplitude of the signal propagating through the path numbered i, ω c is the carrier frequency, τ i (t) represents the time by which the signal transmission time propagating through the path numbered i is extended.

[0084] The random phase of the signal transmitted through the path numbered i received by the receiver can be expressed as:

[0085] r(t)=X(t)cosω c tY(t)sinω c t

[0086]

[0087] In the formula, X(t) and Y(t) are independent random variables. For phase.

[0088] From the central limit theorem, we know that the distribution of the sum of multiple independent random variables approximately follows the normal distribution. Therefore, when n is large, both X(t) and Y(t) can be approximated as normal distributions. In the study, direct paths are usually not considered, so both X(t) and Y(t) have zero mean and the same variance. The representation of r(t) can be further converted to:

[0089]

[0090] Among them, the one-dimensional distribution of V(t) meets the Rayleigh distribution standard, and The one-dimensional distribution of meets the uniform distribution standard, so it can be further formulated as:

[0091]

[0092] Where f(v) and f(φ) are probability density functions, v is the complex fading coefficient of the Rayleigh channel, and σ v is the standard deviation of the Rayleigh channel.

[0093] S3: Construct a signal recognition model based on the CTNN neural network; the CTNN neural network is a convolutional transform neural network.

[0094] See also Figure 3 ,The specific method of constructing the signal recognition model is as follows:

[0095] S31: Build a basic Transformer neural network. The neural network includes an encoder and a decoder. The encoder consists of γ identical layers, each of which contains two sublayers: a multi-head self-attention mechanism and a fully connected feedforward network. In the encoder, the input sequence is fed into multiple attention heads respectively, and the relationship between the elements in the input sequence is modeled to generate a representation of each element. These representations are fed into the feedforward neural network for processing to generate a higher-level semantic representation.

[0096] The decoder also consists of γ identical layers, each with three sublayers. In addition to the two sublayers that are the same as the encoder, a third sublayer is inserted to perform multi-head attention on the encoder's output, combining the representation produced by the encoder with the representation produced by the decoder itself. In the decoder, the multi-head self-attention mechanism is used not only to generate representations for each element, but also to generate relations between output sequences.

[0097] S32: Add position encoding to the input embedding of the encoder and decoder. Position encoding uses sine and cosine functions of different frequencies, and the formula is as follows:

[0098]

[0099] Where PE(pos, 2j) is the encoder position code, PE(pos, 2j+1) is the decoder position code, pos is the position in the sequence, j is the dimension index in the position code vector, and d m is the embedding dimension.

[0100] The formula of the attention mechanism is as follows:

[0101]

[0102] Where Q and K are query matrix and key matrix respectively, d k is the dimension of Q and K, V is the value matrix, its dimension is d v , T is the transposed matrix. Among them, the matrices Q, K, and V are composed of the number of queries m and the dimension d k / d v Sure.

[0103] The calculation formula of the multi-head attention mechanism is as follows:

[0104] MH(Q, K, V) = Concat(MH 1 , H 2 , …, H h )W O

[0105]

[0106] Where W O is the weight matrix, h is the number of heads, are the weights corresponding to Q, K, and V respectively.

[0107] S33: Based on the encoder part of the Transformer model, the input image is segmented into patches of fixed size and the linear embedding sequence of these patches is used as the input of the Transformer encoder.

[0108] In classification tasks, usually only the encoder part of the Transformer is needed, and the decoder is not used. The reason is that the image classification task is a purely supervised learning task that requires classifying the input image without generating the output image. This is different from the natural language processing task. In the image classification task, there is no need to generate sequential outputs, so a decoder is not required.

[0109] Specifically, the signal sequence is first patched, the sequence is divided into patches of fixed size, and the linear embedding sequence of these patches is added with position embedding as the input of the Transformer encoder. In order to perform the classification task, an additional learnable all-zero vector needs to be added to the sequence and used as the final classification vector. The patched dataset is input to the encoder and passed through the multi-head attention sublayer, so that the self-attention mechanism of different parts learns different features in different vector subspaces, thereby extracting richer features. Then, layer normalization is performed to normalize the input vector to a standard normal distribution. Finally, the normalized vector is sent to the feedforward neural sublayer to further extract the features of the input vector. Finally, the encoder output is feature mapped through the fully connected layer, and then probability mapped through the Softmax layer to output the learned modulation type label.

[0110] S34; adding the improved encoder to the CTNN network model, the improved CTNN network model includes 4 convolutional layers, 1 encoder module and 1 fully connected layer.

[0111] The first convolution layer group contains a convolution layer with a convolution kernel size of 2×8, a batch normalization layer, a ReLU activation layer, and a maximum pooling layer. The second convolution layer group contains only one point-by-point convolution layer, which can reduce the number of channels of the input features. The first and second convolution layers are used to extract local features of the signal.

[0112] γ represents the number of improved encoders in the encoder module. The third convolution layer group is a point-by-point convolution layer, which is used to map the feature vector at each position in the global feature map from high dimension to low dimension. The point-by-point convolution layer is used to extract the global features of the signal.

[0113] The extracted local features and global features are fused using residual connections. After the fused features pass through the fourth convolutional layer group, the fully connected layer and Softmax activation function are used to map the output to probability values ​​of different modulation methods to obtain the identification label.

[0114] Assuming that the input of the current layer is x and the output is F(x), the residual connection can be expressed as:

[0115] y=F(x)+x

[0116] In the formula, y is the predicted value.

[0117] Residual connections can ensure the integrity and fluidity of information, effectively avoid the loss and deformation of information in deep networks, effectively reduce network degradation, and improve the accuracy and generalization ability of the network.

[0118] S4: Divide the data set into a training set and a test set, use the signal data in the training set to train the signal recognition model, and use the signal data in the test set to verify it, retaining the signal recognition model parameters that meet the test accuracy. Modulate the received wireless communication signal and input it into the signal recognition model, and output the corresponding analysis results.

[0119] The analysis method of this embodiment adopts an improved CTNN model, combining the convolutional layer with the encoder to improve the model's acquisition of local information and improve the model's recognition performance. The modulation recognition method based on Transformer is more efficient and more in line with parallel distributed GPU (graphics processing unit) calculations. In addition, by constructing a USD data set, Gaussian white noise with a certain signal-to-noise ratio is added to the frequency hopping signal to more realistically simulate the frequency hopping signal, and the frequency hopping signal is passed through a Rayleigh channel. The receiving end collects the frequency hopping signal and maps it into a time domain signal image. The present invention simulates and generates signals by itself, adds annotations to construct a data set, and expands the signal data set sufficient for neural network testing, thereby meeting the testing requirements of neural networks and improving the accuracy of neural network recognition analysis.

[0120] Embodiment 2: This embodiment provides an analysis system for wireless communication signals, which can be controlled by the analysis method for wireless communication signals of Embodiment 1, and includes a signal receiving module, a preprocessing module, a data set construction module, and a signal recognition model.

[0121] The signal receiving module is used to receive wireless communication signals.

[0122] The preprocessing module is used to perform time domain filtering and frequency domain filtering on the wireless communication signal.

[0123] The dataset construction module is used to modulate the wireless communication signal and use the multiplier to calculate the frequency hopping signal. The frequency hopping signal is collected through the Rayleigh channel and mapped into a time domain signal image to obtain the UAD signal dataset.

[0124] The signal recognition model includes a CTNN network model composed of a Transformer encoder and a convolutional neural network. The CTNN network model includes 4 convolutional layer groups, 1 encoder module and 1 fully connected layer group. Among them, the first convolutional layer group includes a convolutional layer with a convolution kernel size of 2×8, a batch normalization layer, a ReLU activation layer, and a maximum pooling layer. The second convolutional layer group includes a point-by-point convolution layer. The encoder module includes γ improved encoders. The third convolutional layer group includes a point-by-point convolution layer. The fourth convolutional layer group includes a convolutional layer with a convolution kernel size of 1×4. The fully connected layer group includes an average pooling layer, a flattening layer, a fully connected layer and a Softmax activation function.

[0125] The signal recognition model is trained using the UAD signal data set, and the model parameters that meet the test accuracy are retained. After the signal recognition model is trained, it takes the preprocessed wireless communication signal as input and outputs the signal analysis results.

[0126] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0127] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for analyzing wireless communication signals, characterized in that: The steps include: S1: receiving a wireless communication signal and preprocessing the wireless communication signal; S2: Collect wireless communication signal data sets, modulate the wireless communication signals in the data sets, and construct a UAD signal data set; the specific method is as follows: S21: using a modulator to modulate the wireless communication signal in the data set to obtain a modulated signal; S22: multiplying the modulated signal and the carrier signal by a multiplier to obtain a frequency hopping signal; S23: collecting the frequency hopping signal through a Rayleigh channel and mapping it into a time domain signal image; S3: Construct a signal recognition model based on a CTNN neural network, where the CTNN neural network is a convolutional transform neural network. The specific method is as follows: S31: Get the encoder of the Transformer neural network; S32: adding the position code to the input embedding of the encoder to obtain an improved encoder; S33: segmenting the time domain signal image into patches of fixed size, and using a linear embedding sequence of the patches as input of the encoder; S34: Add the improved encoder to the CTNN network model; S4: Use the UAD signal data set to train the signal recognition model, retain the model parameters that meet the test accuracy; modulate the preprocessed wireless communication signal and input it into the signal recognition model, and output the corresponding analysis result.

2. A method for analyzing wireless communication signals according to claim 1, characterized in that: In step S1, the signal preprocessing method is specifically as follows: S11: signal sampling; S12: Time domain filtering: Using FIR filter for time domain filtering, the output b(n) is expressed as: Where h(q) is the coefficient of the filter, a(np) is the signal after a(n) is delayed by p cycles, and P is the number of zero points in the plane, p = 0, 1, ..., P-1; S13: frequency domain filtering; convert the signal to the frequency domain by Fourier transform and perform frequency domain operation; convert the signal back to the time domain by inverse Fourier transform; wherein the formula for converting the signal to the frequency domain is expressed as: The formula to convert the signal back to the time domain is: Wherein, a and b are spatial variables, a=0, 1,…, N-1, b=0, 1,…, M-1, u and v are frequency variables, u=0, 1,…, N-1, v=0, 1,…, M-1, and j is an imaginary number.

3. The method for analyzing a wireless communication signal according to claim 1, characterized in that: In step S22, the calculation formula of the frequency hopping signal is expressed as: S(t)=d(t)FH(t) Where FH(t) is the frequency hopping baseband signal, expressed as: In the formula, t is the time variable, A represents the amplitude, α is the number of frequency points, ω T is a rectangular window, T h is the period of the frequency hopping signal, which is also the width of the rectangular window, f0, f1, ..., f k represents the frequency hopping frequency set of the FM signal, k+1 is the number of frequency hopping frequencies, represents the initial phase, n=0, 1, 2,…, α-1.

4. A method for analyzing wireless communication signals according to claim 3, characterized in that: In step S23, the formula for the wireless communication signal received through the Rayleigh channel is expressed as: r(t)=X(t)cosω c t-Y(t)sinω c t In the formula, X(t) and Y(t) are independent random variables. is the phase, a i (t) represents the amplitude of the signal propagating through the path numbered i, ω c is the carrier frequency.

5. The method for analyzing a wireless communication signal according to claim 1, characterized in that: In step S32, the position coding formula is expressed as: Where PE(pos, 2j) is the encoder position code, PE(pos, 2j+1) is the decoder position code, pos is the position in the sequence, j is the dimension index in the position code vector, and d m is the embedding dimension.

6. The method for analyzing a wireless communication signal according to claim 1, characterized in that: In step S34, the improved CTNN network model includes 4 convolutional layer groups, 1 encoder module and 1 fully connected layer group; Among them, the first convolution layer group includes a convolution layer with a convolution kernel size of 2×8, a batch normalization layer, a ReLU activation layer, and a maximum pooling layer; the second convolution layer group includes a point-by-point convolution layer; the encoder module includes γ improved encoders; the third convolution layer group includes a point-by-point convolution layer; the fourth convolution layer group includes a convolution layer with a convolution kernel size of 1×4; the fully connected layer group includes an average pooling layer, a flattening layer, a fully connected layer, and a Softmax activation function.

7. A wireless communication signal analysis system, which adopts a wireless communication signal analysis method as claimed in any one of claims 1 to 6, characterized in that: The analysis system comprises: A signal receiving module, used for receiving wireless communication signals; A preprocessing module, used for performing time domain filtering and frequency domain filtering on the wireless communication signal; The data set construction module is used to modulate the wireless communication signal and calculate the frequency hopping signal using the multiplier. The frequency hopping signal is collected through the Rayleigh channel and mapped into a time domain signal image to obtain the UAD signal data set. A signal recognition model, comprising a CTNN network model consisting of a Transformer encoder and a convolutional neural network, wherein the CTNN network model comprises 4 convolutional layer groups, 1 encoder module and 1 fully connected layer group; wherein the first convolutional layer group comprises a convolutional layer with a convolution kernel size of 2×8, a batch normalization layer, a ReLU activation layer, and a maximum pooling layer; the second convolutional layer group comprises a point-by-point convolutional layer; the encoder module comprises γ improved encoders; the third convolutional layer group comprises a point-by-point convolutional layer; the fourth convolutional layer group comprises a convolutional layer with a convolution kernel size of 1×4; the fully connected layer group comprises an average pooling layer, a flattening layer, a fully connected layer and a Softmax activation function; The signal recognition model is trained using the UAD signal data set, and model parameters that meet the test accuracy are retained; after completing the training, the signal recognition model uses the preprocessed wireless communication signal as input and outputs a signal analysis result.

Images