Communication signal multi-parameter intelligent parallel extraction method

Abstract

The embodiment of the invention discloses a communication signal multi-parameter intelligent parallel extraction method. The method comprises the following steps: firstly, acquiring a to-be-measured signal and extracting an I / Q component; processing the I / Q components through a multi-scale feature extraction structure to obtain high-dimensional features; extracting shared features from the high-dimensional features through a plurality of residual block groups, inputting a plurality of parallel task branches, and enhancing the shared features in the branches by using a channel attention mechanism to obtain output features; the output features of all task branches are transformed through a learnable linear layer to form a key matrix and a value matrix, the current task features are transformed through a learnable linear layer to form a query matrix, and the input features are segmented into a plurality of heads by means of a multi-head attention mechanism; each head obtains output of a single head based on zoom dot product self-attention operation, and results of all the heads are integrated to obtain enhanced features of the current task; and determining a communication signal parameter prediction label according to the enhanced features, and realizing multi-parameter efficient parallel extraction.

Description

Technical Field

[0001] This invention relates to the field of communication signal processing technology, and in particular to a method for intelligent parallel extraction of multiple parameters of communication signals. Background Technology

[0002] With the development of wireless communication technology, the scarcity of spectrum resources has made spectrum monitoring and spectrum sensing particularly important. In spectrum sensing, communication signal parameter extraction is a crucial step in signal detection and analysis, and its core tasks typically include modulation type identification, coding parameter extraction, and signal-to-noise ratio (SNR) evaluation. Specifically, modulation type identification aims to determine the modulation scheme used by the signal; coding parameter extraction involves identifying the channel coding scheme in the signal, such as polar codes and convolutional codes; and SNR estimation is a quantitative assessment of the received signal quality, providing an important basis for subsequent signal processing. The accurate extraction of these parameters collectively forms the foundation for intelligent understanding and reconstruction of communication signals, and is of critical significance in fields such as military reconnaissance, spectrum surveillance, and intelligent communication.

[0003] Traditional processing methods mostly employ a hierarchical approach: first, the modulation type of the signal is identified; then, based on the demodulated bitstream, its channel coding method is further analyzed; and finally, communication parameters (such as signal-to-noise ratio) are estimated. However, this hierarchical approach has significant drawbacks. First, it increases the design complexity and computational load of the algorithm, leading to low processing efficiency. Second, because the identification results of each stage are interdependent, inaccuracy in any stage will affect the performance of subsequent stages, resulting in accumulated errors. In recent years, with the development of intelligent methods such as deep learning, some studies have attempted to introduce neural networks to improve recognition performance. However, most models only focus on extracting single parameters, failing to fully explore the inherent correlations and shared features between multiple parameters in the signal, resulting in model redundancy and difficulty in meeting the practical needs of real-time, efficient, and integrated analysis. Summary of the Invention

[0004] Based on this, it is necessary to propose a multi-parameter intelligent parallel extraction method for communication signals to address the above problems.

[0005] A method for intelligent parallel extraction of multiple parameters of communication signals, the method comprising: Acquire the signal to be tested and extract the I / Q components of the signal to be tested; The I / Q components are processed using a multi-scale feature extraction structure to obtain high-dimensional features; The shared features of the high-dimensional features are extracted by multiple residual block groups. Each shared feature is input into multiple parallel task branches. Within each task branch, the shared features are enhanced by a channel attention mechanism to obtain the output features of the current task. Each task branch corresponds to a communication signal parameter to be extracted. Using a multi-head attention mechanism, the output features of all task branches are transformed through a learnable linear layer to form a key matrix and a value matrix. The current task features are transformed through a learnable linear layer to form a query matrix. The key matrix, value matrix, and query matrix are divided into multiple heads. Each head is output based on a scaling dot product self-attention operation. The outputs of multiple heads are integrated to obtain the enhanced features of the current task after interaction of different task features. The communication signal parameter prediction label is determined based on the enhancement features of each of the task branches.

[0006] The multi-scale feature extraction structure includes: a first convolutional layer, a first feature extraction structure, a second convolutional layer, and a second feature extraction structure. The process of using the multi-scale feature extraction structure to process the I / Q components to obtain high-dimensional features specifically includes: The I / Q components are processed by the first convolutional layer to obtain initial features. The initial features are nonlinearly mapped by the parallel sub-paths of the first feature extraction structure. Each path extracts the local structural information of the initial features through convolutional kernels of different preset sizes. The local structural information of the initial features output by each path is concatenated by the channel dimension and then downsampled by max pooling to obtain the primary representation. The primary representation is processed by the second convolutional layer to obtain intermediate features. A max pooling operation path is further introduced on the first feature extraction structure to form a second feature extraction structure. The intermediate features are processed by the second feature extraction structure. The outputs of each path are concatenated by the channel dimension and then downsampled by the max pooling operation to obtain high-dimensional features.

[0007] The residual block group includes two residual blocks. The process involves extracting shared features from the high-dimensional features using multiple residual block groups, inputting these shared features into multiple parallel task branches, and enhancing the shared features within each task branch using a channel attention mechanism to obtain the output features of the current task. Specifically, this includes: The shared features of the high-dimensional features are extracted by the two residual blocks of the residual block group, and the shared features include a first shared feature and a second shared feature. The shared features are input into multiple parallel task branches, and within each task branch, the first shared features are processed through a channel attention mechanism to generate channel attention weights. The channel attention weights are multiplied element-wise with the second shared feature to obtain the enhanced output features of the current task.

[0008] Specifically, the step of dividing the key matrix, value matrix, and query matrix into multiple headers, and obtaining the output of a single header based on a scaled dot product self-attention operation, includes: according to Determine the attention weight matrix, where A is the attention weight matrix, Q is the query matrix, and K is the transpose of the key matrix; according to The output of a single head is obtained, Z is the output of a single head, A is the attention weight matrix, and V is the value matrix.

[0009] Specifically, acquiring the signal under test and extracting the I / Q components of the signal under test includes: Acquire the signal to be tested and normalize it.

[0010] according to Determine the I component of the signal.

[0011] according to Determine the Q component of the signal, where, For the I component of the signal, For the Q component of the signal, is the normalized signal, and * represents the complex conjugate.

[0012] A multi-parameter intelligent parallel extraction system for communication signals, the system comprising: The I / Q component extraction module is used to acquire the signal under test and extract the I / Q components of the signal under test.

[0013] The high-dimensional feature determination module is used to process the I / Q components through a multi-scale feature extraction structure to obtain high-dimensional features.

[0014] The output feature determination module is used to extract shared features of the high-dimensional features through multiple residual block groups, input each shared feature to multiple parallel task branches, and enhance the shared features through a channel attention mechanism within each task branch to obtain the output features of the current task. Each task branch corresponds to a communication signal parameter to be extracted.

[0015] The enhanced feature determination module utilizes a multi-head attention mechanism to transform the output features of all task branches into a key matrix and a value matrix through a learnable linear layer. The current task features are transformed into a query matrix through a learnable linear layer. The key matrix, value matrix, and query matrix are divided into multiple heads. Each head is used to obtain the output of a single head based on a scaled dot product self-attention operation. The outputs of multiple heads are integrated to obtain the enhanced features of the current task after interaction with features from different tasks.

[0016] A communication signal parameter prediction module is used to determine communication signal parameter prediction labels based on the enhancement features of each of the task branches.

[0017] The system also includes: The training module is used to take a bitstream of a preset length as the original signal, process the original signal according to different signal processing methods, and obtain a channel simulation dataset. The channel simulation dataset includes the I / Q components of the original signal under different signal processing methods and the parameter configurations corresponding to the generation of the I / Q components. The parameter configurations are used as the true parameter labels. The module determines the corresponding communication signal parameter prediction label for each I / Q component in the channel simulation data, defines a total loss function based on the communication signal parameter prediction label and the true parameter label, and updates the parameters of the high-dimensional feature determination module, the output feature determination module, the enhanced feature determination module, and the communication signal parameter prediction module through the total loss function.

[0018] Specifically, defining the total loss function based on the predicted labels and the true labels of the communication signal parameters includes: according to Define the total loss function, where X is the I / Q component. The actual label for the parameter corresponding to the i-th task branch. As task weight, Let cross-entropy be the loss function. Where C is the total number of categories for the i-th branch task. This represents the true probability that the parameter's true label belongs to the c-th class on the i-th task branch. This represents the predicted probability that the parameter prediction label belongs to class c on the i-th task branch.

[0019] A computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the steps of the method described above.

[0020] A computer device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method described above.

[0021] The embodiments of the present invention have the following beneficial effects: This invention processes initial and intermediate features using a first feature extraction structure and a second feature extraction structure, respectively, and then concatenates the results. This reduces model complexity and improves real-time processing capabilities. Utilizing a parallel branching structure, different types of convolutional operations can be performed simultaneously for efficient feature extraction with low complexity. Furthermore, by simultaneously extracting multiple parameters under a parallel task branching architecture, joint perception of different communication parameters is achieved, reducing redundant computations and avoiding the gradual accumulation of errors, thus significantly improving processing efficiency. Simultaneously, through a multi-head attention mechanism, dynamic feature interaction and information fusion between different task branches are realized, improving the overall classification accuracy of the model and enhancing its recognition precision and generalization ability. This results in efficient and accurate extraction of communication signal parameters, enabling rapid and accurate prediction results for communication signal parameters, greatly improving practicality and convenience in real-world applications. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] in: Figure 1 This is a flowchart illustrating an embodiment of the intelligent parallel extraction method for multiple parameters of communication signals provided by the present invention. Figure 2 This is a schematic diagram of an embodiment of an intelligent parallel extraction system for multiple parameters of communication signals provided by the present invention; Figure 3 This is a schematic diagram of the structure of an embodiment of the initial feature extraction module provided by the present invention; Figure 4 The multi-head attention mechanism structure diagram provided by this invention; Figure 5 This is a schematic diagram of an embodiment of an intelligent parallel extraction system for multiple parameters of communication signals provided by the present invention; Figure 6 The confusion matrix of different channel coding methods provided by this invention on the verification set; Figure 7 The confusion matrix of different modulation types on the verification set provided by this invention; Figure 8 This invention provides confusion matrices for different signal-to-noise ratios on a validation set; Figure 9 A schematic diagram of the structure of an embodiment of the device provided by the present invention; Figure 10 A schematic diagram of the structure of an embodiment of the medium provided by the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] like Figure 1 As shown, Figure 1 This is a flowchart illustrating an embodiment of an intelligent parallel extraction method for multiple parameters of communication signals provided by the present invention. The method includes: S101: Acquire the signal under test and extract the I / Q components of the signal under test.

[0026] For example, the signal to be measured is first acquired and then normalized. The I component of the signal is determined according to the following formula: ; The Q component of the signal is determined using the formula shown below: ; in, For the I component of the signal, For the Q component of the signal, is the normalized signal, and * represents the complex conjugate.

[0027] S102: The I / Q components are processed through a multi-scale feature extraction structure to obtain high-dimensional features.

[0028] For example, in conjunction with reference Figure 2 and Figure 3 , Figure 2 This is a schematic diagram of an embodiment of the intelligent parallel extraction system for multi-parameter communication signals provided by the present invention. Figure 3This is a schematic diagram of an embodiment of the initial feature extraction module provided by the present invention. The multi-scale feature extraction structure includes a first feature extraction structure and a second feature extraction structure. First, the I / Q components are processed by the initial feature extraction module to obtain high-dimensional features. The initial feature extraction module includes: a first convolutional layer, a first feature extraction structure, a second convolutional layer, and a second feature extraction structure. Specifically, the I / Q components are processed by the first convolutional layer to obtain initial features. The initial features are then processed by the first feature extraction structure. The outputs of the first feature extraction structure are concatenated and then subjected to max pooling to obtain the first concatenated features. The first feature extraction structure includes: a 1×1 convolution (Conv1×1), a 3×1 convolution (Conv3×1), and a 5×1 convolution (Conv5×1). The first convolutional layer includes: a 31×1 convolution (Conv31×1), a batch normalization layer (BN), and a ReLU activation function.

[0029] Furthermore, the first concatenated features are processed by the second convolutional layer to obtain intermediate features. The intermediate features are then processed by the second feature extraction structure. The outputs of the second feature extraction structure are concatenated and then subjected to max pooling (MaxPool2D2×1) to obtain high-dimensional features. The second feature extraction structure includes: 1×1 convolution (Conv1×1), 3×1 convolution (Conv3×1), 5×1 convolution (Conv5×1), and 3×1 max pooling (MaxPool3×1). The second convolutional layer includes: 7×1 convolution (Conv7×1), batch normalization (BN) layer, and ReLU activation function.

[0030] By using 1×1, 3×1, and 5×1 convolutional kernels and pooling layers in parallel, the network can focus on details (small receptive field), grasp outlines (medium receptive field), and coarsely perceive patterns over a wider range (large receptive field) at the same level. Parallel branches are introduced twice in the primary feature extraction module, embedding multi-scale analysis capabilities into the initial features, which can provide higher quality and more information-density input for subsequent deep networks.

[0031] S103: Extract shared features of high-dimensional features through multiple residual block groups, input each shared feature into multiple parallel task branches, and enhance the shared features through a channel attention mechanism within each task branch to obtain the output features of the current task. Each task branch corresponds to a communication signal parameter to be extracted.

[0032] For example, in conjunction with reference Figure 2Two residual blocks are grouped together, and the number of feature channels is continuously increased. By changing the step size, the length and width of the feature matrix are reduced, thereby completing the dimensionality matching between residual block groups. High-dimensional features are progressively abstracted and mapped into a series of multi-level shared features through multiple residual block groups. Specifically, for each group of residual blocks, shared features of the high-dimensional features are extracted from the two residual blocks of the group. These shared features include a first shared feature and a second shared feature. The shared features are input to at least three parallel task branches. Within each task branch, the first shared feature is processed through a channel attention mechanism to generate channel attention weights. The channel attention weights are then multiplied element-wise with the second shared feature to obtain the enhanced output features. Each task branch corresponds to a communication signal parameter to be extracted.

[0033] S104: Utilizing a multi-head attention mechanism, the output features of all task branches are transformed through a learnable linear layer to form a key matrix and a value matrix. The current task features are transformed through a learnable linear layer to form a query matrix. The key matrix, value matrix, and query matrix are divided into multiple heads. Each head is used to obtain the output of a single head based on a scaling dot product self-attention operation. The outputs of multiple heads are integrated to obtain the enhanced features of the current task after interaction with features from different tasks.

[0034] For example, in conjunction with reference Figure 4 , Figure 4 This is a structural diagram of the multi-head attention mechanism provided by the present invention. Multi-head attention is an improvement on traditional attention mechanisms, aiming to improve the model's expressive and learning capabilities by segmenting the input features into multiple "heads" and processing each head independently. For an output feature f... i First, the output feature is linearly transformed using a learnable weight matrix, as shown in the following equation: ; ; ; Among them, f i This represents the output feature of the i-th task. , , It is the projection weight matrix corresponding to task i. For the input feature dimension, The projected feature dimensions, , , It is the bias term corresponding to task i.

[0035] Stacking the projection vectors of each output feature yields the query matrix (Q), key matrix (K), and value matrix (V): ; ; ; The transformation is performed using multiple sets of linear layers, dividing the Q, K, and V matrices into multiple heads, each with different linear transformation parameters.

[0036] Using the projection vector matrices of the aforementioned output features, the attention weight matrices of different output features for the current output feature can be calculated through scaled dot-product attention. : ; ; Among them, the attention weight matrix elements in The feature matrix represents the attention weights of the current output feature i for the output feature j. elements in The current output feature i is enhanced by interacting with all tasks, i.e., the output of each head. Finally, the outputs of all heads are concatenated, and the resulting long vector is mapped back to the original feature dimension (consistent with the current output feature). This vector is added to the current output feature and input to the normalization layer to integrate information from different heads, thus obtaining a single output feature after fusing features from different tasks at each stage.

[0037] A feed-forward network (FFN) consists of two fully connected neural networks, and its mathematical form is shown below.

[0038] ; Here, x represents the output of the multi-head attention mechanism described above after the residual connection. and These are the weights and biases of the first fully connected layer, respectively. , These are the weights and biases of the second fully connected layer, respectively. For hidden layer dimensions, usually It is four times that of the previous model. The feedforward module enables the interaction and recombination of feature dimensions within the same location, while introducing non-linear activation (ReLU), which allows the model to learn more complex feature representations and function mappings.

[0039] And, as Figure 4As shown, by using the residual connection again after using the feedforward module, this structure can effectively avoid the loss of the specific representation of the current task during the interaction.

[0040] It is worth noting that, after multiple experiments, it has been shown that using a multi-head attention mechanism for information interaction works better when applied to relatively shallow features. However, when each task has extracted its own specific features, using this structure for information interaction can actually negatively impact parameter extraction accuracy. Figure 2 The branch task only uses a multi-head attention mechanism for the first two sets of feature information, uses a weaker linear parameter for the interaction of the third set of feature information, and does not design a branch task interaction structure for the fourth set of feature information.

[0041] S105: Determine the predicted labels for communication signal parameters based on the enhanced features of each task branch.

[0042] For example, the parameter extraction results are output through the classification head in each task branch. For instance, the prediction results are output by the three task branches respectively: Modulation type: Directly output the category name (e.g., "QPSK") or the category index (e.g., 2); Channel coding method: Output coding name (e.g., "LDPC (3,2)") or category index (e.g., 5); Signal-to-noise ratio: Output level index (e.g., 3) or specific value (e.g., "10dB").

[0043] As described above, this invention processes initial and intermediate features using a first feature extraction structure and a second feature extraction structure, respectively, and then concatenates the results. This reduces model complexity and improves real-time processing capabilities. The parallel branching structure allows for efficient feature extraction of different types of convolutional operations simultaneously with low complexity. Furthermore, by simultaneously extracting multiple parameters under a parallel task branching architecture, joint perception of different communication parameters is achieved, reducing redundant computations and avoiding the gradual accumulation of errors, thus significantly improving processing efficiency. Simultaneously, the multi-head attention mechanism enables dynamic feature interaction and information fusion between different task branches, improving the overall classification accuracy of the model and enhancing its recognition precision and generalization ability. This results in efficient and accurate extraction of communication signal parameters, enabling rapid and accurate prediction of communication signal parameters in a short time, greatly improving practicality and convenience in real-world applications.

[0044] like Figure 5 As shown, Figure 5 This is a schematic diagram of an embodiment of an intelligent parallel extraction system for multi-parameter communication signals provided by the present invention. The intelligent parallel extraction system 10 for multi-parameter communication signals includes: The I / Q component extraction module 11 is used to acquire the signal under test and extract the I / Q components of the signal under test.

[0045] The high-dimensional feature determination module 12 is used to process the I / Q components through a multi-scale feature extraction structure to obtain high-dimensional features.

[0046] The output feature determination module 13 is used to extract the shared features of the high-dimensional features through multiple residual block groups, input each shared feature to multiple parallel task branches, and enhance the shared features through a channel attention mechanism within each task branch to obtain the output features of the current task. Each task branch corresponds to a communication signal parameter to be extracted.

[0047] The enhanced feature determination module 14 is used to utilize a multi-head attention mechanism to transform the output features of all task branches through a learnable linear layer to form a key matrix and a value matrix. The current task features are transformed through a learnable linear layer to form a query matrix. The key matrix, value matrix, and query matrix are divided into multiple heads. Each head is used to obtain the output of a single head based on a scaled dot product self-attention operation. The output results of multiple heads are integrated to obtain the enhanced features of the current task after interaction of different task features.

[0048] The communication signal parameter prediction module 15 is used to determine the communication signal parameter prediction label based on the enhancement features of each task branch.

[0049] For example, in the I / Q component extraction module 11, the signal to be measured is acquired, and the signal to be measured is normalized; according to Determine the I component of the signal; based on Determine the Q component of the signal, where, For the I component of the signal, For the Q component of the signal, is the normalized signal, and * represents the complex conjugate.

[0050] In the high-dimensional feature determination module 12, the multi-scale feature extraction structure includes: a first convolutional layer, a first feature extraction structure, a second convolutional layer, and a second feature extraction structure. The first convolutional layer processes the I / Q components to obtain initial features. The first feature extraction structure processes these initial features separately, concatenates the outputs of the first feature extraction structure, and then performs max pooling to obtain the first concatenated features. The first feature extraction structure includes: 1×1 convolution, 3×1 convolution, and 5×1 convolution. The second convolutional layer processes the first concatenated features to obtain intermediate features. The second feature extraction structure processes these intermediate features separately, concatenates the outputs of the second feature extraction structure, and then performs max pooling to obtain high-dimensional features. The second feature extraction structure includes: 1×1 convolution, 3×1 convolution, 5×1 convolution, and 3×1 max pooling.

[0051] In the output feature determination module 13, the residual block group includes two residual blocks. The two residual blocks of the residual block group are used to extract the shared features of the high-dimensional features. The shared features include the first shared feature and the second shared feature. Each shared feature is input to at least three parallel task branches. In each task branch, the first shared feature is processed by the channel attention mechanism to generate channel attention weights. The channel attention weights are multiplied element-wise with the second shared feature to obtain the enhanced output features.

[0052] In the enhanced feature determination module 14, a multi-head attention mechanism is used to transform the output features of all task branches into a key matrix and a value matrix through a learnable linear layer. The current task features are transformed into a query matrix through a learnable linear layer. The key matrix, value matrix, and query matrix are divided into multiple heads. Each head is output based on a scaling dot product self-attention operation. The output results of multiple heads are integrated to obtain the enhanced features of the current task after interaction of different task features.

[0053] In the communication signal parameter prediction module 15, the communication signal parameter prediction label is determined based on the enhancement features of each task branch.

[0054] The intelligent parallel extraction system for multi-parameter communication signals also includes: The training module (not shown in the figure) is used to take a bit stream of a preset length as the original signal and process the original signal according to different signal processing methods to obtain a channel simulation dataset. The channel simulation dataset includes the I / Q components of the original signal under different signal processing methods and the corresponding parameter configurations when generating the I / Q components. The parameter configurations are used as the true parameter labels. The corresponding communication signal parameter prediction label is determined according to each I / Q component in the channel simulation data. The total loss function is defined according to the communication signal parameter prediction label and the true parameter label. The parameters of the high-dimensional feature determination module, the output feature determination module, the enhanced feature determination module, and the communication signal parameter prediction module are updated through the total loss function.

[0055] For example, a random 0 and 1 bit stream of a specified length is first generated as the original signal b[n]. The original signal is then processed based on different signal processing methods to obtain a channel simulation dataset. The channel simulation dataset includes the I / Q components of the original signal under different signal processing methods and the corresponding parameter configurations when generating the I / Q components, as shown in Table 1. The channel simulation dataset has 6 digital modulation categories, 7 channel coding methods, and 11 signal-to-noise ratios distributed between 0dB and 20dB. Each combination of digital modulation, channel coding method, and signal-to-noise ratio has 1000 samples.

[0056] Table 1 Channel Simulation Dataset

[0057] Specifically, channel coding is performed on the original signal to obtain the coded signal c[k]; ; in, K represents the encoding function, and K represents the length of the encoded signal.

[0058] The encoded signal is mapped to a modulation symbol to obtain the modulation signal s[m];

[0059] in, Let M represent the modulation mapping function, and M represent the length of the modulated signal.

[0060] Furthermore, a square root raised cosine FIR filter with a roll-off factor of β is designed, and its standard expression for the time-domain impulse response h(t) is as follows:

[0061] in, It is the symbol period, the roll-off factor β is a random number in the range of 0.2 to 0.7, the filter spans 6 symbol periods, and each symbol contains 8 sampling points, that is, the filter length is 48.

[0062] The modulated signal is upsampled and convolved with the above filter to complete the pulse shaping filter, thus obtaining the filtered signal. ;

[0063] Where the upsampling factor sps is 8, * represents convolution operation, δ is the unit impulse function, n represents the sampling point number, and m represents the modulation signal number.

[0064] Furthermore, random phase rotation and frequency shift are added to the filtered signal, and additive white Gaussian noise is added according to the corresponding signal-to-noise ratio to simulate the signal in the actual channel;

[0065]

[0066]

[0067] in, To add phase rotation and frequency shift to the signal, This indicates an operation that randomly selects values ​​within a specified range. This represents the actual phase rotation value. This indicates the actual frequency offset value, and the range of frequency offset values. Set it to 0.1.

[0068]

[0069]

[0070] in, It is the average power of the signal before noise is added. It is the average power of the noise. It is additive white Gaussian noise, and its average power setting is obtained according to the formula for calculating the signal-to-noise ratio (SNR).

[0071] Furthermore, the noisy signal is normalized and the I / Q components are extracted.

[0072] The I component of the signal is determined using the formula shown below: ; The Q component of the signal is determined using the formula shown below: ; in, For the I component of the signal, For the Q component of the signal, is the normalized signal, and * represents the complex conjugate.

[0073] Furthermore, the channel simulation dataset is divided into a training set and a test set in a 7:3 ratio. The training set data is then fed into the high-dimensional feature determination module, the output feature determination module, the enhanced feature determination module, and the communication signal parameter prediction module in batches. Based on each I / Q component in the training set, the corresponding communication signal parameter prediction label is determined.

[0074] The total loss function is defined based on the predicted labels and the actual labels of the communication signal parameters, as shown in the following formula: ; Where X is the I / Q component, The actual label for the parameter corresponding to the i-th task branch. As task weight, Let cross-entropy be the loss function. Where C is the total number of categories for the i-th branch task. This represents the true probability that the parameter's true label belongs to the c-th class on the i-th task branch. This represents the predicted probability that the parameter prediction label belongs to class c on the i-th task branch.

[0075] Furthermore, the parameters of the high-dimensional feature determination module, the output feature determination module, the enhanced feature determination module, and the communication signal parameter prediction module are updated using the total loss function.

[0076] Furthermore, the performance of each task in the multi-task parallel parameter extraction system was tested on the validation set; in conjunction with references Figure 6 , Figure 7 and Figure 8 , Figure 6 This is the confusion matrix of different channel coding methods provided by the present invention on the verification set. Figure 7 This invention provides confusion matrices for different modulation types on the verification set. Figure 8 This invention provides confusion matrices for different signal-to-noise ratios on the validation set. For all signal-to-noise ratio data, the channel coding method extraction accuracy is 82.35%, the modulation type extraction accuracy is 98.98%, and the signal-to-noise ratio extraction accuracy is 98.50%.

[0077] The experimental results above demonstrate that the method proposed in this invention can efficiently and accurately perform parallel joint identification and extraction of channel coding scheme, modulation type and signal-to-noise ratio parameters under signal-to-noise ratio conditions of 0dB and above. Furthermore, this multi-task parallel extraction framework has strong scalability and can be further extended to include other communication parameter extraction branch tasks, thus possessing strong engineering practical value.

[0078] As described above, the multi-task parallel branch designed in this invention not only significantly reduces the number of model parameters and computational overhead, but also addresses scenarios with limited training samples, improving the model's robustness and generalization ability in complex electromagnetic environments. Furthermore, by decoupling and collaboratively optimizing each parameter extraction task within a unified framework, the error accumulation problem in traditional serial processes is avoided, significantly improving the overall efficiency and accuracy of communication signal parameter extraction, and possessing significant engineering application value.

[0079] like Figure 9 As shown, Figure 9 This is a schematic diagram of an embodiment of the device provided by the present invention. The device 20 includes a memory 21 and a processor 22. The memory 21 stores a computer program, and the processor 22 executes the computer program during operation to achieve, for example... Figure 1 The method shown.

[0080] The specific technical details of the intelligent parallel extraction method for multiple parameters of communication signals implemented by the above-mentioned device 20 when executing a computer program have been discussed in detail in the above method steps, and therefore will not be repeated here.

[0081] like Figure 10 As shown, Figure 10 This is a schematic diagram of the structure of an embodiment of the medium provided by the present invention. The medium 30 stores at least one computer program 31, which is executed by the processor 22 to perform the following... Figure 1 The method shown is detailed above and will not be repeated here. In one embodiment, the medium 30 can be a storage chip, hard disk, portable hard disk, USB flash drive, optical disk, or other read / write storage device, or even a server, etc.

[0082] Furthermore, the processes depicted in the accompanying drawings do not necessarily have to be performed in the specific or sequential order shown to achieve the desired result. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0083] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer-readable storage media are basically similar to the method embodiments, and therefore described more simply; relevant parts can be referred to the descriptions of the method embodiments.

[0084] The apparatus, device, non-volatile computer-readable storage medium and method provided in the embodiments of this specification are corresponding. Therefore, the apparatus, device and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, device and non-volatile computer storage medium will not be repeated here.

[0085] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0086] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components. Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0091] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0092] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0093] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0094] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0095] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0096] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for intelligent parallel extraction of multiple parameters of communication signals, characterized in that, The method includes: Acquire the signal to be tested and extract the I / Q components of the signal to be tested; The I / Q components are processed using a multi-scale feature extraction structure to obtain high-dimensional features; The shared features of the high-dimensional features are extracted by multiple residual block groups. Each shared feature is input into multiple parallel task branches. Within each task branch, the shared features are enhanced by a channel attention mechanism to obtain the output features of the current task. Each task branch corresponds to a communication signal parameter to be extracted. Using a multi-head attention mechanism, the output features of all task branches are transformed through a learnable linear layer to form a key matrix and a value matrix. The current task features are transformed through a learnable linear layer to form a query matrix. The key matrix, value matrix, and query matrix are divided into multiple heads. Each head is output based on a scaling dot product self-attention operation. The outputs of multiple heads are integrated to obtain the enhanced features of the current task after interaction of different task features. The communication signal parameter prediction label is determined based on the enhancement features of each of the task branches.

2. The intelligent parallel extraction method for multiple parameters of communication signals according to claim 1, characterized in that, The multi-scale feature extraction structure includes: a first convolutional layer, a first feature extraction structure, a second convolutional layer, and a second feature extraction structure. The process of processing the I / Q components through the multi-scale feature extraction structure to obtain high-dimensional features specifically includes: The I / Q components are processed by the first convolutional layer to obtain initial features. The initial features are nonlinearly mapped by the parallel sub-paths of the first feature extraction structure. Each path extracts the local structural information of the initial features through convolutional kernels of different preset sizes. The local structural information of the initial features output by each path is concatenated by the channel dimension and then downsampled by max pooling to obtain the primary representation. The primary representation is processed by the second convolutional layer to obtain intermediate features. A max pooling operation path is further introduced on the first feature extraction structure to form a second feature extraction structure. The intermediate features are processed by the second feature extraction structure. The outputs of each path are concatenated by the channel dimension and then downsampled by the max pooling operation to obtain high-dimensional features.

3. The intelligent parallel extraction method for multiple parameters of communication signals according to claim 1, characterized in that, The residual block group includes two residual blocks. The process involves extracting shared features from the high-dimensional features using multiple residual block groups, inputting these shared features into multiple parallel task branches, and enhancing the shared features within each task branch using a channel attention mechanism to obtain the output features of the current task. Specifically, this includes: The shared features of the high-dimensional features are extracted by the two residual blocks of the residual block group, and the shared features include a first shared feature and a second shared feature. The shared features are input into multiple parallel task branches, and within each task branch, the first shared features are processed through a channel attention mechanism to generate channel attention weights. The channel attention weights are multiplied element-wise with the second shared feature to obtain the enhanced output features of the current task.

4. The intelligent parallel extraction method for multiple parameters of communication signals according to claim 1, characterized in that, The step of dividing the key matrix, value matrix, and query matrix into multiple headers, and obtaining the output of a single header based on a scaled dot product self-attention operation, specifically includes: according to Determine the attention weight matrix, where A is the attention weight matrix, Q is the query matrix, and K is the transpose of the key matrix; according to The output of a single head is obtained, Z is the output of a single head, A is the attention weight matrix, and V is the value matrix.

5. The intelligent parallel extraction method for multiple parameters of communication signals according to claim 1, characterized in that, The step of acquiring the signal under test and extracting the I / Q components of the signal under test specifically includes: Acquire the signal to be tested and normalize the signal to be tested; according to Determine the I component of the signal; according to Determine the Q component of the signal, where, For the I component of the signal, For the Q component of the signal, is the normalized signal, and * represents the complex conjugate.

6. A multi-parameter intelligent parallel extraction system for communication signals, characterized in that, The system includes: The I / Q component extraction module is used to acquire the signal under test and extract the I / Q components of the signal under test. The high-dimensional feature determination module is used to process the I / Q components through a multi-scale feature extraction structure to obtain high-dimensional features; The output feature determination module is used to extract the shared features of the high-dimensional features through multiple residual block groups, input each of the shared features into multiple parallel task branches, and enhance the shared features through a channel attention mechanism within each task branch to obtain the output features of the current task. Each task branch corresponds to a communication signal parameter to be extracted. The enhanced feature determination module is used to utilize a multi-head attention mechanism to transform the output features of all task branches through a learnable linear layer to form a key matrix and a value matrix. The current task features are transformed through a learnable linear layer to form a query matrix. The key matrix, value matrix, and query matrix are divided into multiple heads. Each head is used to obtain the output of a single head based on a scaled dot product self-attention operation. The output results of multiple heads are integrated to obtain the enhanced features of the current task after interaction of different task features. A communication signal parameter prediction module is used to determine communication signal parameter prediction labels based on the enhancement features of each of the task branches.

7. The intelligent parallel extraction system for multiple parameters of communication signals according to claim 6, characterized in that, The system also includes: The training module is used to take a bitstream of a preset length as the original signal, process the original signal according to different signal processing methods, and obtain a channel simulation dataset. The channel simulation dataset includes the I / Q components of the original signal under different signal processing methods and the parameter configurations corresponding to the generation of the I / Q components. The parameter configurations are used as the true parameter labels. The module determines the corresponding communication signal parameter prediction label for each I / Q component in the channel simulation data, defines a total loss function based on the communication signal parameter prediction label and the true parameter label, and updates the parameters of the high-dimensional feature determination module, the output feature determination module, the enhanced feature determination module, and the communication signal parameter prediction module through the total loss function.

8. The intelligent parallel extraction system for multiple parameters of communication signals according to claim 7, characterized in that, The step of defining the total loss function based on the predicted labels and the true labels of the communication signal parameters specifically includes: according to Define the total loss function, where X is the I / Q component. The actual label for the parameter corresponding to the i-th task branch. As task weight, Let cross-entropy be the loss function. Where C is the total number of categories for the i-th branch task. This represents the true probability that the parameter's true label belongs to the c-th class on the i-th task branch. This represents the predicted probability that the parameter prediction label belongs to class c on the i-th task branch.

9. A computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the steps of the method as claimed in any one of claims 1 to 5.

10. A computer device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method as claimed in any one of claims 1 to 5.

Images