CN-121997124-A - Multi-parameter intelligent parallel extraction method for communication signals

CN121997124ACN 121997124 ACN121997124 ACN 121997124ACN-121997124-A

Abstract

The embodiment of the invention discloses a multi-parameter intelligent parallel extraction method for communication signals. The method comprises the steps of firstly obtaining a signal to be tested and extracting I/Q components of the signal to be tested, 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, enhancing the shared features through a channel attention mechanism in the branches to obtain output features, transforming the output features of all the task branches through a learnable linear layer to form a key matrix and a value matrix, transforming the current task features through the learnable linear layer to form a query matrix, dividing the input features into a plurality of heads through a multi-head attention mechanism, obtaining output of a single head through self-attention operation of each head based on scaling dot product, integrating the results of all the heads to obtain enhancement features of a current task, and determining a communication signal parameter prediction tag according to the enhancement features to realize multi-parameter efficient parallel extraction.

Inventors

QI PEIHAN
PAN CHENLU
YIN KAI
SUN WEN
JIANG TAO
LIANG LINLIN
MENG YONGCHAO
Meng Menyu

Assignees

西安电子科技大学

Dates

Publication Date: 20260508
Application Date: 20251231

Claims (10)

1. A method for intelligent parallel extraction of multiple parameters of a communication signal, the method comprising: Acquiring a signal to be detected, and extracting an I/Q component of the signal to be detected; Processing the I/Q component through a multi-scale feature extraction structure to obtain a high-dimensional feature; extracting shared features of the high-dimensional features through a plurality of residual block groups, respectively inputting each shared feature into a plurality of parallel task branches, and enhancing the shared features in each task branch through a channel attention mechanism to obtain output features of a current task, wherein each task branch corresponds to a communication signal parameter to be extracted; Transforming output features of all task branches through a learnable linear layer by utilizing a multi-head attention mechanism to form a key matrix and a value matrix, transforming current task features through the learnable linear layer to form a query matrix, dividing the key matrix, the value matrix and the query matrix into a plurality of heads, obtaining output of a single head by each head based on scaling dot product self-attention operation, and integrating output results of the heads to obtain enhancement features of the current task after interaction of different task features; And determining a communication signal parameter prediction tag according to the enhancement features of each task branch.
2. The method for intelligent parallel extraction of multiple parameters of communication signals according to claim 1, wherein the multi-scale feature extraction structure comprises a first convolution layer, a first feature extraction structure, a second convolution layer and a second feature extraction structure, and the processing of the I/Q components through the multi-scale feature extraction structure to obtain high-dimensional features specifically comprises: Processing the I/Q component through the first convolution layer to obtain initial characteristics, respectively carrying out nonlinear mapping on the initial characteristics through parallel sub-paths of a first characteristic extraction structure, extracting local structure information of the initial characteristics through convolution kernels with different preset sizes by each path, splicing the local structure information of the initial characteristics output by each path through channel dimensions, and then carrying out downsampling through a maximum pooling operation to obtain primary characterization; And processing the primary characterization through the second convolution layer to obtain intermediate features, further introducing a maximum pooling operation path into the first feature extraction structure to form a second feature extraction structure, respectively processing the intermediate features through the second feature extraction structure, splicing the output of each path through the channel dimension, and then performing downsampling through the maximum pooling operation to obtain high-dimensional features.
3. The method for intelligent parallel extraction of multiple parameters of communication signals according to claim 1, wherein the residual block group includes two residual blocks, the shared feature of the high-dimensional feature is extracted by multiple residual block groups, the shared feature is input into multiple parallel task branches, and in each task branch, the shared feature is enhanced by a channel attention mechanism to obtain an output feature of a current task, and the method specifically includes: Respectively extracting sharing features of the high-dimensional features through two residual blocks of the residual block group, wherein the sharing features comprise a first sharing feature and a second sharing feature; Inputting the shared features into a plurality of parallel task branches, and processing the first shared features in each task branch through a channel attention mechanism to generate channel attention weights; and multiplying the channel attention weight with the second shared feature element by element to obtain the enhanced output feature of the current task.
4. The method for intelligent parallel extraction of multiple parameters of communication signals according to claim 1, wherein the dividing the key matrix, the value matrix and the query matrix into a plurality of heads, each head obtaining an output of a single head based on scaling dot product self-attention operation, specifically comprises: According to Determining an attention weight matrix, wherein 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 method for intelligent parallel extraction of multiple parameters of communication signals according to claim 1, wherein the steps of obtaining a signal to be detected and extracting an I/Q component of the signal to be detected specifically include: obtaining a signal to be detected, and carrying out normalization processing on the signal to be detected; According to Determining an I component of the signal; According to The Q component of the signal is determined, wherein, As an I-component of the signal, As the Q-component of the signal, Is the normalized signal, is the complex conjugate.
6. A communication signal multiparameter intelligent parallel extraction system, the system comprising: The I/Q component extraction module is used for acquiring a signal to be detected and extracting an I/Q component of the signal to be detected; The high-dimensional feature determining module is used for processing the I/Q component through a multi-scale feature extraction structure to obtain high-dimensional features; The output characteristic determining module is used for extracting the shared characteristic of the high-dimensional characteristic through a plurality of residual error block groups, inputting each shared characteristic into a plurality of parallel task branches respectively, and enhancing the shared characteristic in each task branch through a channel attention mechanism to obtain the output characteristic of the current task, wherein each task branch corresponds to one communication signal parameter to be extracted respectively; The enhanced feature determining module is used for transforming the output features of all task branches through a learnable linear layer by utilizing a multi-head attention mechanism to form a key matrix and a value matrix, transforming the current task features through the learnable linear layer to form a query matrix, dividing the key matrix, the value matrix and the query matrix into a plurality of heads, obtaining the output of a single head by self-attention operation of each head based on a scaling dot product, and integrating the output results of the heads to obtain the enhanced features of the current task after interaction of different task features; and the communication signal parameter prediction module is used for determining a communication signal parameter prediction label according to the enhancement characteristic of each task branch.
7. The communication signal multi-parameter intelligent parallel extraction system of claim 6, further comprising: The system comprises a training module, a high-dimensional feature determining module, an output feature determining module, an enhancement feature determining module and a communication signal parameter predicting module, wherein the training module is used for processing a bit stream with preset length as an original signal based on different signal processing modes to obtain a channel simulation data set, the channel simulation data set comprises I/Q components of the original signal under the different signal processing modes and corresponding parameter configurations when the I/Q components are generated, the parameter configurations are used as parameter real tags, the corresponding communication signal parameter predicting tags are determined according to each I/Q component in the channel simulation data, a total loss function is defined according to the communication signal parameter predicting tags and the parameter real tags, and the parameter updating is carried out on the high-dimensional feature determining module, the output feature determining module, the enhancement feature determining module and the communication signal parameter predicting module through the total loss function.
8. The communication signal multi-parameter intelligent parallel extraction system according to claim 7, wherein the prediction tag and the parameter real tag define a total loss function according to the communication signal parameters, and specifically comprising: According to Defining a total loss function, wherein X is an I/Q component, The parameter corresponding to the ith task branch is the real label, As a weight of the task(s), In order to cross-entropy loss function, Wherein C is the total category number of the ith branch task, Representing the true probability that the parameter true label belongs to class c on the ith task branch, Representing the prediction probability that the parameter prediction tag belongs to class c on the ith task branch.
9. A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method of 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 of any of claims 1 to 5.

Description

Multi-parameter intelligent parallel extraction method for communication signals Technical Field The invention relates to the technical field of communication signal processing, in particular to a multi-parameter intelligent parallel extraction method for communication signals. Background With the development of wireless communication technology, the scarcity of spectrum resources makes spectrum monitoring and spectrum sensing particularly important. In spectrum sensing, communication signal parameter extraction is a key link for realizing signal detection and analysis, and its core tasks generally include modulation type identification, coding parameter extraction, signal-to-noise ratio evaluation, and the like. Specifically, the modulation type identification aims at judging the modulation mode adopted by the signal, the coding parameter extraction involves identifying the channel coding mode in the signal, such as polarization codes, convolution codes and the like, and the signal-to-noise ratio estimation is the quantitative evaluation of the quality of the received signal and provides an important basis for the subsequent signal processing. The accurate extraction of the parameters forms the basis of intelligent understanding and reconstruction of the communication signals, and has key significance in the fields of military reconnaissance, spectrum supervision, intelligent communication and the like. The traditional processing method mostly adopts a hierarchical processing mode, namely, firstly, the modulation type of a signal is identified, then, the channel coding mode is further analyzed based on the demodulated bit stream, and finally, the communication parameters (such as signal to noise ratio and the like) are estimated. However, this mode of hierarchical processing has significant drawbacks. Firstly, the algorithm design complexity and the calculation amount are increased, the processing efficiency is low, and secondly, the inaccuracy of any ring can influence the performance of the subsequent links due to the mutual dependence of the identification results of all links, thereby causing accumulated errors. In recent years, with the development of intelligent methods such as deep learning, partial research attempts are made to introduce a neural network to improve recognition performance, but most models are focused on extracting only a single parameter, and intrinsic association and shared features among multiple parameters in signals cannot be fully mined, so that model redundancy is caused, and the actual requirements of real-time, efficient and integrated analysis are difficult to meet. Disclosure of Invention Based on this, it is necessary to provide a multi-parameter intelligent parallel extraction method for communication signals in order to solve the above problems. A method for multi-parameter intelligent parallel extraction of communication signals, the method comprising: Acquiring a signal to be detected, and extracting an I/Q component of the signal to be detected; Processing the I/Q component through a multi-scale feature extraction structure to obtain a high-dimensional feature; extracting shared features of the high-dimensional features through a plurality of residual block groups, respectively inputting each shared feature into a plurality of parallel task branches, and enhancing the shared features in each task branch through a channel attention mechanism to obtain output features of a current task, wherein each task branch corresponds to a communication signal parameter to be extracted; Transforming output features of all task branches through a learnable linear layer by utilizing a multi-head attention mechanism to form a key matrix and a value matrix, transforming current task features through the learnable linear layer to form a query matrix, dividing the key matrix, the value matrix and the query matrix into a plurality of heads, obtaining output of a single head by each head based on scaling dot product self-attention operation, and integrating output results of the heads to obtain enhancement features of the current task after interaction of different task features; And determining a communication signal parameter prediction tag according to the enhancement features of each task branch. The multi-scale feature extraction structure comprises a first convolution layer, a first feature extraction structure, a second convolution layer and a second feature extraction structure, wherein the I/Q component is processed through the multi-scale feature extraction structure to obtain a high-dimensional feature, and the multi-scale feature extraction structure specifically comprises the following steps: Processing the I/Q component through the first convolution layer to obtain initial characteristics, respectively carrying out nonlinear mapping on the initial characteristics through parallel sub-paths of a first characteristic extraction structure, extracting local structure informatio