CN114925720A - Small sample modulation signal identification method based on space-time mixed feature extraction network - Google Patents
Small sample modulation signal identification method based on space-time mixed feature extraction network Download PDFInfo
- Publication number
- CN114925720A CN114925720A CN202210416209.0A CN202210416209A CN114925720A CN 114925720 A CN114925720 A CN 114925720A CN 202210416209 A CN202210416209 A CN 202210416209A CN 114925720 A CN114925720 A CN 114925720A
- Authority
- CN
- China
- Prior art keywords
- modulation
- modulation signal
- temporal
- class
- spatial
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Signal Processing (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention discloses a small sample modulation signal identification method based on a space-time mixed feature extraction network, which comprises the following steps: constructing an initial space-time mixed feature extraction network; inputting the training small sample modulation signal set to an initial space-time mixed feature extraction network for training to obtain a target space-time mixed feature extraction network; and inputting the modulation signal set to be identified into a target space-time mixed feature extraction network for identification to obtain the modulation category of each modulation signal in the modulation signal set to be identified. The invention can still ensure the accuracy of signal modulation type identification under the condition of small sample number.
Description
Technical Field
The invention belongs to the technical field of signal modulation identification, and particularly relates to a small sample modulation signal identification method based on a space-time mixed feature extraction network.
Background
Automatic Modulation Classification (AMC) is a basic signal processing technique that aims to identify the Modulation type of a received wireless communication signal. Due to its military significance, it has received much attention and has played an important role in various applications, such as spectrum monitoring, spectrum interference detection and electronic warfare.
AMC is a multi-class classification task from a machine learning perspective. Conventional AMC algorithms fall into two categories, namely likelihood-based methods and feature-based methods. Both methods can achieve efficient identification of AMCs. While semi-supervised AMC methods may be effective in avoiding over-reliance on labeled training samples, they still require a large number of training samples for each class to assist in model training. In practical AMC tasks, it can be difficult and costly to collect enough signal samples for some types of modulated signals. In more extreme cases, only a small number of modulation class signal samples are available, the existing AMC method is no longer effective, and the lack of samples presents a challenge to the existing AMC method.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a small sample modulation signal identification method based on a space-time mixed feature extraction network. The technical problem to be solved by the invention is realized by the following technical scheme:
the embodiment of the invention provides a small sample modulation signal identification method based on a space-time mixed feature extraction network, which comprises the following steps:
constructing an initial space-time mixed feature extraction network;
inputting a training small sample modulation signal set to the initial space-time mixed feature extraction network for training to obtain a target space-time mixed feature extraction network;
inputting a modulation signal set to be identified into the target space-time mixed feature extraction network for identification to obtain the modulation category of each modulation signal in the modulation signal set to be identified;
the modulation category of each type of modulation signal in the modulation signal set to be identified is different from the modulation category of each type of modulation signal in the training small sample modulation signal set; the training small sample modulation signal set comprises a training support set and a training query set; the modulation signal set to be identified comprises a tagged support set and a query set to be identified; the corresponding training process and the recognition process both comprise:
extracting spatial features and temporal features of a modulation signal set by using a space-time mixed feature extraction network; respectively calculating class prototypes corresponding to the spatial features and the temporal features according to the support set; respectively calculating Euclidean distances between the query set and class prototypes corresponding to the spatial features and the temporal features of the query set; converting the corresponding Euclidean distance into the probability of the corresponding modulation category by utilizing a softmax function; jointly calculating a hybrid prediction probability according to the probabilities of the converted modulation classes; wherein the modulation signal set comprises the training small sample modulation signal set and the modulation signal set to be identified; the support set comprises the training support set and the tagged support set; the query set comprises the training query set and the query set to be identified;
the training process further comprises: iteratively adjusting the initial spatio-temporal mixed feature extraction network according to the mixed prediction probability until an iteration stop condition is met to obtain the target spatio-temporal mixed feature extraction network;
the identification process further comprises: and searching the maximum probability value in the mixed prediction probability, and taking the modulation category corresponding to the maximum probability value as the modulation category of the query set to be identified.
In one embodiment of the invention, the spatio-temporal hybrid feature extraction network comprises a spatial feature extractor, a temporal feature extractor, an Euclidean classifier, and a hybrid inference module, wherein,
the spatial feature extractor is used for extracting spatial features of the modulation signal set;
the time characteristic extractor is used for extracting the time characteristics of the modulation signal set;
the Euclidean classifier is used for calculating class prototypes corresponding to the spatial features and the temporal features of the Euclidean classifier according to the support set; respectively calculating Euclidean distances between the query set and class prototypes corresponding to the spatial features and the time features of the query set;
the hybrid reasoning module is used for converting the corresponding Euclidean distance into the prediction probability of the corresponding modulation category by utilizing a softmax function; and jointly calculating the mixed prediction probability according to the prediction probabilities of the converted modulation classes.
In one embodiment of the invention, the spatial feature extractor comprises a first subspace feature extractor, a second subspace feature extractor, and a stitching module, wherein,
the first subspace feature extractor is used for extracting the spatial features of the I-path modulation signals in the modulation signal set;
the second subspace feature extractor is used for extracting the spatial features of the Q-path modulation signals in the modulation signal set;
and the splicing module is used for splicing the spatial characteristics of the I-path modulation signal and the spatial characteristics of the Q-path modulation signal to obtain the spatial characteristics output by the spatial characteristic extractor.
In one embodiment of the present invention, each of the first subspace feature extractor and the second subspace feature extractor comprises a plurality of convolution pooling units connected in sequence, and a full connection layer connected to the last convolution pooling unit; wherein, the first and the second end of the pipe are connected with each other,
each convolution pooling unit comprises a convolution layer, a ReLU activation layer, a batch normalization layer and a maximum pooling layer which are connected in sequence.
In one embodiment of the invention, the temporal feature extractor comprises a plurality of sequentially connected residual blocks, each residual block comprising a plurality of sequentially connected extended causal convolution activation units, and an adder connected to the last extended causal convolution activation unit; wherein, the first and the second end of the pipe are connected with each other,
each extended causal convolution activation unit comprises an extended causal convolution layer, a batch normalization layer, a ReLU activation layer and a Dropout layer which are connected in sequence; wherein the input end of the adder is connected with the input end of the extended cause and effect convolution layer in the first extended cause and effect convolution activation unit and the output end of the Dropout layer in the last extended cause and effect convolution activation unit.
In an embodiment of the present invention, the spatial feature extractor extracts the spatial features by mapping, and the corresponding mapping function formula is expressed as:
wherein x is i Representing the i-th modulation signal, f, of said set of modulation signals Spatial (x i ) Represents x i The spatial feature extractor of (2) extracts a mapping function of spatial features, encoder (x) i ) Representing a functional mapping, particularly 2 x 128 x i Mapped as 1 x 100 in spatial feature space
In one embodiment of the present invention, the time feature extractor extracts the time feature by a mapping method, and the corresponding mapping function formula is expressed as:
wherein f is Temporal (x i ) Represents x i The time feature extractor of (2) extracts a mapping function of the time feature, specifically 2 × 128 x i Mapped to 1 x 16 in the temporal feature space
In one embodiment of the present invention, the class prototype formula corresponding to the spatial feature of the support set is calculated as follows:
the prototype formula of the class corresponding to the time characteristic of the support set is calculated according to the support set and is expressed as follows:
wherein, c k-S Representing class prototypes, S, in the space of features supporting the k-th class modulation signals in the set k Represents the kth class modulation signal set, | S, in the support set k I represents S k The total number of modulation signals in the medium,
denotes S k The (i) th modulation signal of (a),
to represent
The corresponding label is marked with a corresponding label,
to represent
A mapping function of the spatial feature extractor of c k-T Representing a class prototype of a k-th class modulation signal in the support set in a temporal feature space,
represent
The mapping function of the temporal feature extractor of (1).
In one embodiment of the present invention, the predictive probability formula for converting the corresponding euclidean distance into the corresponding modulation class using the softmax function is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
representing the predicted probability that the spatial features of the k-th class of modulation signals in the query set correspond to the modulation class, C representing the number of modulation classes of the signals in the query set,
representing a query set
The euclidean distance to the class prototype corresponding to the spatial signature of its class k modulation signal,
·|| 2 the vector 2-norm is evaluated,
representing the predicted probability that the temporal characteristic of the kth class of modulated signals in the query set corresponds to the modulation class,
representing a query set
The euclidean distance to the class prototype corresponding to the temporal characteristics of its class k modulated signal,
in one embodiment of the present invention, the hybrid predictive probability formula is jointly calculated from the predictive probabilities of the modulation classes translated at that time as:
wherein the content of the first and second substances,
the probability of a hybrid prediction is represented,
the invention has the beneficial effects that:
the invention provides a small sample modulation signal recognition method based on a space-time mixed feature extraction network, which provides a new network structure, namely a space-time mixed feature extraction network, extracts the space feature and the time feature of a modulation signal set in the training or recognition process through the space-time mixed feature extraction network, calculates class prototypes corresponding to the space feature and the time feature by using a support set in the modulation signal set, calculates Euclidean distances between a query set in the modulation signal set and the class prototypes corresponding to the space feature and the time feature, calculates prediction probabilities in space feature space and time feature space by using the Euclidean distances, calculates mixed prediction probabilities by combining the prediction probabilities in space feature space and space feature space, and updates the network weight of the space-time mixed feature extraction network according to the mixed prediction probabilities in the training, and in the identification, the modulation category of the modulation signal in the query set to be identified is determined according to the mixed prediction probability.
Therefore, the method can fully extract the characteristics in the time characteristic space and the space characteristic space, and has excellent automatic characteristic extraction and characteristic fusion capabilities compared with the conventional mode of simply extracting the characteristics in the space or the time characteristic space, so that the accuracy of signal modulation type identification can be effectively improved by utilizing more signal characteristics; based on a new network structure, the invention also adopts a new reasoning strategy: the prediction probabilities in the time characteristic space and the space characteristic space are synthesized to jointly calculate a mixed prediction probability, and the mixed prediction probability is used for updating the network weight of the space-time mixed feature extraction network so as to train to obtain a more accurate target space-time mixed feature extraction network for subsequent recognition; in the identification process, a more accurate target space-time mixed feature extraction network obtained by training is used for identification, the category with the maximum mixed prediction probability is used as the modulation category of the to-be-identified inquiry concentrated modulation signal, and the identification accuracy of the to-be-identified inquiry concentrated modulation signal can be ensured.
Meanwhile, because the spatio-temporal mixed feature extraction network is adopted, under the condition that the number of modulation signal training samples is deficient, even if spatial or temporal features are not fully extracted, the class prototype is calculated by using the support set, and the prediction probability of the query set is calculated by using the class prototype calculated by the support set, so that the small sample learning mode can still ensure the identification accuracy of the spatio-temporal mixed feature extraction network on the signal modulation classes; and the difficulty of extracting the network model training by the space-time mixed characteristics can be reduced by utilizing fewer training samples to carry out network training.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a schematic flow chart of a small sample modulated signal identification method based on a space-time hybrid feature extraction network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a network structure of a spatial feature extractor provided in an embodiment of the present invention;
FIG. 3 is a schematic diagram of a network structure of a temporal feature extractor provided in an embodiment of the present invention;
FIG. 4 is a schematic diagram of a training process of a spatio-temporal hybrid feature extraction network according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of an identification process of extracting a network by using spatio-temporal mixed features according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
Example one
In order to ensure the accuracy of signal modulation type identification under a small sample, the embodiment of the invention provides a small sample modulation signal identification method based on a space-time mixed feature extraction network, which specifically comprises the following steps:
and S10, constructing an initial space-time mixed feature extraction network.
Specifically, when modulation signal identification is usually implemented, an identification signal is mapped to a spatial feature space or a temporal feature space, and since only the spatial feature or the temporal feature mapped by the modulation signal is considered in the spatial feature space or the temporal feature space, and the characteristics of the modulation signal are limited, it is inevitable that the modulation type of the modulation signal cannot be effectively identified by using the extracted spatial feature or temporal feature.
Based on the above analysis, the embodiment of the present invention provides a new network structure, i.e., an instant space-time hybrid feature extraction network, and compared with the prior art, because the modulated signals are mapped to the spatial feature space and the temporal feature space at the same time, more modulated signals can be better demodulated geographically, the features in time and space can be fully extracted, and the accuracy of signal modulation category identification can be still effectively improved even if a small number of signal samples are used. Specifically, the embodiment of the invention provides a novel spatio-temporal mixed feature extraction network, which comprises a spatial feature extractor, a temporal feature extractor, an Euclidean classifier and a mixed reasoning module, wherein,
a spatial feature extractor for extracting spatial features of the modulated signal set;
a time feature extractor for extracting time features of the modulated signal set;
an Euclidean classifier for calculating class prototypes corresponding to the spatial features and the temporal features of the Euclidean classifier according to the support set; respectively calculating Euclidean distances between the query set and class prototypes corresponding to the spatial features and the time features of the query set;
the mixed reasoning module is used for converting the corresponding Euclidean distance into the prediction probability of the corresponding modulation category by utilizing the softmax function; and jointly calculating the mixed prediction probability according to the prediction probabilities of the converted modulation classes.
Therefore, the embodiment of the invention not only considers the spatial characteristic and the time characteristic of the modulation signal at the same time, but also adopts a new reasoning strategy based on the spatial characteristic and the time characteristic by utilizing the small sample learning thought, and specifically comprises the following steps: the support set is used for calculating class prototypes corresponding to the spatial features and the temporal features, then the Euclidean distance between the query set and the corresponding class prototypes is calculated, the Softmax function is used for converting the Euclidean distance into the prediction probability, the smaller the Euclidean distance is, the larger the prediction probability is, therefore, the type of the class prototypes which the modulation signals in the query set are closer to can be predicted, and classification of the modulation signals in the query set is achieved. On the basis, the embodiment of the invention also provides a mixed reasoning strategy, the prediction probability under the spatial characteristic and the prediction probability under the temporal characteristic are comprehensively considered, a mixed prediction probability function is designed, the prediction probability is calculated through the function, and the larger the prediction probability is, the closer the corresponding prediction query concentrated modulation signal is to which type of prototype is, and at the moment, the classification of the query concentrated modulation signal is more accurate. Therefore, by using the inference strategy, even if the spatial characteristics or the time characteristics are not fully extracted, the identification performance of the modulation signals can still be kept by using the small sample signals, and meanwhile, the difficulty of extracting the network training by using the space-time mixed characteristics can be reduced due to the adoption of the small sample signals.
Referring to fig. 2, the spatial feature extractor according to an embodiment of the present invention may include a first subspace feature extractor, a second subspace feature extractor, and a concatenation module, wherein,
the first subspace characteristic extractor is used for extracting the spatial characteristics of the I-path modulation signals in the modulation signal set;
the second subspace characteristic extractor is used for extracting the spatial characteristics of the Q-path modulation signals in the modulation signal set;
and the splicing module is connected with the first subspace feature extractor and the second subspace feature extractor and is used for splicing the spatial features of the I-path modulation signals and the spatial features of the Q-path modulation signals to obtain the spatial features output by the spatial feature extractor.
Therefore, in the spatial feature extraction process, the support set and the query set are divided into the I-path modulation signal and the Q-path modulation signal, the I-path modulation signal and the Q-path modulation signal are respectively input into the first subspace feature extractor and the second subspace feature extractor, the I-path modulation signal and the Q-path modulation signal are mapped to two subspace feature spaces, the amplitude and the phase of the modulation signals can be better demodulated through the I, Q two paths of modulation signals, and the accuracy of the modulation category identification of the modulation signals is effectively improved.
Referring to fig. 2 again, in the embodiment of the present invention, each of the first subspace feature extractor and the second subspace feature extractor includes a plurality of sequentially connected convolution pooling units, and a full connection layer connected to the last convolution pooling unit; wherein the content of the first and second substances,
each convolution pooling unit comprises a convolution layer, a ReLU activation layer, a batch normalization layer and a maximum pooling layer which are connected in sequence.
For example, taking 3 convolution pooling units as an example, each convolution pooling unit has a specific structure as follows: convolution layer (Conv) → ReLU active layer → Batch normalization layer (Batch Norm) → Max Pooling layer (Max pool) → convolution layer → ReLU active layer → Batch normalization layer → Max Pooling layer → Fully Connected layer (full Connected).
In the embodiment of the invention, the sizes of convolution kernels in each convolution pool unit are 2 multiplied by 2, the step length is 3, and the number of convolution kernels of convolution layers is 16; the size of the largest pooling layer in each convolution pool unit is 2 multiplied by 2, and the sliding step length is 2; the signal characteristics after passing through the convolution pool unit are connected to 50 nodes after being paved on a full connection layer; finally, the output of the I, Q two-path modulation signal is spliced on one dimension and then input into an Euclidean classifier.
Referring to fig. 3, the time feature extractor according to the embodiment of the present invention includes a plurality of sequentially connected residual blocks, each of which includes a plurality of sequentially connected extended causal convolution activation units, and an adder connected to the last extended causal convolution activation unit; wherein the content of the first and second substances,
each extended causal convolution activation unit comprises an extended causal convolution layer, a batch normalization layer, a ReLU activation layer and a Dropout layer which are connected in sequence; the input end of the adder is connected with the input end of the extended cause and effect convolution layer in the first extended cause and effect convolution activation unit and the output end of the Dropout layer in the last extended cause and effect convolution activation unit.
For example, taking 3 residual blocks as an example, the structures of the 3 residual blocks are the same, each residual block includes 2 extended causal convolution activating units and an adder, and the specific structure of each residual block is as follows: extended causal convolutional layer (connecting adder) → batch normalization layer → ReLU activation layer → Dropout layer → extended causal convolutional layer → batch normalization layer → ReLU activation layer → Dropout layer → adder. Finally, the outputs of the 3 residual blocks are input to the Euclidean classifier.
The random truncation probability parameters of the Dropout layer in the embodiment of the invention are all 0.2; the extended amount of the extended causal convolutional layer in the first residual block is 1; the extension of the extended causal convolutional layer in the second residual block is 2; the extension of the extended causal convolutional layer in the third residual block is 4.
In the time characteristic extraction process, the embodiment of the invention inputs the modulation signals in the support set and the inquiry set into the time characteristic extractor as complete IQ modulation signals.
It should be noted that the space-time mixed feature extraction network constructed in S10 is used for training, the modulation signal set used in the training process specifically corresponds to a training small sample modulation signal set, the support set specifically corresponds to a training support set in the training small sample modulation signal set, and the query set specifically corresponds to a training query set in the training small sample modulation signal set. And the modulation signal set for identification subsequently corresponds to a modulation signal set to be identified, the support set corresponds to a labeled support set in the modulation signal set to be identified, the query set corresponds to a query set S20 to be identified in the modulation signal set to be identified, and the training small sample modulation signal set is input to the initial space-time mixed feature extraction network for training to obtain a target space-time mixed feature extraction network.
Specifically, in the training process of the embodiment of the invention, the target space-time mixed feature extraction network is obtained by training the initial space-time mixed feature extraction network, and the structures of the initial space-time mixed feature extraction network and the target space-time mixed feature extraction network are shown in fig. 2 and 3.
In the training process, the training data is a training small sample modulation signal set, and the training small sample modulation signal set is divided into a training support set and a training query set. Selecting C x m modulation signals from the training small sample modulation signal set according to categories to be used as a training support set, and the rest
Taking the modulation signals as a training query set, C represents the number of signal modulation classes of a training small sample modulation signal set, m represents the number of modulation signals of each class of modulation classes in a training support set,
represents the total number of modulation signals of which the training small sample modulation signal set type is the k-th type,
values greater than m, for example,
the value of 1000, the value of m of 100,
at 900, the support set is small in number of samples, and the query set is much larger in number of samples than the support set. Wherein, for an IQ-modulated signal, the in-phase part and the quadrature part thereof are independent of each other, both processed into a 2 × 128 shape; training small sample modulation signal set: for the same class of modulated Signal tagsThe same, different classes of modulated signal labels are different.
Correspondingly, referring to fig. 4, the training process includes the following steps:
s201, extracting the space characteristic and the time characteristic of the training small sample modulation signal set by using the initial space-time mixed characteristic extraction network.
Specifically, for the training process, if the spatio-temporal mixed feature extraction network described in S201 is an initial spatio-temporal mixed feature extraction network, the initial spatio-temporal mixed feature extraction network is used to extract the spatial features and the temporal features of the training small sample modulation signal set, specifically:
and (3) sequentially sending each modulation signal in the training support set and the training query set in a signal form with the shape of 2 x 128 into a spatial feature extractor in I, Q paths to obtain two mapping expressions with 50 dimensions, and finally connecting the two mapping expressions in one dimension to obtain a mapping expression with the dimension of 1 x 100. Then, in the embodiment of the present invention, the spatial feature extractor shown in fig. 2 is used to extract the spatial feature through a mapping manner, and the corresponding mapping function formula is expressed as:
wherein x is i Represents the ith modulation signal, f, in the training small sample modulation signal set Spatial (x i ) Represents x i The spatial feature extractor of (2) extracts a mapping function of spatial features, encoder (x) i ) Representing a functional mapping, in particular 2 x 128 x i Mapped as 1 x 100 in spatial feature space
Modulation signal x in each training support set and training query set i Entering the spatial feature extractor will correspondingly obtain a mapping expression
Meanwhile, each modulation signal in the training support set and the training inquiry set is sent to the time characteristic extractor through an integral IQ signal, and the sent signal is mapped into a time characteristic space. Then, in the embodiment of the present invention, the time feature extractor shown in fig. 3 is used to extract the time feature in a mapping manner, and the corresponding mapping function formula is expressed as:
wherein f is Temporal (x i ) Denotes x i The time feature extractor of (2) extracts a mapping function of the time feature, specifically 2 × 128 x i Mapped to 1 x 16 in the temporal feature space
And S202, respectively calculating class prototypes corresponding to the spatial features and the temporal features according to the training support set.
Specifically, the embodiment of the present invention calculates a class prototype formula corresponding to the spatial feature according to the training support set, and the formula is represented as follows:
meanwhile, a class prototype formula corresponding to the time characteristic is calculated according to the training support set and is expressed as follows:
wherein, c k-S Represents a class prototype of a k (k is 1,2, … …, C) class modulation signal in a training support set in a space characteristic space, S k Represents the kth class modulation signal set, | S, in the training support set k | denotes S k The total number of modulation signals in (a) is,
denotes S k The ith modulation signalThe number of the mobile station is,
to represent
The corresponding label is marked with a corresponding label,
represent
A mapping function of the spatial feature extractor of c k-T Represents a class prototype of the k-th class modulation signal in the training support set in the temporal feature space,
to represent
The mapping function of the temporal feature extractor of (1).
And S203, respectively calculating Euclidean distances between the training query set and class prototypes corresponding to the spatial features and the temporal features of the training query set.
Specifically, the euclidean distance formula between the training query set and the class prototype corresponding to the spatial feature of the training query set is calculated and expressed as follows:
meanwhile, the Euclidean distance formula between the training query set and the class prototype corresponding to the spatial feature is calculated and expressed as follows:
wherein, the first and the second end of the pipe are connected with each other,
representing a training query set
The euclidean distance to the class prototype corresponding to the spatial signature of its class k modulation signal,
representing a training query set
Euclidean distance to the class prototype corresponding to the time characteristic of the kth class modulation signal, | | · | calcualting 2 Representing the vector 2-norm.
And (3) finding the modulation signal corresponding to the minimum distance value of the C class prototypes in the training query set through the formula (5) and the formula (6), and taking the class corresponding to the modulation signal as the prediction class.
In order to further improve the identification accuracy of the small sample modulation signal, the embodiment of the invention also introduces a hybrid reasoning module in a new space-time hybrid feature extraction network, and predicts the probability of modulation categories by using Euclidean distances corresponding to the spatial features and the temporal features together to guide the adjustment of the network weight of the space-time hybrid feature extraction network, trains to obtain the optimal space-time hybrid feature extraction network as a target space-time hybrid feature extraction network, and the specific corresponding training process comprises S204 and S205.
And S204, converting the corresponding Euclidean distance into the prediction probability of the corresponding modulation category by utilizing a softmax function.
Specifically, in the embodiment of the present invention, the euclidean distance corresponding to the spatial feature is converted into the prediction probability formula of the corresponding modulation class by using the softmax function, and the prediction probability formula is expressed as:
meanwhile, the Euclidean distance corresponding to the time characteristic is converted into a prediction probability formula of the corresponding modulation category by utilizing a softmax function, and the prediction probability formula is expressed as follows:
wherein the content of the first and second substances,
the prediction probability of the modulation class corresponding to the spatial feature of the kth class modulation signal in the training query set is represented, C represents the number of the modulation classes of the signal in the training query set,
representing a training query set
The euclidean distance to the class prototype corresponding to the spatial signature of its class k modulation signal,
representing the predicted probability that the temporal features of the kth class of modulation signals in the training query set correspond to the modulation class,
representing a training query set
The euclidean distance to the class prototype corresponding to the temporal characteristics of its class k modulated signal.
And S205, jointly calculating a mixed prediction probability according to the prediction probabilities of the converted modulation classes.
Specifically, according to the probability that the spatial feature converted by the formula (7) corresponds to the modulation class and the prediction probability that the temporal feature converted by the formula (8) corresponds to the modulation class, the embodiment of the present invention jointly calculates the hybrid prediction probability formula as follows:
wherein the content of the first and second substances,
the probability of a hybrid prediction is represented,
and S206, iteratively adjusting the initial space-time mixed feature extraction network according to the mixed prediction probability until an iteration stop condition is met to obtain a target space-time mixed feature extraction network.
Specifically, in the training process of the embodiment of the present invention, the initial iteration number t is 0, the maximum iteration number t is set, for example, t is 1000, when the iteration number t is greater than or equal to 1000, the iteration stop condition is satisfied, the currently modulated initial spatio-temporal mixed feature extraction network is used as the target spatio-temporal mixed feature extraction network, and when the iteration number t is greater than or equal to 1000, the initial spatio-temporal mixed feature extraction network is used as the target spatio-temporal mixed feature extraction network<At 1000 deg.C, the spatial features are assigned to probabilities of modulation classes
Predictive probability of time characteristic corresponding to modulation class
And mixed prediction probabilities
Is used as the loss error L of the initial space-time mixed feature extraction network Hybrid The calculation formula is expressed as:
calculating the formula (10) to obtain a loss value L Hybrid And (4) back propagation, updating the network weights of the spatial feature extractor and the time feature extractor in the iteration process, repeating the steps S201 to S205 until a maximum iteration time stopping condition is met, and taking the adjusted initial space-time mixed feature extraction network as a target space-time mixed feature extraction network. The loss value can be calculated by adopting a cross entropy loss function mode, and a random gradient descent algorithm is adopted in the iteration process, but the method is not limited to the loss function and the iteration algorithm.
And S30, inputting the modulation signal set to be identified into a target space-time mixed feature extraction network for identification to obtain the modulation category of each modulation signal in the modulation signal set to be identified.
Specifically, the modulation signal set to be identified in the embodiment of the present invention includes a tagged support set and a query set to be identified. The modulation category of each type of modulation signal in the modulation signal set to be recognized is different from the modulation category of each type of modulation signal in the training small sample modulation signal set, for example, the training small sample modulation signal set selects RML2016.10A data sets of BPSK, CPFSK, GFSK, PAM4, QAM16, QAM64, QPSK, and WBFM as the training categories, and the modulation signal set to be recognized selects RML2016.10A data sets of 8PSK, AM-DSB, and AM-SSB as the test categories. Selecting a small number of modulation signals from the modulation signal set to be identified corresponding to each type of test category as a tagged support set, for example, only 5 modulation signals need to be selected from each type, and using the rest modulation signals in the modulation signal set to be identified as a query set to be identified; and (3) concentrating the modulation signals to be identified: the labels are the same for the same class of modulated signals, and different classes of modulated signals.
Correspondingly, referring to fig. 5, the identification process includes the following steps:
s301, extracting the spatial characteristics and the time characteristics of a modulation signal set to be identified by using a target space-time mixed characteristic extraction network;
s302, respectively calculating class prototypes corresponding to the spatial features and the time features of the tagged support set;
s303, respectively calculating Euclidean distances between the query set to be identified and class prototypes corresponding to the spatial features and the time features of the query set to be identified;
s304, converting the corresponding Euclidean distance into the prediction probability of the corresponding modulation type by utilizing a softmax function;
and S305, jointly calculating a mixed prediction probability according to the prediction probabilities of the converted modulation classes.
S306, searching the maximum probability value in the mixed prediction probability, and taking the modulation category corresponding to the maximum probability value as the modulation category of the query set to be identified.
S301 to S305 are realized in S201 to S205 in a similar manner, except that S201 to S205 are directed to a training small sample modulation signal set, a training support set and a training inquiry set in the training small sample modulation signal set are both signals with tags, S301 to S305 are directed to a modulation signal set to be recognized, the modulation signal set to be recognized is a modulation signal set with tags in the modulation signal set to be recognized, the inquiry set to be recognized is a signal without tags, and finally, modulation categories of the modulation signals in the inquiry sets to be recognized are also recognized. S305 finds a maximum probability value in the hybrid prediction probability after calculating the hybrid prediction probability, where the maximum probability value corresponds to a minimum euclidean distance and indicates that the probability of the modulation type is the maximum, so that the modulation type corresponding to the maximum probability value can be used as the modulation type of the query set to be identified. In the identification process, the sample size of the support set with the label is very small, but the identification accuracy of the modulation signals in the query set to be identified can still be ensured.
In order to verify the effectiveness of the small sample modulation signal identification method based on the space-time mixed feature extraction network provided by the embodiment of the invention, the following experiments are carried out for verification.
1. Conditions of the experiment
The hardware platform of the simulation experiment of the invention is as follows: the GPU is NVIDIA GeForce RTX 2080Ti and 20 cores, the main frequency is 2.2GHz, and the memory size is 128 GB; the video memory size is 11 GB. The software platform of the simulation experiment of the invention is as follows: the operating system is windows 10.
The training sample set of the simulation experiment of the invention is selected from 8 modulation signals with the signal-to-noise ratio of more than 0dB in the RML2016.10.a data set, namely BPSK, CPFSK, GFSK, PAM4, QAM16, QAM64, QPSK and WBFM.
The test sample set of the simulation experiment of the invention is selected from three modulation signals in the RML2016.10.a data set, namely AM-DSB, AM-SSB and 8 PSK.
The rml2016.10.a data set involves 10 different SNRs, varying from 0dB to 18dB, spaced 2dB apart. Each modulation class contains 1000 samples at each signal-to-noise ratio. Complex-valued IQ signal samples are processed in a 2 x 128 shape, with their in-phase and quadrature parts being independent of each other.
2. Emulated content
In order to verify the identification effect of the method under the condition of limited samples, the method is tested by utilizing a labeled support set and a query set to be identified in the identification process. Table 1 and table 2 list the details of the data sets used in the training phase and the testing phase, respectively. It should be noted that although the network model parameters remain unchanged, different tagged support sets may result in different classification results. Thus, each classification result was obtained from a separate monte carlo trial of 1000 randomly selected tagged support sets.
TABLE 1 data set of training phase
TABLE 2 data set of test phases
The final experimental identification accuracy can reach 98.49%. The experimental result proves that the novel small sample learning method provided by the invention can simultaneously extract the spatial characteristics and the time characteristics from the signal samples by adopting the space-time mixed characteristic extraction network structure, and can realize effective automatic modulation classification only by a small amount of training samples with the label marks. The experimental result on the radio ML data set shows the effectiveness and robustness of the method in a small amount of AMC tasks.
To sum up, the method for identifying small sample modulation signals based on the spatio-temporal mixed feature extraction network proposed by the embodiments of the present invention proposes a new network structure, namely, the spatio-temporal mixed feature extraction network, extracts spatial features and temporal features of a modulation signal set in a training or identification process through the spatio-temporal mixed feature extraction network, calculates class prototypes corresponding to the spatial features and the temporal features by using a support set in the modulation signal set, calculates euclidean distances between a query set in the modulation signal set and class prototypes corresponding to the spatial features and the temporal features, calculates prediction probabilities in spatial feature spaces and temporal feature spaces by using the euclidean distances, calculates a mixed prediction probability by combining the prediction probabilities in the temporal feature spaces and the spatial feature spaces, and updates a network weight of the spatio-temporal mixed feature extraction network according to the mixed prediction probability in the training, and in the identification, the modulation category of the modulation signal in the query set to be identified is determined according to the mixed prediction probability.
Therefore, the embodiment of the invention can fully extract the characteristics in the time characteristic space and the space characteristic space, and has excellent automatic characteristic extraction and characteristic fusion capabilities compared with the existing mode of simply extracting the characteristics in the space or the time characteristic space, thereby effectively improving the accuracy of signal modulation type identification by using more signal characteristics; based on a new network structure, the embodiment of the invention also adopts a new inference strategy: the prediction probabilities in the time characteristic space and the space characteristic space are synthesized to jointly calculate a mixed prediction probability, and the mixed prediction probability is used for updating the network weight of the space-time mixed feature extraction network so as to train to obtain a more accurate target space-time mixed feature extraction network for subsequent recognition; in the identification process, a more accurate target space-time mixed feature extraction network obtained by training is used for identification, the category with the maximum mixed prediction probability is used as the modulation category of the to-be-identified inquiry concentrated modulation signal, and the identification accuracy of the to-be-identified inquiry concentrated modulation signal can be ensured.
Meanwhile, the space-time mixed feature extraction network is adopted, so that even though space or time features are not fully extracted under the condition that the number of modulation signal training samples is deficient, the class prototype is calculated by using the support set, and the prediction probability of the query set is calculated by using the class prototype calculated by using the support set, so that the small sample learning mode can still ensure the identification accuracy of the space-time mixed feature extraction network on the signal modulation classes; and the difficulty of extracting the network model training by the space-time mixed characteristics can be reduced by utilizing fewer training samples to carry out network training.
Referring to fig. 6, an embodiment of the present invention provides an electronic device, including a processor 601, a communication interface 602, a memory 603, and a communication bus 604, where the processor 601, the communication interface 602, and the memory 603 complete mutual communication through the communication bus 604;
the memory 603 is used for storing computer programs;
the processor 601 is configured to implement the steps of the method for identifying a small sample modulated signal based on a spatio-temporal mixed feature extraction network when executing the program stored in the memory 603.
The embodiment of the invention provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the small sample modulation signal identification method based on the space-time mixed feature extraction network are realized.
For the apparatus/electronic device/storage medium embodiment, since it is substantially similar to the method embodiment, the description is relatively simple, and reference may be made to part of the description of the method embodiment for relevant points.
In the description of the present invention, it is to be understood that the terms "first", "second" and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
While the present application has been described in connection with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed application, from a review of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the word "a" or "an" does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.
The foregoing is a further detailed description of the invention in connection with specific preferred embodiments and it is not intended to limit the invention to the specific embodiments described. For those skilled in the art to which the invention pertains, numerous simple deductions or substitutions may be made without departing from the spirit of the invention, which shall be deemed to belong to the scope of the invention.
Claims (10)
1. A small sample modulation signal identification method based on a space-time mixed feature extraction network is characterized by comprising the following steps:
constructing an initial space-time mixed feature extraction network;
inputting a training small sample modulation signal set to the initial space-time mixed feature extraction network for training to obtain a target space-time mixed feature extraction network;
inputting a modulation signal set to be identified into the target space-time mixed feature extraction network for identification to obtain the modulation category of each modulation signal in the modulation signal set to be identified;
the modulation category of each type of modulation signal in the modulation signal set to be identified is different from the modulation category of each type of modulation signal in the training small sample modulation signal set; the training small sample modulation signal set comprises a training support set and a training query set; the modulation signal set to be identified comprises a tagged support set and a query set to be identified; the corresponding training process and the recognition process both comprise:
extracting spatial features and temporal features of a modulation signal set by utilizing a space-time mixed feature extraction network; respectively calculating class prototypes corresponding to the spatial features and the temporal features according to the support set; respectively calculating Euclidean distances between the query set and class prototypes corresponding to the spatial features and the temporal features of the query set; converting the corresponding Euclidean distance into the probability of the corresponding modulation category by utilizing a softmax function; jointly calculating a hybrid prediction probability according to the probabilities of the converted modulation classes; wherein the modulation signal set comprises the training small sample modulation signal set and the modulation signal set to be identified; the support set comprises the training support set and the tagged support set; the query set comprises the training query set and the query set to be identified;
the training process further comprises: iteratively adjusting the initial spatio-temporal mixed feature extraction network according to the mixed prediction probability until an iteration stop condition is met to obtain the target spatio-temporal mixed feature extraction network;
the identification process further comprises: and searching the maximum probability value in the mixed prediction probability, and taking the modulation category corresponding to the maximum probability value as the modulation category of the query set to be identified.
2. The small sample modulated signal recognition method based on spatio-temporal mixed feature extraction network as claimed in claim 1, wherein the spatio-temporal mixed feature extraction network comprises a spatial feature extractor, a temporal feature extractor, an Euclidean classifier and a mixed inference module, wherein,
the spatial feature extractor is used for extracting spatial features of the modulation signal set;
the time characteristic extractor is used for extracting the time characteristics of the modulation signal set;
the Euclidean classifier is used for calculating class prototypes corresponding to the spatial features and the temporal features of the Euclidean classifier according to the support set; respectively calculating Euclidean distances between the query set and class prototypes corresponding to the spatial features and the time features of the query set;
the hybrid reasoning module is used for converting the corresponding Euclidean distance into the prediction probability of the corresponding modulation category by utilizing a softmax function; and jointly calculating a hybrid prediction probability according to the prediction probabilities of the converted modulation classes.
3. The small sample modulation signal identification method based on space-time hybrid feature extraction network as claimed in claim 2, wherein said spatial feature extractor comprises a first subspace feature extractor, a second subspace feature extractor and a concatenation module, wherein,
the first subspace feature extractor is used for extracting the spatial features of the I-path modulation signals in the modulation signal set;
the second subspace feature extractor is used for extracting the spatial features of the Q-path modulation signals in the modulation signal set;
and the splicing module is used for splicing the spatial characteristics of the I-path modulation signal and the spatial characteristics of the Q-path modulation signal to obtain the spatial characteristics output by the spatial characteristic extractor.
4. The method for identifying small sample modulation signals based on the spatio-temporal mixed feature extraction network as claimed in claim 3, wherein the first subspace feature extractor and the second subspace feature extractor each comprise a plurality of convolution pooling units connected in sequence, and a full connection layer connected with the last convolution pooling unit; wherein, the first and the second end of the pipe are connected with each other,
each convolution pooling unit comprises a convolution layer, a ReLU activation layer, a batch normalization layer and a maximum pooling layer which are connected in sequence.
5. The method for identifying small sample modulation signals based on the spatio-temporal hybrid feature extraction network as claimed in claim 2, wherein the temporal feature extractor comprises a plurality of sequentially connected residual blocks, each residual block comprises a plurality of sequentially connected extended causal convolution activation units, and an adder connected with the last extended causal convolution activation unit; wherein the content of the first and second substances,
each extended causal convolution activation unit comprises an extended causal convolution layer, a batch normalization layer, a ReLU activation layer and a Dropout layer which are connected in sequence; the input end of the adder is connected with the input end of the extended causal convolution layer in the first extended causal convolution activation unit and the output end of the Dropout layer in the last extended causal convolution activation unit.
6. The small sample modulation signal recognition method based on spatio-temporal mixed feature extraction network as claimed in claim 2, wherein the spatial feature extractor extracts the spatial features by mapping, and the corresponding mapping function formula is expressed as:
f Spatial (x i )＝encoder(x i ):
wherein x is i Representing the i-th modulation signal, f, of said set of modulation signals Spatial (x i ) Represents x i The spatial feature extractor of (2) extracts a mapping function of spatial features, encoder (x) i ) Representing a functional mapping, in particular 2 x 128 x i Mapped as 1 x 100 in spatial feature space
7. The small sample modulation signal recognition method based on spatio-temporal mixed feature extraction network as claimed in claim 6, wherein the time feature extractor extracts the time feature by mapping, and the corresponding mapping function formula is expressed as:
f Temporal (x i ):
wherein f is Temporal (x i ) Denotes x i The time feature extractor of (2) extracts a mapping function of the time feature, specifically 2 × 128 x i Mapped to 1 x 16 in the temporal feature space
8. The method for identifying small sample modulation signals based on the spatio-temporal mixed feature extraction network as claimed in claim 7, wherein the prototype-like formula corresponding to the spatial features calculated according to the support set is represented as:
the prototype formula of the class corresponding to the time characteristic of the support set is calculated according to the support set and is expressed as follows:
wherein, c k-S Representing class prototypes, S, in the space feature space of the kth class modulation signal in the support set k Represents the kth class modulation signal set, | S, in the support set k I represents S k The total number of modulation signals in the medium,
denotes S k The (i) th modulation signal of (b),
to represent
The corresponding label is marked with a corresponding label,
to represent
A mapping function of the spatial feature extractor of c k-T Representing a class prototype of a k-th class modulation signal in the support set in a temporal feature space,
to represent
The mapping function of the temporal feature extractor of (1).
9. The small sample modulation signal recognition method based on the spatio-temporal hybrid feature extraction network of claim 8, wherein the prediction probability formula for converting the corresponding euclidean distance into the corresponding modulation class using a softmax function is expressed as:
wherein the content of the first and second substances,
representing the predicted probability that the spatial features of the k-th class of modulation signals in the query set correspond to the modulation class, C representing the number of modulation classes of the signals in the query set,
representing a query set
Spatial signature to its class k modulation signalCharacterizing the Euclidean distance between corresponding class prototypes,
||·|| 2 the vector 2-norm is evaluated,
representing the predicted probability that the temporal characteristic of the kth class of modulated signals in the query set corresponds to the modulation class,
representing a query set
The euclidean distance to the class prototype corresponding to the temporal characteristics of its class k modulated signal,
10. the method for identifying small sample modulation signals based on the spatio-temporal mixed feature extraction network as claimed in claim 9, wherein the formula for jointly calculating the mixed prediction probability according to the prediction probability of the modulation classes converted at the time is represented as:
wherein, the first and the second end of the pipe are connected with each other,
the probability of a hybrid prediction is represented,
α 1 ,α 2 ,…,α C a prediction probability, max {. cndot.) table, representing the corresponding modulation class output by the C-class modulation signal through the spatial feature extractor in the query setMaximum value, beta, is shown 1 ,β 2 ,…,β C Representing the predicted probability of the class C modulated signal in the query set passing the corresponding modulation class output by the temporal feature extractor,
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210416209.0A CN114925720A (en) | 2022-04-20 | 2022-04-20 | Small sample modulation signal identification method based on space-time mixed feature extraction network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210416209.0A CN114925720A (en) | 2022-04-20 | 2022-04-20 | Small sample modulation signal identification method based on space-time mixed feature extraction network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114925720A true CN114925720A (en) | 2022-08-19 |
Family
ID=82806217
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210416209.0A Pending CN114925720A (en) | 2022-04-20 | 2022-04-20 | Small sample modulation signal identification method based on space-time mixed feature extraction network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114925720A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115720184A (en) * | 2022-10-08 | 2023-02-28 | 西安电子科技大学 | Small sample signal modulation type identification method based on characteristic distribution |
| CN117354106A (en) * | 2023-12-06 | 2024-01-05 | 中国海洋大学 | Communication signal modulation identification method and system based on heavy parameter causal convolution network |
-
2022
- 2022-04-20 CN CN202210416209.0A patent/CN114925720A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115720184A (en) * | 2022-10-08 | 2023-02-28 | 西安电子科技大学 | Small sample signal modulation type identification method based on characteristic distribution |
| CN115720184B (en) * | 2022-10-08 | 2024-04-19 | 西安电子科技大学 | Small sample signal modulation type identification method based on characteristic distribution |
| CN117354106A (en) * | 2023-12-06 | 2024-01-05 | 中国海洋大学 | Communication signal modulation identification method and system based on heavy parameter causal convolution network |
| CN117354106B (en) * | 2023-12-06 | 2024-03-01 | 中国海洋大学 | Communication signal modulation identification method and system based on heavy parameter causal convolution network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110086737B (en) | Communication signal modulation mode identification method based on graph neural network | |
| US11909563B2 (en) | Method and apparatus for modulation recognition of signals based on cyclic residual network | |
| CN110113288B (en) | Design and demodulation method of OFDM demodulator based on machine learning | |
| CN114925720A (en) | Small sample modulation signal identification method based on space-time mixed feature extraction network | |
| CN112491442B (en) | Self-interference elimination method and device | |
| US11700156B1 (en) | Intelligent data and knowledge-driven method for modulation recognition | |
| WO2021088465A1 (en) | Fast modulation recognition method using multilayer perceptron, and employing data fusion of multiple distribution tests | |
| CN116628566A (en) | Communication signal modulation classification method based on aggregated residual transformation network | |
| CN114943245A (en) | Automatic modulation recognition method and device based on data enhancement and feature embedding | |
| CN114896887A (en) | Frequency-using equipment radio frequency fingerprint identification method based on deep learning | |
| CN113657517A (en) | Attention mechanism and metric learning based few-sample power defect detection method | |
| CN115001937B (en) | Smart city Internet of things-oriented fault prediction method and device | |
| CN116070136A (en) | Multi-mode fusion wireless signal automatic modulation recognition method based on deep learning | |
| CN114818785A (en) | Signal modulation mode identification method based on multi-model fusion | |
| Yang et al. | Conventional Neural Network‐Based Radio Frequency Fingerprint Identification Using Raw I/Q Data | |
| CN114785433A (en) | Channel scene recognition method, network device and storage medium | |
| Li et al. | Signal modulation classification based on deep belief network | |
| Bomberger et al. | Bayesian program learning for modeling and classification of RF emitters | |
| KR102284007B1 (en) | Method and apparatus for intelligent digital modulation classification through k-symbol differential operation preprocessing | |
| CN114285701B (en) | Method, system, equipment and terminal for identifying transmitting power of main user | |
| CN117835246B (en) | Task-oriented privacy semantic communication method | |
| CN115828154B (en) | LPI radar signal identification method, system, equipment and storage medium | |
| CN115630289B (en) | Target identification method and device based on evidence theory | |
| CN114997299B (en) | Radio frequency fingerprint identification method in resource limited environment | |
| CN117975163A (en) | Improved transducer-based wireless spectrum state image classification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |