A wireless radiation source individual identification method

Abstract

The present application relates to the technical field of wireless communication signal processing, and particularly relates to a wireless radiation source individual identification method, comprising: converting augmented samples of received wireless signals into first signal embeddings by using a signal encoder; judging the signal source and generating a seen radiation source identifier by using a seen radiation source identification module; generating a high-distinguishability signal hash code by using an online signal hash module according to the signal embedding; concatenating the signal hash code and the seen radiation source identifier to generate a collision mitigation hash code; and predicting a radiation source individual identity tag online through an index hash table, if the collision mitigation hash code corresponding to the current signal does not exist in the hash table, it is considered that the signal comes from a new radiation source and is stored in the hash table for updating. Through multi-task learning, online signal processing and feature optimization technology, the present application significantly improves the identification accuracy and robustness of the model in the actual complex environment, and has wide practical application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication signal processing, and particularly relates to a wireless radiation source individual identification method. BACKGROUND

[0002] With the explosive growth of the number of access devices and the extensive expansion of application dimensions of modern wireless communication systems, the potential threat of malicious devices is increasingly serious, and accurately distinguishing the individual identity of the device to protect the security of the communication system has become a key challenge. Radiation source individual identification technology aims to distinguish the signals of different devices using the inherent radio frequency fingerprint in wireless signals and has been widely applied in many fields such as cognitive radio, intrusion detection, and identity authentication. With the continuous progress of artificial intelligence theory and technology, radiation source individual identification technology based on deep learning has emerged in recent years. By training a neural network model on a large-scale data set composed of wireless received signals, the identification features in the received signals are automatically extracted and classified, which significantly improves the identification ability of the individual identity of the wireless radiation source.

[0003] However, in practical applications, many communication devices operate in a non-cooperative manner, making it difficult to collect signal samples of all potential radiation sources on a large scale to build a training data set. A large number of radiation sources provide only a small amount of training data or no training data. These radiation sources with few samples can be referred to as new radiation sources, and the identification of these new radiation sources can be modeled as a small sample and zero sample radiation source individual identification problem.

[0004] In the aspect of identifying new radiation sources, most deep learning-based radiation source individual identification methods use a large number of training samples from seen radiation sources to build an identification model, and identify signals from seen radiation sources in the inference stage. When encountering signals from new radiation sources, such methods can only classify all new radiation source signals into a unified unknown category, describing an open set radiation source individual identification patent, but such a solution cannot effectively distinguish the individual identity of new radiation sources. In addition, some existing works can identify new radiation sources with only a small amount of training samples, such as a simple description of a small sample radiation source individual identification patent, but such a solution cannot effectively distinguish the individual identity of new radiation sources without training samples.

[0005] In the aspect of real-time classification and identification of radiation sources, many application scenarios require fast feedback, and the model must be able to process received signals online. Existing clustering-based identification methods learn feature extraction models from seen radiation source signal samples and cluster the features of signals from new radiation sources to identify new radiation sources. For example, a simple description of one or two zero sample identification patents based on clustering, but such a solution follows the offline identification paradigm and requires other new radiation source signals as a reference, which cannot meet the requirements of online identification.

[0006] In the aspect of collision problem of signal hash coding, hash coding is an effective feature representation method, which can map signals to low-dimensional binary space. Although hash coding can preliminarily distinguish the individual identity of the radiation source, due to the lack of sufficient training data of the new radiation source, the existing method is prone to overfitting, so that the model is biased to the distribution of the seen radiation source data in the inference stage, resulting in a high hash code collision probability, and then affecting the classification accuracy. SUMMARY

[0007] The present application aims at the technical bottlenecks of the conventional radiation source individual identification method in the aspects of difficulty in online identification of new radiation sources, serious hash code collision and limited generalization ability, and proposes a wireless radiation source individual identification method, comprising the following steps:

[0008] The augmented sample of the received wireless signal is converted into a first signal embedding by using a signal encoder;

[0009] The signal source is judged by using a seen radiation source identification module, and a seen radiation source identifier is generated;

[0010] The signal hash code with high distinguishability is generated by using an online signal hash module according to the signal embedding;

[0011] The collision relief hash code is generated by cascading the signal hash code and the seen radiation source identifier;

[0012] The radiation source individual identity label is predicted online by using an index hash table, and if the collision relief hash code corresponding to the current signal does not exist in the hash table, it is considered that the signal comes from a new radiation source and is stored in the hash table for updating.

[0013] Further, the signal encoder takes a supervised contrast loss function as an optimization objective in the training process, and optimizes the signal embedding by maximizing the similarity between the features of the same class samples and the difference between the features of different class samples, and the supervised contrast loss function is represented as:

[0014]

[0015] Wherein, The supervised contrast loss function is represented as: N is the number of original signal samples, and after two random slicing operations, each original sample will generate two augmented signal samples, so that the total number of augmented signal samples is 2N; I(p) is the index set of positive samples, and |I(p)| is the cardinality of the index set I(p) of positive samples, that is, the number of samples belonging to the same category as sample p in the present application; e p The signal embedding of the pth positive sample is represented as The augmented signal sample corresponding to the pth positive sample is represented as To generate signal embedding by mapping the signal using a signal encoder e q is the signal embedding of the qth positive sample, q e {I(p) \ p}, i.e., q belongs to the index set of positive samples except the index number p; e k is the signal embedding of the kth sample, k e {(I(p) U I(n)) \ p}, i.e., k is the index set of all samples except the current signal sample index p, and the index set of all samples includes the positive sample index set I(p) and the negative sample index set I(n); represents that the signal embedding e p is enhanced processed; represents that the signal embedding e q is enhanced processed; represents that the signal embedding e k is enhanced processed.

[0016] Further, when the radiation source identification module judges the signal source, the maximum distance of the sample from the reciprocal point in the feature space is taken as the decision metric to judge the sample attribution, and a radiation source identifier a is generated, where a = 1 indicates that the signal comes from a known radiation source, and a = -1 indicates that the signal comes from a new radiation source

[0017] Further, the process of judging whether the signal comes from a known radiation source includes:

[0018] Let each known radiation source be associated with a reciprocal point, i.e., associate the radiation source of the known category u with the reciprocal point P u ;

[0019] Project the first signal embedding through a separation projector to obtain a corresponding second signal embedding;

[0020] Calculate the distance between the second signal embedding and the reciprocal point, if the distance meets the set parameter, then the signal source belongs to the radiation source corresponding to the reciprocal point, and if the distance between the second signal embedding and all known reciprocal points does not meet the set parameter, then a new reciprocal point is associated with the signal.

[0021] Further, the separation projector is trained using an adversarial reciprocal point loss, and the adversarial reciprocal point loss is represented as:

[0022]

[0023] wherein, is the adversarial reciprocal point loss; is the classification error of the separation projector; is the adversarial boundary constraint of the separation projector; and λ is a weight coefficient; yp y p is the true class label for the p-th augmented signal sample; p (u | e p ) represents the first signal embedding e p for the p-th sample; represents the second signal embedding z p and the point P ; represents the class label y p for the sample; R is a learnable model parameter, which in the present application represents the boundary radius of the signal samples in the feature space, and the specific value of the parameter is obtained in the model training process.

[0024] Further, the calculation process of the distance between the signal embedding and the reciprocal point includes:

[0025] d (z p , P u ) = d s (z p , P u ) + d a (z p , P u )

[0026]

[0027] d a (z p , P u ) = -z p ·P u

[0028] wherein d (z p , P u ) represents the distance between the second signal embedding z p and the point P u ; d s (z p , P u ) represents the spatial distance between the second signal embedding z p and the point P u ; n represents the dimension of the embedding space, and the present application takes as a normalization factor to prevent the magnitude of the distance value from being too large under different dimensions, thereby improving the calculation stability; ||·|| 2 represents the calculation of L2 norm; d a (z p , P u ) represents the angular distance between the second signal embedding z p and the point P u .

[0029] Further, the online signal hash module is trained by using a supervised contrast loss function, so that when the online signal hash module performs hash coding according to the signal embedding, the hash codes of signals of the same category are close, and the hash codes of different categories are different. The supervised contrast loss function is represented as:

[0030]

[0031] wherein, is the supervised contrast loss function; z' p is the embedding feature vector of the pth sample, and is represented as h p represents the hash code generated by the online signal hash module according to the pth sample, c p represents the confidence generated by the online signal hash module according to the pth sample, represents the Kronecker product; z' q represents the embedding feature vector of the qth positive sample; z' k is the embedding feature vector of the kth sample; z' p ·z' k represents the inner product operation between the vector z' p and the vector z' k , which is used to calculate the similarity between the embedding feature vectors.

[0032] Further, the first signal embedding is data enhanced by using the embedding enhancer, and the enhanced data is used as the input of the online signal hash module.

[0033] Further, a binary constraint is introduced for the signal hash code, which is used to ensure that the generated signal hash code is close to the binary code, so as to improve the efficiency and stability of the model training. The binary constraint regular term is represented as:

[0034]

[0035] wherein, is the binary constraint regular term; ||h p || represents the L1 norm calculation on the hash code h p ; L represents the length of the hash code, that is, how many binary bits are contained in the hash code of each sample. In the present application, this parameter is a learnable model parameter, and the best value of the parameter is obtained through the training process. The person skilled in the art can also set it according to experience.

[0036] Further, a similarity constraint is introduced for the signal hash code, which ensures that the signal hash codes from the same radiation source have high similarity, and the signal hash codes from different radiation sources are separated as much as possible. The similarity constraint regular term is represented as:

[0037]

[0038] wherein,

[0039] is a similarity constraint regularization term; h q represents the hash code generated by the online signal hash module according to the qth sample; h k represents the hash code generated by the online signal hash module according to the kth sample; h p ·h q represents h p and h q inner product operation, used to calculate the similarity between embedded feature vectors.

[0040] Compared with the prior art, the present application has the following beneficial effects:

[0041] 1. The present application adopts an online radiation source individual identification method based on signal hash, learns general knowledge from seen emitters, assigns discriminative hash codes to signals from different emitters and indexes hash table for classification decision, providing a unified method for online identification of new radiation source individuals suitable for small sample and zero sample, which can provide online feedback and meet the demand of efficient and accurate radiation source individual identification in practical application.

[0042] 2. The present application designs a collision mitigation signal hash model, which uses a signal coding module to extract discriminative embedded features and uses a seen radiation source identification module to generate a seen radiation source identifier representing whether the signal is from a seen radiation source, and cascades to the signal hash code output by the online signal hash module to generate a collision mitigation hash code, thereby reducing the hash code collision probability between seen radiation sources and new radiation sources, improving the identification accuracy of new radiation sources, and enabling the model to efficiently identify signals from new radiation sources even in the absence of new radiation source training data.

[0043] 3. The present application adopts a multi-task learning strategy in the training process, combines embedded feature extraction, hash coding optimization, seen radiation source identification and other tasks to design a loss function, and further enhances the generalization ability of embedded features through joint optimization of adversarial mutual point loss and supervised contrast loss, so that the model can adapt to different channel environments and multiple signal systems, and improve the overall identification performance.

[0044] In summary, the present application aims at the technical bottlenecks of the conventional radiation source individual identification method in terms of difficulty in online identification of new radiation sources, serious hash code collision and limited generalization ability, and proposes an improved scheme based on a collision alleviation signal hash (CASH) model. Through multi-task learning, online signal processing and feature optimization technology, the present application significantly improves the identification accuracy and robustness of the model in actual complex environments, and has wide practical application value. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 A wireless radiation source individual identification method flowchart of the present application;

[0046] Figure 2 A module architecture diagram of the wireless radiation source individual identification method of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0048] The present application proposes a wireless radiation source individual identification method, as shown in Figure 1 , comprising the following steps:

[0049] The augmented sample of the received wireless signal is converted into a first signal embedding by using a signal encoder;

[0050] The signal source is judged by using a seen radiation source identification module, and a seen radiation source identifier is generated;

[0051] The signal embedding is used to perform hash coding by using an online signal hash module, and a high-distinguishability signal hash code is generated;

[0052] The signal hash code and the seen radiation source identifier are concatenated to generate a collision alleviation hash code;

[0053] The radiation source individual identity tag is predicted online by using an index hash table. If the collision alleviation hash code corresponding to the current signal does not exist in the hash table, it is considered that the signal comes from a new radiation source, and the hash table is updated by storing the new radiation source.

[0054] In this embodiment, random slicing, a common augmentation method, is used when acquiring augmented samples. During model training, each signal sample is randomly sliced ​​twice to generate two augmented samples from the original signal sample. Augmented samples from the same radiation source are positive samples, and augmented samples from different radiation sources are negative samples, thereby improving the translation invariance of the signal embedding. During recognition, an augmented sample of the same length is extracted from the center of each signal sample. The signal encoder extracts signal embeddings with strong generalization and class discrimination capabilities to support subsequent hash encoding and recognition tasks. The signal encoder consists of an embedding extractor and an embedding enhancer. The embedding extractor generates a signal representation related to the radio frequency fingerprint as the signal embedding; the embedding enhancer further maps the signal embedding to low-dimensional features to enhance the generalization ability of the signal embedding.

[0055] During model training, the signal encoder primarily uses the supervised contrastive loss function as its optimization objective. This function optimizes signal embedding by maximizing the similarity between features of samples from the same class and the difference between features of samples from different classes. The supervised contrastive loss function is:

[0056]

[0057] in, The supervised contrastive loss function is used; N is the number of original signal samples. After two random slicing operations, each original sample generates two augmented signal samples, so 2N is the total number of augmented signal samples; I(p) is the index set of positive samples, and |I(p)| is the cardinality of the index set I(p) of positive samples, i.e., the number of samples belonging to the same class as sample p; e p The signal embedding for the p-th positive sample is denoted as: This represents the augmented signal sample corresponding to the p-th positive sample. To utilize signal encoders for Perform mapping to generate signal embedding; e q The signal embedding for the q-th positive sample, where q is the value of e. p The embedding indexes of positive sample signals belonging to the same category, where q comes from the set of positive sample indices I(p); e k The signal embedding for the k-th sample is given, where k is the set of indices of all samples except the current signal sample index p, including positive and negative sample indices. Indicates the embedding of the signal e p Enhancement processing is performed; Embedded signal e q Enhancement processing is performed; Indicates the embedding of the signal e k Enhancement processing is performed.

[0058] In the embodiment, the seen radiation source identification module generates a seen radiation source identifier, which reacts whether the received signal belongs to any seen radiation source. The application uses the anti-reciprocal point loss guide model to learn the feature space in which the seen categories and new categories can be obviously separated, so that the embedding of the same category signal sample is more compact, and the embedding of the new category signal is obviously separated from it. When identifying, the maximum distance of the sample from the reciprocal point in the feature space is used as the decision metric to determine the sample attribution, and a radiation source identifier a is generated, where a = 1 indicates that the signal comes from a seen radiation source, and a = -1 indicates that the signal comes from a new radiation source. The identification process includes:

[0059] Let each known radiation source be associated with a reciprocal point, that is, associate the radiation source of the known category u with the reciprocal point P u ;

[0060] Embed the first signal through a separation projector Project to obtain the corresponding second signal embedding;

[0061] Calculate the distance between the second signal embedding and the reciprocal point. If the distance meets the set parameter, the signal source belongs to the radiation source corresponding to the reciprocal point. If the distance between the second signal embedding and all known reciprocal points does not meet the set parameter, a new reciprocal point is associated with the signal.

[0062] The distance between the signal embedding and the category reciprocal point can be divided into spatial distance and angle distance, as follows:

[0063] d(z p ,P u ) = d s (z p ,P u ) + d a (z p ,P u )

[0064]

[0065] d a (z p ,P u ) = -z p ·P u

[0066] Where d(z p ,P u ) represents the distance between the second signal embedding z p and the point P u ; d s (z p ,P u ) represents the distance between the second signal embedding z p and the point P uThe spatial distance between them; n represents the dimension of the embedded space; The normalization factor is used to prevent the distance values ​​from becoming too large across different dimensions, thereby improving computational stability; ||·||2 represents the calculation of the L2 norm; d a (z p ,P u ) indicates that the second signal is embedded in z p With point P u The angular distance between them.

[0067] To ensure the compactness of the seen class samples, the adversarial reciprocity point loss includes a classification error component:

[0068]

[0069] Among them, y p Let p(u|e) be the class label of the p-th augmented signal sample. p ) represents the first signal embedding e of the p-th sample. p The probability of belonging to category u is specifically expressed as:

[0070]

[0071] in, This represents the set of all categories, including both known and newly added radiation source categories; d(z) p ,P v ) indicates that the second signal is embedded in z p With point P v The distance between them, P v This represents the reciprocal point corresponding to category v.

[0072] Furthermore, this invention introduces adversarial boundary constraints to ensure reliable separation of the new radiation source signal, specifically as follows:

[0073]

[0074] in, This indicates that the second signal is embedded in z. p With point The distance between them; This indicates that the category label is y. p The feature center point of the sample; R is a learnable model parameter, representing the boundary radius of the signal sample in the feature space.

[0075] Ultimately, the adversarial reciprocity loss can be expressed as:

[0076]

[0077] Where λ is a weighting coefficient, used to balance the strength of adversarial constraints.

[0078] To reduce training complexity, this embodiment utilizes a multi-task training mode to simultaneously train the signal encoder module and the observed radiation source identification module, thereby enhancing the generalization ability of signal embedding and reducing training costs. The loss in the first stage can be expressed as:

[0079]

[0080] in, The supervised contrastive loss function is used to encode the signal; α is an adjustable weight used for balancing. and The relative importance between them; This represents the final adversarial reciprocity point loss for the identified radiation source recognition module.

[0081] In this embodiment, if multiple signal samples come from the same radiation source, they will be assigned the same hash code; signal samples from different radiation sources will be assigned different hash codes. Specifically, the training process for the online signal hashing module includes:

[0082] Signal embedding is processed through a symbol projector and a confidence projector;

[0083] The signal hash code is optimized by supervised comparison loss to ensure that the generated hash code has strong distinguishability;

[0084] By combining binary constraints and similarity constraints for model regularization, the distinguishability of hash codes can be improved.

[0085] Specifically, signal embedding is processed through a symbol projector and a confidence projector. The symbol projector generates attribute identifiers as signal hash codes, with each dimension representing whether the signal possesses a certain radiation source attribute. The confidence projector calculates the confidence level for the signal attribute decision based on the signal embedding. To ensure the generated hash codes have strong discriminative power, this invention uses supervised contrastive loss to optimize the signal hash codes. The purpose of this loss function is to maximize the similarity of signal hash codes from the same radiation source and minimize the similarity of signal hash codes from different radiation sources. The supervised contrastive loss optimizes the relative distance of signal hash codes, making the hash codes of similar signals as close as possible and the hash codes of different signals as far apart as possible. The specific supervised contrastive loss function can be expressed as follows:

[0086]

[0087] in, is the supervision contrast loss function; N is the number of original signal samples, after two random slicing operations on all original signal samples, two augmented signal samples are generated for each original sample, and therefore 2N is the total number of augmented signal samples; I(p) is an index set of positive samples, |I(p)| is the cardinality of the index set I(p) of positive samples, that is, the number of samples belonging to the same category as the sample p; z p is the embedding feature vector of the pth sample, and is represented as h p represents the hash code generated by the online signal hash module according to the pth sample; c p represents the confidence generated by the online signal hash module according to the pth sample, represents the Kronecker product, which combines the hash code and the confidence to improve the discrimination; q is a positive sample signal embedding index belonging to the same category as e p ; z q represents the embedding feature vector of the qth positive sample; k is an index set of all samples except the current signal sample index p, including positive sample indexes and negative sample indexes; z k is the embedding feature vector of the kth sample; · represents the inner product operation, which is used to calculate the similarity between embedding feature vectors.

[0088] In order to further improve the discrimination of the hash code, the application additionally introduces binary constraint and similarity constraint for the signal hash code. The binary constraint is used to ensure that the generated signal hash code is close to the binary code, so as to improve the efficiency and stability of model training. The similarity constraint ensures that the signal hash codes from the same radiation source have high similarity, while the signal hash codes from different radiation sources are separated as much as possible. The specific binary constraint and similarity constraint regular terms are represented as follows:

[0089]

[0090]

[0091] wherein, is the binary constraint regular term; ||h p || represents the L1 norm calculation on the hash code; L is an adjustable hyperparameter, representing the length of the hash code, that is, how many binary bits are contained in the hash code of each sample. is the similarity constraint regular term; h q represents the hash code generated by the online signal hash module according to the qth sample; h k represents the hash code generated by the online signal hash module according to the kth sample; · represents the inner product operation, which is used to calculate the similarity between embedding feature vectors.

[0092] To ensure that the signal hash codes from different emission sources have good discrimination ability, the embodiment combines the binary constraint regular term and the similarity constraint regular term into the regular term of the online signal hash module, ensures effective avoidance of hash code collision problems, and improves the robustness and precision of the model when facing new signal sources. Specifically, it can be expressed as:

[0093]

[0094] In combination with the regular term composed of the symbol projector, the confidence projector and the double constraint of the online signal hash module, the application significantly reduces the influence of intraclass disturbance, improves the discrimination degree of the signal hash code, and thus improves the accuracy and robustness of the model. Finally, the loss function of the second stage can be expressed as:

[0095]

[0096] Wherein, β is an adjustable weight for balancing the supervised contrast loss and the regularization term, so as to determine whether to focus more on learning the discriminative features of the signal or to strengthen the generalization ability of the model in the optimization process.

[0097] Cascade the signal hash code and the seen radiation source identifier to generate a collision mitigation hash code, which can be specifically expressed as:

[0098] h c = [h, a];

[0099] The collision mitigation hash code not only contains the feature information of the signal, but also contains the additional information of whether the signal comes from the seen radiation source.

[0100] When multiple signal samples come from the same radiation source, their hash codes should be similar, and the collision mitigation hash code after splicing can still effectively distinguish them. However, if signals from different radiation sources have similar hash codes, the identifier will effectively avoid the hash code collision problem, ensuring that even if the hash codes are the same, the source of the signal can still be distinguished based on the identifier.

[0101] In the embodiment, the hash table is used to index the collision mitigation hash code. When a new signal sample arrives, we will check whether the current hash code already exists in the hash table S: if it does not exist in the hash table S, it means that the signal comes from a new radiation source, and we will add it as a new item to the hash table and perform label prediction on the signal; if it already exists in the hash table, it means that the signal comes from a seen radiation source, and the hash table will return the stored identity label. The specific expression is as follows:

[0102]

[0103] wherein, Ind(S, h c ) is the corresponding label of the collision resolution hash code h c in the hash table S, if h c is not in the hash table, the model will regard it as a new radiation source and update it.

[0104] According to the complete process of the present application,

[0105] The present embodiment is directed to the training process and real-time data processing process in the present application. As Figure 2 , the training process specifically includes:

[0106] The signal samples obtained from different radiation sources are processed based on the signal augmentation technology to obtain augmented samples;

[0107] The embedding extractor of the signal encoder obtains the first signal embedding from the augmented samples, the embedding enhancer of the signal encoder performs data enhancement on the first signal embedding, and the supervised contrast loss function is used to train the signal encoder;

[0108] The separation projector of the known radiation source identification module obtains the corresponding second signal embedding according to the input first signal embedding, inputs the second signal embedding into the classifier or other decision module to judge whether the class of the signal belongs to the known class, and marks the identifier, and the known radiation source identification module is trained according to the classification error and the adversarial boundary constraint;

[0109] The output of the embedding enhancer is taken as the input of the online signal hash module, the confidence projector and the symbol projector are used to obtain the confidence and the hash code of the input signal respectively, and the supervised contrast loss function is combined with the binary constraint and the similarity constraint to train the online signal hash module.

[0110] As Figure 2 , the real-time data processing process specifically includes:

[0111] The signal samples obtained from different radiation sources are processed based on the signal augmentation technology to obtain augmented samples;

[0112] The pre-trained embedding extractor obtains the first signal embedding from the augmented samples, and the pre-trained embedding enhancer performs data enhancement on the first signal embedding;

[0113] The known radiation source identification module marks the identifier for the signal according to the first signal embedding, when the signal comes from the known radiation source, the identifier a = 1, and when the signal comes from the unknown radiation source, the identifier a = -1;

[0114] The output of the embedding enhancer is taken as the input of the symbol projector in the online signal hash module, and the hash code corresponding to the signal is obtained;

[0115] The identifier and the hash code of the signal are concatenated together as a collision mitigation hash code for the signal.

[0116] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely divergences of the principles and spirit of the application and that numerous modifications, changes, substitutions, and alterations can be undertaken by one skilled in the art without departing from the principles and spirit of the application, which is defined by the claims as follows and their equivalents.

Claims

1. A method for identifying individual wireless radiation sources, characterized in that, Includes the following steps: An augmented sample of the received wireless signal is converted into a first signal embedding using a signal encoder. The known radiation source identification module determines the source of the signal and generates a known radiation source identifier. When the known radiation source identification module determines the source of the signal, it uses the maximum distance of the sample from the reciprocal point in the feature space as the decision metric to determine the sample's ownership and generates a radiation source identifier a. When a=1, it means that the signal comes from a known radiation source, and when a=-1, it means that the signal comes from a newly added radiation source. The online signal hashing module is used to perform hash encoding based on the signal embedding to generate a highly distinguishable signal hash code; The cascaded signal hash code is combined with the known radiation source identifier to generate a collision mitigation hash code; The individual identity label of the radiation source is predicted online by indexing a hash table. If the collision mitigation hash code corresponding to the current signal does not exist in the hash table, the signal is considered to come from a newly added radiation source and is stored in the hash table for updating.

2. The method for individual identification of wireless radiation sources according to claim 1, characterized in that, During training, the signal encoder uses a supervised contrastive loss function as its optimization objective. This function optimizes signal embedding by maximizing the similarity between features of samples from the same class and the difference between features of samples from different classes. The supervised contrastive loss function is expressed as follows: in, For supervised contrastive loss function; This represents the number of original signal samples. The set of indices for positive samples. The set of indices for positive samples The cardinality; For the first The signal embedding of a positive sample is denoted as... , Indicates the first The augmented signal samples corresponding to each positive sample To utilize signal encoders for Perform mapping to generate signal embedding. ; For the first Signal embedding of a positive sample, ; For the first Signal embedding of each sample, , The set of indices for negative samples; Indicates signal embedding Enhancement processing is performed; signal embedding Enhancement processing is performed; Indicates signal embedding Enhancement processing is performed.

3. The method for individual identification of wireless radiation sources according to claim 1, characterized in that, The process of determining whether a signal originates from a known radiation source includes: Associating each known radiation source with a reciprocal point, that is, associating radiation sources of known class u with reciprocal points. ; The first signal is embedded through the split projector. Projection is performed to obtain the corresponding second signal embedding; Calculate the distance between the second signal embedding and the reciprocal point. If the distance meets the set parameters, the signal source belongs to the radiation source corresponding to the reciprocal point. If the distance between the second signal embedding and all known reciprocal points does not meet the set parameters, a new reciprocal point is added for the signal to be associated.

4. The method for individual identification of wireless radiation sources according to claim 3, characterized in that, Adopting adversarial reciprocity point loss for separate projectors During training, the adversarial reciprocity loss is represented as: in, To counteract the loss of reciprocal points; To separate the classification error of the projector; To overcome the boundary constraints of the separate projector; These are the weighting coefficients; This represents the number of original signal samples. For the first The true category label of each augmented signal sample; Indicates the first The first signal embedding of each sample The probability of belonging to category u; Indicates the second signal embedding With point The distance between them; Indicates category label as The feature center points of the sample; R is the learnable model parameters.

5. A method for identifying an individual wireless radiation source according to claim 3 or 4, characterized in that, The calculation process for the distance between the signal embedding and the reciprocity point includes: in, Indicates the second signal embedding With point The distance between them; Indicates the second signal embedding With point Spatial distance between them; Indicates the dimension of the embedding space; This indicates the calculation of the L2 norm; Indicates the second signal embedding With point The angular distance between them.

6. The method for individual identification of wireless radiation sources according to claim 1, characterized in that, The online signal hashing module is trained using a supervised contrastive loss function. This ensures that when the online signal hashing module performs hash encoding based on signal embedding, the hash codes of signals of the same class are similar, while the hash codes of signals of different classes differ. The supervised contrastive loss function is expressed as follows: in, For supervised contrastive loss function; This represents the number of original signal samples. The set of indices for positive samples. The set of indices for positive samples The cardinality; For the first The embedded feature vector of each sample is represented as: , The online signal hash module indicates that, according to the first... The hash code generated for each sample The online signal hash module indicates that, according to the first... The confidence score generated for each sample Indicates the Kronecker product. ; Indicates the first Embedded feature vectors of positive samples, ; For the first The embedded feature vector of each sample, , The set of indices for negative samples; Representing vectors with vector Inner product operations between them.

7. The method for individual identification of wireless radiation sources according to claim 6, characterized in that, The first signal embedding is enhanced using an embedding enhancer, and the enhanced data is used as the input to the online signal hashing module.

8. The method for individual identification of wireless radiation sources according to claim 6, characterized in that, A binary constraint is introduced for the signal hash code. The binary constraint is used to ensure that the generated signal hash code is close to the binary code, thereby improving the efficiency and stability of model training. The binary constraint regularization term is expressed as: in, For binary constraint regularization terms; The online signal hash module indicates that, according to the first... The hash code generated for each sample Represents the hash code Perform L1 norm calculation; Indicates the length of the hash code.

9. A method for identifying an individual wireless radiation source according to claim 6 or 8, characterized in that, A similarity constraint is introduced for the signal hash codes. This constraint ensures that the hash codes of signals from the same radiation source have high similarity, while the hash codes of signals from different radiation sources are separated as much as possible. The regularization term of the similarity constraint is expressed as follows: in, For similarity constraint regularization terms; The online signal hash module indicates that, according to the first... Hash codes generated from each sample; This represents the hash code generated by the online signal hashing module based on the k-th sample; express and Inner product operation.

Images