Cross-modulation robust radio frequency fingerprint identification method based on time-frequency double-branch hybrid enhancement

Abstract

The invention discloses a cross-modulation robust radio frequency fingerprint identification method based on time-frequency double-branch hybrid enhancement. According to the method, by constructing a time domain and frequency domain double-branch feature extraction model, the features of radio frequency signals are effectively extracted, so that high-robustness and high-accuracy identification of radio frequency equipment in a cross-modulation environment is realized. The method comprises the following steps: firstly, constructing a time domain and frequency domain feature extraction model for source domain radio frequency equipment identification, training through a multi-source domain data set and by introducing a data enhancement technology, and learning universal domain invariant features; then, the source domain depth feature extraction model is directly applied to a target domain, and a target domain sample does not need to be known in advance. Finally, radio frequency fingerprint identification in a cross-modulation scene is realized, and the problem that the equipment category cannot be accurately identified due to data distribution difference is solved.

Description

Technical Field

[0001] The present invention relates to the fields of big data, signal processing and Internet of Things security, and in particular to a cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement. Background Art

[0002] With the rapid development of wireless communication technology, radio frequency fingerprint recognition, as an emerging device authentication method, has received widespread attention. Radio frequency fingerprint recognition uniquely identifies a device by analyzing the characteristics of the radio frequency signal emitted by the device during wireless communication, and is then used in application scenarios such as network security and device authentication. However, radio frequency fingerprint recognition faces significant challenges in terms of robustness and recognition accuracy in cross-modulation situations. Traditional radio frequency fingerprint recognition methods usually rely on data from a single field for training, and are difficult to cope with changes in signal characteristics under different modulation modes, resulting in a significant drop in recognition performance in cross-modulation environments.

[0003] Therefore, it is of great practical application value to develop a radio frequency fingerprint recognition method that can maintain high robustness and high accuracy in a cross-modulation environment. The present invention constructs a dual-branch feature extraction model in the time domain and frequency domain, and combines hybrid enhancement technology to effectively extract the characteristics of the radio frequency signal, thereby achieving high robustness and high accuracy recognition in a cross-modulation environment. Deep learning and the source domain IoT device modulation signal data set are used to construct a deep feature extraction model and a feature recognition model suitable for source domain IoT device identification, and then the data enhancement technology is used to realize IoT device identification in different data distribution scenarios. This method can uniquely identify IoT devices for IoT device access authentication, prevent malicious devices from accessing the IoT, and ensure the security of the IoT. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement, which solves the problem that in the identification of IoT devices such as WiFi and LoRa, the modulation signal data distribution differences of the target domain IoT devices are large and cannot be accurately identified.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement specifically comprises the following steps:

[0007] Step 1: construct a time domain and frequency domain feature extraction model and feature recognition model in the source domain;

[0008] Step 2: Use multi-source domain IoT device modulation signal datasets to jointly train time domain and frequency domain feature extraction models, and introduce data enhancement technology to obtain source domain deep feature extraction models and feature recognition models;

[0009] Step 3: directly apply the source domain deep feature extraction model to the target domain to obtain the target domain deep feature extraction model without knowing the target domain samples in advance;

[0010] Step 4: Input the modulated signal of the IoT device to be identified into the target domain feature recognition model to obtain the recognition result of the RF device to be identified.

[0011] Furthermore: in step 1, the time domain and frequency domain feature extraction model of the source domain as a whole is composed of a group of InceptionBlock modules, a global average pooling layer and a feature fusion layer, and the time domain and frequency domain feature recognition model of the source domain includes 1 layer of random dropout operation and 1 layer of full connection operation connected in sequence.

[0012] Further: In step 2, the source domain IoT device modulation signal dataset is composed of several groups of complex baseband signals x of source domain IoT devices. s and its corresponding category label y in one-hot encoding form s constitute.

[0013] Further: The joint training specifically includes the following steps:

[0014] Step 2-1: transform the complex baseband signal x of the source domain IoT device s Input initializes the source domain deep feature extraction model and obtains x s The depth feature is denoted as x s,L-2 , x s,L-2 Input initializes the source domain feature recognition model and obtains x s The predicted probability value of The minimization of the cross entropy loss function is used as the objective function for initializing the source domain deep feature extraction model and feature recognition model.

[0015]

[0016] Where C is the number of IoT device categories in the source domain, for y s The i-th code element in For x s The predicted probability value for the i-th category;

[0017] Step 2-2: Generate new samples by performing data enhancement on the source domain IoT device modulated signal data. Suppose there are two samples (x i ,y i ) and (x j ,y j), where x is the input data and y is the corresponding label, the method is as follows:

[0018]

[0019] Among them, λ~β(α,α), α is a hyperparameter that controls the randomness of the interpolation weight;

[0020] Step 2-3: The objective function z obtained from the source domain IoT device modulation signal dataset is s Considered as the L-th layer error of the initial source domain deep feature extraction model and feature recognition model, the L-th layer error is calculated for the L-th layer parameter ω s,L The derivative of Implement the Lth layer parameter ω s,L Updates, there are

[0021]

[0022] Where μ is the update step size;

[0023] Step 2-4, calculate the L-th layer error for the L-th layer input data x s,L-1 The derivative of It is regarded as the error transmitted from the Lth layer to the L-1th layer, and back-propagated until it is updated to the 1st layer, thereby realizing the parameter update of the deep feature extraction model and the recognition model, and obtaining the source domain deep feature extraction model and feature recognition model.

[0024] Furthermore: the target domain deep feature extraction model in step three is the source domain deep feature extraction model trained in step two.

[0025] Further: Step 3 specifically includes the following steps:

[0026] Step 3-1: The complex baseband signal x of the IoT device to be identified o Input the target domain deep feature extraction model to obtain its deep feature x o,L-2 ;

[0027] Further: the specific steps of step 4 include:

[0028] Step 4-1, change x o,L-2 Input the target domain feature recognition model to obtain its predicted probability value x o,L , the predicted label of the IoT device to be identified is obtained by the following formula:

[0029] argmax i x o,L i=1,2,...,C.

[0030] The beneficial effects achieved by the present invention are as follows: In order to solve the problem that IoT devices such as WiFi and LoRa cannot be accurately identified due to the large distribution difference between the modulation signal of the target domain IoT device and the training data, the present invention creatively introduces domain generalization, effectively realizes the recognition of the target domain by the knowledge between the source domain models, and can accurately identify the category of IoT devices in the target domain scenario with large data distribution difference, with high accuracy, stability and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is the identification flow chart of the present invention;

[0032] Figure 2 It is a model block diagram of the present invention;

[0033] Figure 3 It is a feature extraction schematic diagram of the present invention. DETAILED DESCRIPTION

[0034] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0035] like Figure 1 As shown, a cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement specifically includes the following steps:

[0036] Step 1: Use deep learning related knowledge to build the time domain and frequency domain feature extraction model and feature recognition model of the source domain, such as Figure 2 As shown in the figure, the source domain feature extraction model as a whole consists of a group of InceptionBlock modules, a global average pooling layer and a feature fusion layer. The feature recognition model of the source domain in step 1 includes a layer of random inactivation operation and a layer of full connection operation connected in sequence. The parameters in the first layer operation of the corresponding initialization source domain feature extraction model and feature recognition model are denoted as ω s,l , where l = 1, 2, ..., L, L represents the total number of layers of the initialized source domain deep feature extraction model and feature recognition model, and the source domain IoT device modulation signal datasets such as WiFi and LoRa are used to jointly train the two to obtain the source domain feature extraction model and feature recognition model;

[0037] Step 2: Use multi-source domain IoT device modulation signal datasets to jointly train time domain and frequency domain feature extraction models, and introduce data enhancement technology to obtain source domain deep feature extraction models and feature recognition models;

[0038] Step 3: directly apply the source domain deep feature extraction model to the target domain, so that the source domain model can directly extract features from the target domain data and obtain the features of the IoT devices in the target domain.

[0039] Step 4: Input the modulated signal of the IoT device to be identified into the target domain feature recognition model to obtain the recognition result of the RF device to be identified.

[0040] In step 2, the deep learning related knowledge is used to construct the initial source domain deep feature extraction model and feature recognition model, and the source domain IoT device modulation signal data set such as WiFi and LoRa is used to train the two to obtain the source domain deep feature extraction model and feature recognition model. The specific steps include:

[0041] Step 2-1: denote the complex baseband signal of source domain IoT devices such as WiFi and LoRa as x s , the category label of the source domain IoT device is recorded as y s , x s Input the deep feature extraction model and obtain its deep feature as x s,L-2 , x s,L-2 Input the feature recognition model and get its predicted probability value x s,L , using the minimization of the cross entropy loss function as the objective function of the model, we have

[0042]

[0043] Where C is the number of IoT device categories in the source domain, and the result of the above formula obtained by batch processing is recorded as z s ;

[0044] Step 2-2: Generate new samples by performing data enhancement on the source domain IoT device modulated signal data. Suppose there are two samples (x i ,y i ) and (x j ,y j ), where x is the input data and y is the corresponding label, the method is as follows:

[0045]

[0046] Among them, λ~β(α,α), α is a hyperparameter that controls the randomness of the interpolation weight.

[0047] Step 2-3, using the obtained z s Considered as the Lth layer error, the derivative of the Lth layer error with respect to the Lth layer parameter is calculated To update the parameters of layer L, we have

[0048]

[0049] Where μ is the update step size, and the derivative of the Lth layer error to the Lth layer input data is calculated It is regarded as the error transmitted from the Lth layer to the L-1th layer, and back-propagated until it is updated to the 1st layer, thereby realizing the parameter update of the deep feature extraction model and the recognition model, and obtaining the source domain deep feature extraction model and feature recognition model.

[0050] like Figure 3 As shown, assuming that the complex baseband signal x s The dimension of is (2, N). After being processed by the source domain feature extraction model and the feature recognition model, the complex baseband signal is compressed into a vector of dimension (1, C), which can be expressed as

[0051] In the present invention, the category labels of IoT devices are encoded in the form of one-hot encoding. Assuming that the number of categories of IoT devices in the source domain is 10, the category labels of these 10 IoT devices are defined as 1-10 in sequence. If the category label of an IoT device is 2, it is compiled into 010000000 through one-hot encoding, that is, y s .

[0052] for y s The i-th code element in .

[0053] In step three, if Figure 2 As shown, the source domain deep feature extraction model is used as the target domain deep feature extraction model. The steps are as follows:

[0054] Step 3-1: The complex baseband signal x of the IoT device to be identified o Input the target domain deep feature extraction model to obtain its deep feature x o,L-2 ;

[0055] In step 4, the modulation signal feature of the IoT device to be identified is input into the feature recognition model to obtain the identification result of the IoT device to be identified. The specific steps include:

[0056] Step 4-1, change x o,L-2 Input the target domain feature recognition model to obtain its predicted probability value x o,L , the predicted label of the IoT device to be identified is obtained by the following formula:

[0057] argmax i x o,L i=1,2,...,C.

[0058] The beneficial effects achieved by the present invention are as follows: In order to solve the problem that IoT devices such as WiFi and LoRa cannot be accurately identified due to the large distribution difference between the modulation signal of the target domain IoT device and the training data, the present invention creatively introduces domain generalization, effectively realizes the recognition of the target domain by the knowledge between the source domain models, and can accurately identify the category of IoT devices in the target domain scenario with large data distribution difference, with high accuracy, stability and robustness.

[0059] Based on the same technical solution, the present invention also discloses a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by a computing device, the computing device executes the above-mentioned cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement.

[0060] Based on the same technical solution, the present invention also discloses a computing device, including one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the above-mentioned cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement.

[0061] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0063] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0064] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0065] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement, characterized in that: The specific steps include: Step 1: construct a time domain and frequency domain feature extraction model and feature recognition model in the source domain; Step 2: Use multi-source domain IoT device modulation signal datasets to jointly train time domain and frequency domain feature extraction models, and introduce data enhancement technology to obtain source domain deep feature extraction models and feature recognition models; Step 3: directly apply the source domain deep feature extraction model to the target domain to obtain the target domain deep feature extraction model without knowing the target domain samples in advance; Step 4: Input the modulated signal of the IoT device to be identified into the target domain feature recognition model to obtain the recognition result of the RF device to be identified.

2. According to claim 1, a cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement is characterized in that: The time domain and frequency domain feature extraction model of the source domain in the step 1 is composed of a group of InceptionBlock modules, a global average pooling layer and a feature fusion layer. The time domain and frequency domain feature recognition model of the source domain includes 1 layer of random deactivation operation and 1 layer of full connection operation connected in sequence.

3. According to claim 1, a cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement is characterized in that: The source domain IoT device modulation signal dataset in step 2 is composed of several groups of complex baseband signals x s and its corresponding category label y in one-hot encoding form s The joint training specifically includes the following steps: Step 2-1: transform the complex baseband signal x of the source domain IoT device s Input initializes the source domain deep feature extraction model and obtains x s The depth feature is denoted as x s，L-2 , x s，L-2 Input initializes the source domain feature recognition model and obtains x s The predicted probability value of The minimization of the cross entropy loss function is used as the objective function for initializing the source domain deep feature extraction model and feature recognition model. Where C is the number of IoT device categories in the source domain, for y s The i-th code element in For x s The predicted probability value for the i-th category; Step 2-2: Generate new samples by performing data enhancement on the source domain IoT device modulated signal data. Suppose there are two samples (x i ,y i ) and (x j ,y j ), where x is the input data and y is the corresponding label, the method is as follows: Among them, λ~β(α,α), α is a hyperparameter that controls the randomness of the interpolation weight; Step 2-3: The objective function value z obtained from the source domain IoT device modulation signal dataset is s Considered as the L-th layer error of the initial source domain deep feature extraction model and feature recognition model, the L-th layer error is calculated for the L-th layer parameter ω s，L The derivative of Implement the Lth layer parameter ω s，L Updates, there are Where μ is the update step size, and L is the total number of layers of the initialization source domain deep feature extraction model and feature recognition model; Step 2-4, calculate the L-th layer error for the L-th layer input data x s，L-1 The derivative of It is regarded as the error transmitted from the Lth layer to the L-1th layer, and back-propagated until it is updated to the 1st layer, thereby initializing the parameter update of the source domain deep feature extraction model and the feature recognition model, and obtaining the source domain deep feature extraction model and the feature recognition model.

4. According to claim 1, a cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement is characterized in that: The target domain deep feature extraction model in step three is the source domain deep feature extraction model trained in step two.

5. According to claim 1, a cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement is characterized in that: The step three specifically includes the following steps: Step 3-1: The complex baseband signal x of the IoT device to be identified o Input the target domain deep feature extraction model to obtain its deep feature x o，L-2 .

6. According to claim 1, a cross-modulation robust radio frequency fingerprint recognition method based on time-frequency dual-branch hybrid enhancement is characterized in that: The step 4 specifically includes the following steps: Step 4-1, change x o，L2 Input the target domain feature recognition model to obtain its predicted probability value x o，L , the predicted label of the IoT device to be identified is obtained by the following formula: arg max i x o，L i=1,2,...,C。 7. A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, characterized in that: When the instructions are executed by a computing device, the computing device is caused to perform the method according to any one of claims 1 to 6.

8. An electronic device, characterized in that: The method comprises one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method as claimed in any one of claims 1 to 6.

Images