Radio frequency fingerprint model updating method and device, and processing equipment

Abstract

The invention discloses a radio frequency fingerprint model updating method and device, processing equipment and a computer readable storage medium, and the method comprises the steps: building a reference model corresponding to a first task, the reference model is used for radio frequency fingerprint recognition, and the first task is used for recognizing a radio frequency fingerprint of initial equipment; first information and second information are obtained, the first information is information which corresponds to a first task and is used for updating the reference model, the second information is information which corresponds to a second task and is used for updating the reference model, and the second task is old equipment data updating or new equipment introduction; and updating the reference model according to the first information and the second information to obtain a first model.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of communication security, and specifically to a method, apparatus, processing device, and computer-readable storage medium for updating a radio frequency fingerprint model. Background Art

[0002] RF fingerprinting is a technology for identifying wireless devices by extracting unique structures in the electromagnetic waves emitted by the transmitter. RF fingerprinting uses the hardware differences in the physical layer as fingerprint features between different devices to achieve device identification. RF fingerprinting has been used as an additional security layer for wireless devices, and its uniqueness can be used to identify wireless devices to avoid spoofing or simulation attacks. RF fingerprinting can be achieved through the RF fingerprinting model. However, over time, due to long-term use and accidental collisions, the characteristics of the original device components may change, causing the RF fingerprint characteristics of the old device to change, or new devices are added. At this time, the RF fingerprinting model may no longer be applicable, resulting in poor RF fingerprinting results. Summary of the invention

[0003] To solve the above technical problems, the embodiments of the present application provide a method, an apparatus, a processing device and a computer-readable storage medium for updating a radio frequency fingerprint model.

[0004] The present invention provides a method for updating a radio frequency fingerprint model, the method comprising:

[0005] Establishing a reference model corresponding to a first task, where the reference model is used for radio frequency fingerprint recognition, wherein the first task is to recognize the radio frequency fingerprint of an initial device;

[0006] Obtaining first information and second information, wherein the first information is information corresponding to the first task and used to update the above-mentioned benchmark model, and the second information is information corresponding to the second task and used to update the above-mentioned benchmark model, and the second task is to update the data of the old device or introduce the new device;

[0007] The benchmark model is updated according to the first information and the second information to obtain a first model.

[0008] The present application embodiment provides a device for updating a radio frequency fingerprint model, the device comprising:

[0009] A first processing unit is used to establish a reference model corresponding to a first task, where the reference model is used for radio frequency fingerprint recognition, and the first task is to recognize the radio frequency fingerprint of an initial device;

[0010] A first acquisition unit is used to obtain first information and second information, wherein the first information is information corresponding to the first task and used to update the above-mentioned benchmark model, and the second information is information corresponding to the second task and used to update the above-mentioned benchmark model, and the second task is to update the data of the old device or introduce the new device;

[0011] The first updating unit is used to update the above-mentioned reference model according to the above-mentioned first information and the second information to obtain a first model.

[0012] The processing device provided in the embodiment of the present application includes: a processor and a memory, the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the above-mentioned method for updating the radio frequency fingerprint model.

[0013] The computer-readable storage medium provided in the embodiment of the present application is used to store a computer program, which enables a computer to execute the above-mentioned method for updating the radio frequency fingerprint model.

[0014] The above technical solution of the embodiment of the present application establishes a reference model corresponding to the first task, and the reference model is used for radio frequency fingerprint identification. The first task is to identify the radio frequency fingerprint of the initial device; obtains first information and second information, wherein the first information is information corresponding to the first task for updating the above reference model, and the second information is information corresponding to the second task for updating the above reference model, and the second task is to update the data of the old device or introduce a new device; updates the reference model according to the first information and the second information to obtain the first model. In this way, by updating the radio frequency fingerprint model through the first information of the first task and the second information of the second task, as much model information corresponding to the first task as possible can be retained, and at the same time, in the face of the emergence of new tasks, different new samples can be quickly learned to achieve effective updating of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0016] Figure 1 The following is a flow chart of the method for updating the radio frequency fingerprint model provided in the embodiment of the present application. Figure 1 ;

[0017] Figure 2 The following is a flow chart of the method for updating the radio frequency fingerprint model provided in the embodiment of the present application. Figure 2 ;

[0018] Figure 3 It is a schematic diagram of the structure of the device for updating the radio frequency fingerprint model provided in an embodiment of the present application;

[0019] Figure 4 It is a schematic structural diagram of a processing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to enable a more detailed understanding of the features and technical contents of the embodiments of the present application, the implementation of the embodiments of the present application is described in detail below in conjunction with the accompanying drawings. The attached drawings are for reference only and are not used to limit the embodiments of the present application.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0022] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0023] It should also be pointed out that the terms "first\second\third" involved in the embodiments of the present application are only used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0024] RF fingerprinting is a technology for identifying wireless devices by extracting the unique structure in the electromagnetic waves emitted by the transmitter. RF fingerprinting uses the hardware differences of this physical layer as fingerprint features between different devices to achieve device identification. RF fingerprinting has been used as an additional security layer for wireless devices, and its uniqueness can be used to identify wireless devices to avoid spoofing or simulation attacks.

[0025] In recent years, with the development of artificial intelligence, intelligent recognition schemes that do not rely on fixed features and directly use time domain data for recognition have received increasing attention. Some studies have divided the collected time domain data into small segments and input the I / Q signals as samples into deep neural networks, achieving good results.

[0026] However, in actual applications, due to the time-varying characteristics of wireless signals emitted by different devices, existing RF fingerprint recognition models often face some challenges. The first is that they cannot cope with the problem of obsolete knowledge caused by aging equipment. Existing models have a certain natural adaptability to the signals of specific devices. However, over time, the wireless signals emitted by specific devices may change. This signal change may cause the model to be unable to accurately identify, thereby affecting the effect of RF fingerprint recognition. The second is the problem of not considering the addition of new devices: when a new device is added to the RF fingerprint recognition model, the model needs to be retrained to identify the signal characteristics of the new device. Traditional methods mainly include transfer learning and retraining a new model. Transfer learning is to input the new collected data into the existing model for training, while retraining a new model is to shuffle all the data and then retrain the new model.

[0027] Existing RF fingerprint models may not be able to identify new devices, or due to the aging of old devices, the RF fingerprint features change, and the recognition accuracy of the model may decrease, resulting in a significant reduction in the actual application effect. Therefore, in response to this situation, the RF fingerprint recognition solution needs to establish a model update mechanism that can quickly adapt to the signals of new and old devices. Existing methods for updating RF fingerprint models include techniques such as retraining and transfer learning. Retraining a new model often leads to the problem of long training time when the amount of data is too large, while transfer learning will cause the problem of forgetting old knowledge, thereby reducing the recognition efficiency and accuracy of the device. Traditional attempts to solve the continuous learning problem of deep neural networks rely on careful selection of network hyperparameters and other standard regularization methods to mitigate catastrophic forgetting, such as traditional dropout regularization. However, using only stochastic gradient descent with dropout regularization has limitations and cannot be extended to more tasks. Studies have shown that they have only achieved satisfactory results on two tasks.

[0028] In order to efficiently update the radio frequency fingerprint recognition model, the following technical solution of the embodiment of the present application is proposed.

[0029] It should be noted that, unless otherwise specified, the “model” described in the embodiments of the present application may be referred to as a “radio frequency fingerprint model”.

[0030] Figure 1 The following is a flow chart of the method for updating the radio frequency fingerprint model provided in the embodiment of the present application. Figure 1 ;like Figure 1 As shown, the method comprises the following steps:

[0031] Step 101: Establish a reference model corresponding to a first task, where the reference model is used for radio frequency fingerprint identification. The first task is to identify the radio frequency fingerprint of an initial device.

[0032] Here, the first task is the initial task, which is to identify the RF fingerprint of the initial device. That is to say, the above-mentioned benchmark model is a model for identifying the RF fingerprint of the initial device in the initial task. In a specific implementation, for the convenience of description, task A is referred to as the initial task. Task A is to use the RF fingerprint feature to realize the identification of device 1, device 2 and device 3. Then the benchmark model can identify device 1, device 2 and device 3 through the RF fingerprint features of device 1, device 2 and device 3. The number of identification devices is set according to the actual situation, and this application does not make specific restrictions on this.

[0033] Establish a benchmark model corresponding to the first task, including:

[0034] Obtaining first training data for the first task;

[0035] The initial model is trained according to the first training data to obtain a benchmark model.

[0036] Among them, the first task is an initial task, the above-mentioned first training data is used to train an initial model, the initial model is trained according to the first training data to obtain a benchmark model, and the benchmark model is used for radio frequency fingerprint recognition.

[0037] In a specific implementation, task A is to use radio frequency fingerprint features to realize the classification of device 1, device 2 and device 3, collect the original I / Q signals of device 1, device 2 and device 3, that is, the first training data, and record the collection time of the original I / Q signals of each device. The collected original I / Q signals are imported into the radio frequency fingerprint library of the device. The radio frequency fingerprint library can be the radio frequency fingerprint library corresponding to each device, or it can be the radio frequency fingerprint library common to all devices. This application does not make specific restrictions on this. Determine whether the collected I / Q signals need to be preprocessed according to actual conditions. This application does not make specific restrictions on this. In a specific implementation, the preprocessing process includes: time-frequency conversion, signal zero padding, setting a sliding window, and other preprocessing methods can also be selected according to actual conditions. This application does not make specific restrictions on this. In a specific implementation, a convolutional neural network model (Convolutional Neural Networks, CNN) is selected as the initial model, and other initial models can also be selected according to actual conditions. This application does not make specific restrictions on this. Since task A is to use radio frequency fingerprint features to realize the identification of device 1, device 2 and device 3, let the training set be where x i represents the sample, y iis its corresponding label (1, 2, 3), where when there is a preprocessing process, the data in the training set is the I / Q signal after preprocessing, if there is no preprocessing process, the data in the training set is the first training data, and this application does not make specific restrictions on this. Since task A is a classification problem, the cross entropy function L is selected here A (w A ) is used as the loss function of the initial model. Of course, the corresponding loss function can also be selected according to the actual situation. This application does not make specific restrictions on this. The expression of the loss function is as follows:

[0038]

[0039] Among them, w A are the parameters of the model, x i Represents the training set D A The i-th data in, N is the training set D A The amount of data in, k is the number of categories to be recognized by the initial model, which is set according to the actual situation. Here, task A is to identify device 1, device 2, and device 3, so the maximum value of k is 3, f k (x i ;w A ) indicates that the initial model is for the input sample x i The probability of predicting the kth category, [y i = = k] means when y i =k is 1, otherwise it is 0.

[0040] According to the training set D A The initial model is trained with the loss function, and finally the RF fingerprint recognition benchmark model based on task A is obtained.

[0041] Step 102: Obtain first information and second information, the first information is information corresponding to the first task for updating the above-mentioned benchmark model, and the second information is information corresponding to the second task for updating the above-mentioned benchmark model, and the second task is to update the old device data or introduce new equipment.

[0042] The first information includes parameters of the benchmark model, first training data, the amount of data corresponding to the first training data, and the Fisher information matrix of the benchmark model.

[0043] Among them, if the collected I / Q signal does not need to be preprocessed, the first training data is the original I / Q signal collected, and the data volume corresponding to the first training data refers to the data volume of the first training data; if the collected I / Q signal needs to be preprocessed, the first training data is the original I / Q signal collected, and the data volume corresponding to the first training data is the data volume of the first training data after preprocessing.

[0044] When the second task is obtained, in order to effectively update the baseline model, the Elastic Weight Consolidation (EWC) algorithm is introduced. This algorithm is a method that can implement incremental learning in neural networks, which enables the neural network to not forget the previously learned knowledge when processing new tasks. The EWC algorithm is based on the Fisher information matrix, which can be calculated by the gradient of the loss function of the old task to the model parameters, and is used to measure the influence of the parameters on the old task. The EWC algorithm associates the Fisher information matrix of the previously learned task with the loss function of the current task. The objective function of training is not only to minimize the loss of the current task, but also to retain the knowledge of the previous task as much as possible, which effectively reduces the conflict between the new task and the old task.

[0045] The Fisher information matrix is ​​defined as follows:

[0046]

[0047] Where N is the number of training samples, j is the jth sample of the training sample, and w i is the parameter of the benchmark model corresponding to the old task, L j is the loss function of the model, is the gradient of the jth sample to the ith parameter. This matrix reflects the influence of the benchmark model on the ith parameter on the old task. In a specific implementation, if the next task needs to update the benchmark model, then in the Fisher information matrix of task A, N is the amount of data corresponding to the first training data, j is the training set D A The jth sample in w i is the parameter of the benchmark model corresponding to task A, L j is the loss function of the model, is the training set D A The matrix is ​​the gradient of the j-th sample in the model to the i-th parameter. This matrix reflects the influence of the i-th parameter of the baseline model on task A.

[0048] The second information includes: a first parameter, a second parameter, a first objective function, and a second objective function; the method further includes:

[0049] If the second task is to update the old device data, then determine the first objective function; and obtain the collection time of the first training data and the collection time of the second training data, and determine the first parameter according to the collection time of the first training data and the collection time of the second training data; the second training data is the training data required for updating the old device data;

[0050] If the second task is to introduce a new device, a second objective function is determined; and the usage frequency of the new device and the importance information of the new device are obtained, and a second parameter is determined according to the usage frequency of the new device and the importance information of the new device.

[0051] As time goes by, the RF fingerprint recognition model may complete different updates according to different needs. This application believes that the update needs mainly come from the following two situations: the need to identify new devices, that is, the introduction of new devices; because the model has been used for a long time, the fingerprint characteristics of the old devices have changed, and the old device data needs to be updated, that is, the old device data is updated.

[0052] In a specific implementation, during the use of the device, due to long-term use and accidental collision, the characteristics of the device components may change, thereby causing the RF fingerprint characteristics of the old device to change. At this time, it is necessary to re-collect the I / Q signal of the device to update the RF fingerprint library.

[0053] If the second task is to update the data of the old device, then determine the first objective function; and, obtain the acquisition time of the aforementioned first training data and the acquisition time of the second training data, and determine the first parameter according to the acquisition time of the aforementioned first training data and the acquisition time of the aforementioned second training data; the second training data is the training data required for updating the data of the old device. The second training data is the I / Q signal of the device to be updated that is collected, and the second training data is used to update the reference model. Which old device needs to be updated is determined according to the actual situation, and this application does not make specific restrictions on this. The first parameter is the regularization parameter required for updating the data of the old device. In a specific implementation, the second task is set as task B, and task B is to update the data of device 1 and re-collect the I / Q signal of device 1, that is, the second training data. In order to update the reference model later, it is necessary to record the acquisition time of the second training data, that is, the time of re-collecting the I / Q signal of device 1. Determine whether the second training data needs to be pre-processed according to the actual situation, and this application does not make specific restrictions on this. In a specific implementation, the pre-processing process is consistent with the aforementioned pre-processing process, which will not be repeated here, and other pre-processing methods can also be selected according to the actual situation, and this application does not make specific restrictions on this. The regularization parameter is determined according to the acquisition time of the first training data and the acquisition time of the second training data. The collected I / Q signal is imported into the RF fingerprint library, the previously collected I / Q signal of device 1 is deleted, and the data in the RF fingerprint library of the device is updated. The RF fingerprint library can be the RF fingerprint library corresponding to device 1, or it can be shared with other devices. This application does not make specific restrictions on this.

[0054] Here, the purpose of task B is still to identify device 1, device 2 and device 3. Let the training set be where x irepresents the sample, y i is its corresponding label (1, 2, 3), where, when there is a preprocessing process, the data in the training set is the I / Q signal after preprocessing, and if there is no preprocessing process, the data in the training set is the second training data, which is not specifically limited in this application. The loss function of task B is consistent with the loss function of task A, that is:

[0055] L B (w B )=L A (w A );

[0056] Specifically, the loss function L of task B is B (w B ) is expressed as follows:

[0057]

[0058] Among them, w B is the parameter of the model corresponding to task B, x i Represents the training set D B The i-th data in, N is the training set D B The amount of data in, k is the number of categories to be recognized by the model, and is set according to the actual situation. Here, task B is to identify device 1, device 2, and device 3. Therefore, the maximum value of k is 3, and f k (x i ;w B ) indicates that the initial model is for the input sample x i The probability of predicting the kth category, [y i = = k] means when y i =k is 1, otherwise it is 0.

[0059] Therefore, the objective function of task B can be expressed as:

[0060] L B (w B )+λ·Ω(w A );

[0061] Among them, L B (w B ) is the loss function of task B, w B is the parameter of the model corresponding to task B, λ is the regularization coefficient, Ω(w A ) is the regularization term, and its expression is:

[0062]

[0063] Among them, w B,i represents the parameters of the model corresponding to task B, w A,iis the parameter of the model corresponding to task A, F i (w A ) reflects the influence of the i-th parameter of the model corresponding to task A on task A, that is, the influence of the i-th parameter of the baseline model on task A.

[0064] The regularization coefficient λ controls the influence of retaining the previous task in the update, that is, when the second task is to update the old device data, the regularization coefficient λ controls the influence of retaining task A in the update. In a specific implementation, the setting of λ is related to the time interval for updating the old device data. As mentioned above, the RF fingerprint library not only contains the original I / Q signals of different devices, but also contains the time of collecting the signals. Obviously, the longer the time interval, the greater the possibility of the device fingerprint feature changing. At this time, the value of λ should be lower. For example, the data of device 1 is updated at time t, and the collection time of the data of device 1 is t. 1 , then λ(tt 1 ) should be a random (tt 1 ) is a monotonically decreasing function, and λ is called the regularization parameter of time correlation. It should be noted that λ is not only related to the update time interval. For example, when the data update of the old device is only a daily update, the value of λ is relatively large, because the old model can still complete the recognition of the old device. However, the recognition rate of the old device is very low. When data update is required, the old model can no longer complete the recognition of the old device. At this time, the value of λ needs to be relatively small.

[0065] In a specific implementation, the setting process of λ is described by taking the time interval as an example. Assume that λ is the regularization parameter of time correlation, t is the current time, and t 1 is the collection time of the old device data, then λ can be calculated using the following formula:

[0066]

[0067] In this calculation formula, as the time interval increases (ie (tt 1 ), the value of λ decreases continuously, reflecting the fact that the influence of old device data on the update gradually weakens. When the time interval approaches infinity, the value of λ approaches 0, indicating that the influence of old model parameters on new task learning is negligible. Among them, c is a constant used to avoid the situation where the denominator is 0, and it can also control the range of λ. Because the setting of λ is related to the time interval, it may also be related to the following factors:

[0068] (1) Size and complexity of the dataset: The larger and more complex the dataset is, the larger the regularization parameter is required to balance the complexity of the model and the noise level of the dataset.

[0069] (2) Complexity of network structure: The number of layers and nodes in a neural network will affect the complexity of the model. When the network structure is more complex, the generalization ability of the model may become worse, and a larger regularization parameter is required.

[0070] (3) Correlation between the previous task and the new task: If the features between the previous task and the new task are similar, a smaller regularization parameter is required to retain the information of the previous task; if the features between the tasks are very different, a larger regularization parameter is required to avoid the information of the previous task interfering with the learning of the new task.

[0071] (4) Choice of loss function: The loss function of the model may also affect the effectiveness of regularization. If the loss function can better measure the importance of the new task, then fewer regularization parameters are needed to balance the old and new tasks; if the loss function cannot directly measure the importance of the new task, then a larger regularization parameter is needed.

[0072] The above factors are not closely related to the RF fingerprint scenario, but only to the data set / model / loss function. Therefore, the empirical value of λ after comprehensively considering the above four factors is recorded as λ 0 .

[0073] In a specific implementation, in the radio frequency fingerprint recognition solution, the system implements secondary authentication by identifying different devices. Due to the continuous introduction of communication devices, such as replacement of mobile phones, the system needs to continuously update the radio frequency fingerprint recognition model in order to realize the recognition of new devices.

[0074] If the second task is the introduction of a new device, the second objective function is determined; and the frequency of use of the new device and the importance information of the new device are obtained, and the second parameter is determined according to the frequency of use of the new device and the importance information of the new device. If the second task is the introduction of a new device, it is necessary to collect the I / Q signal of the new device, that is, the third training data, and record the collection time of the third training data. The third training data is used to update the benchmark model when the new device is introduced. The second parameter is the regularization parameter required to update the model when the new device is introduced. In a specific embodiment, a new device 4 is added, that is, the benchmark model is required to identify the new device 4. For the convenience of description, the second task is set to task C. Task C is to enable the benchmark model to identify device 1, device 2, device 3 and device 4. Therefore, it is necessary to collect the I / Q signal emitted by device 4. The third training data is the collected I / Q signal of device 4. For the subsequent model update, it is necessary to record the collection time of the third training data, that is, the time of collecting the I / Q signal of device 4. It is determined whether the third training data needs to be preprocessed according to the actual situation, and this application does not make specific restrictions on this. In a specific embodiment, the preprocessing process is consistent with the aforementioned preprocessing process, which will not be described in detail here. Other preprocessing methods may also be selected according to actual conditions, and this application does not make specific limitations on this. The regularization parameter is determined according to the frequency of use of the new device and the importance information of the new device. Since task C is the introduction of a new device 4, the regularization parameter is determined according to the frequency of use of the new device 4 and the importance information of the new device 4. The I / Q signal of device 4 is imported into the RF fingerprint library, which may be the RF fingerprint library corresponding to device 4, or may share a RF fingerprint library with other devices, and this application does not make specific limitations on this.

[0075] Here, the purpose of task C is to identify device 1, device 2, device 3, and device 4. Let the training set be where x i represents the sample, y i is its corresponding label (1, 2, 3, 4), where, when there is a preprocessing process, the data in the training set is the I / Q signal after preprocessing, and if there is no preprocessing process, the data in the training set is the third training data, which is not specifically limited in this application. The loss function of task C can also be designed as a cross entropy function, as follows:

[0076]

[0077] Among them, w C is the parameter of the model corresponding to task C, x i Represents the training set D C The i-th data in, N is the training set D CThe amount of data in, k is the number of categories to be recognized by the model, which is set according to the actual situation. Here, task C is to identify device 1, device 2, device 3 and device 4, so the maximum value of k is 4, f k (x i ;w C ) indicates that the initial model is for the input sample x i The probability of predicting the kth category, [y i = = k] means when y i =k is 1, otherwise it is 0.

[0078] Therefore, the objective function of task C can be expressed as:

[0079] L C (w C )+λ·Ω(w A );

[0080] Among them, L C (w C ) is the loss function of task C, w C is the parameter of the model corresponding to task C, λ is the regularization coefficient, Ω(w A ) is the regularization term, and its expression is:

[0081]

[0082] Among them, w C,i represents the parameters of the model corresponding to task C, w A,i is the parameter of the model corresponding to task A, F i (w A ) reflects the influence of the i-th parameter of the model corresponding to task A on task A, that is, the influence of the i-th parameter of the baseline model on task A.

[0083] Among them, the regularization coefficient λ is related to the importance and usage frequency of the new device 4. Since the setting of the λ value is very specific, different application scenarios will have different requirements, and this application does not make specific restrictions on this. In a specific implementation, the importance information F and usage frequency U of the new device 4 are used to calculate the initial λ value:

[0084]

[0085] Wherein, k is a constant that can be set according to empirical rules and experimental results, and this application does not make any specific restrictions on this. 0 The settings are the same as above and will not be repeated here.

[0086] This formula shows that when the importance or frequency of use of the device (task) is high, the λ value should be set smaller to balance the model's adaptation to new tasks and retention of old tasks; when the importance or frequency of use of the device is low, the λ value should be larger to retain the influence of previously learned tasks and make the new model more adaptable to old devices. In practical applications, the specific methods for calculating the importance information F and the frequency of use U vary from case to case. For the importance information of new devices, you can consider using expert opinions, survey results, or determine it based on the functions and applications of the device. For the frequency of use, you can collect information from device log files or user feedback, or calculate it based on device production and sales data. You can also use historical data from model training to infer the relative importance and frequency of new tasks.

[0087] Importance information and usage frequency are related to specific application scenarios and need to be adjusted according to actual conditions. In a specific implementation, if the task is to update a model for a medical device used in a hospital, the importance information of the device may be defined as the degree of impact of the device on life and health, and the frequency of use may be based on the usage rate of the device in the hospital. If the task is to update the model for a device in a smart home environment, the importance information and usage frequency of the device may be based on the function of the device and the usage habits of family members, and this application does not make specific limitations on this.

[0088] Step 103: Update the reference model according to the first information and the second information to obtain a first model.

[0089] The first information is information corresponding to the first task and used to update the above-mentioned benchmark model, wherein the second information is information corresponding to the second task and used to update the above-mentioned benchmark model. After updating the model, the first model is obtained, that is, the model corresponding to the second task.

[0090] If the second task is to update the old device data, then the benchmark model is updated according to the first parameter, the first objective function, the second training data and the Fisher information matrix to obtain the first model;

[0091] If the second task is the introduction of a new device, the benchmark model is updated according to the second parameter, the second objective function, the third training data and the Fisher information matrix to obtain the first model, and the third training data is the training data required for the introduction of the new device.

[0092] In a specific implementation, if the second task is to update the old device data, that is, the aforementioned task B, then according to the regularization parameter λ related to the time interval, the corresponding loss function L when the old device data is updated is B (w B), and the Fisher information matrix associated with task A determines the objective function of task B, namely L B (w B )+λ·Ω(w A ), through the training set D B The reference model is updated to obtain a first model, so that the first model can re-identify device 1, device 2, and device 3.

[0093] In a specific implementation, if the second task is to introduce a new device, that is, the aforementioned task C, then according to the regularization parameter λ related to the importance information of the new device and the frequency of use of the new device, the corresponding loss function L when the new device is introduced is C (w C ), and the Fisher information matrix associated with task A determines the objective function of task C, namely L C (w C )+λ·Ω(w A ), through the training set D C The reference model is updated to obtain a first model, so that the first model can re-identify device 1, device 2, device 3, and device 4.

[0094] Since the security authentication scheme based on RF fingerprint needs to continuously update the model over time, the above embodiments respectively describe how the baseline model is updated and adjusted when introducing new devices and updating the data of old devices. In actual applications, there will be multiple task updates, and the main process includes: determining the task type, determining the regularization coefficient, determining the new objective function, and completing the RF fingerprint model update. After determining the task type, the RF fingerprint model can be updated according to the above method.

[0095] After the reference model is updated according to the first information and the second information and the first model is obtained, the method includes:

[0096] Obtaining third information, where the third information is information corresponding to the third task and used to update the first model;

[0097] The first model is updated according to the first information, the second information and the third information.

[0098] Among them, the first information is the information corresponding to the first task and used to update the above-mentioned baseline model, and the second information is the information corresponding to the second task and used to update the above-mentioned baseline model; the baseline model is updated according to the first information and the second information to obtain the first model, which is the model corresponding to the second task. When determining the third task, it is necessary to obtain the third information, and the third information is the information corresponding to the third task and used to update the above-mentioned first model; the above-mentioned first model is updated according to the aforementioned first information, the aforementioned second information and the aforementioned third information.

[0099] In a specific implementation, assume that the initial task A is to identify device 1, device 2, and device 3; a period of time t 0 After that, the data of device 1 needs to be updated, and devices 1, 2, and 3 need to be distinguished, which is task B. After a period of time t 1 Later, a new device 4 is added to the system. At this time, the model needs to distinguish between device 1, device 2, device 3 and device 4, corresponding to task C. At this time, the latest task is task C.

[0100] In one embodiment, both Task A and Task B may be considered simultaneously, which allows the objective function to be defined as:

[0101]

[0102] Among them, L C (w C ) is the loss function corresponding to task C, that is, the loss function corresponding to the introduction of new equipment,

[0103] L C (w C )=L B (w B );

[0104]

[0105] L C (w C ) are the same as above and will not be repeated here.

[0106] When W = W A hour,

[0107]

[0108] When W = W B hour,

[0109]

[0110] When W = W A When W is a function that is negatively correlated with the importance information and usage frequency of device 4, and is also a function that is related to t 0 Negatively correlated function, when W = W B When W is a 0 The I / Q signal of the newly introduced device 4 is collected, and the training set is where x i represents the sample, y iis its corresponding label (1, 2, 3, 4), where when there is a preprocessing process, the data in the training set is the I / Q signal after preprocessing. If there is no preprocessing process, the data in the training set is the I / Q signal of the collected device 4. This application does not make specific limitations on this, and the first model is updated through the training set D.

[0111] In a specific implementation, since task A has been considered when training task B, when task C appears, only the training model parameters of task B can be directly considered. At this time, the objective function can be defined as:

[0112] L C (w C )+λ B Ω(W B );

[0113] in

[0114]

[0115]

[0116] λ B is a function that is negatively correlated between the importance information of device 4 and its usage frequency. Collect the I / Q signals of the newly introduced device 4 and let the training set be where x i represents the sample, y i is its corresponding label (1, 2, 3, 4), where when there is a preprocessing process, the data in the training set is the I / Q signal after preprocessing. If there is no preprocessing process, the data in the training set is the I / Q signal of the collected device 4. This application does not make specific limitations on this, and the first model is updated through the training set D.

[0117] The above technical solution of the embodiment of the present application establishes a reference model corresponding to the first task, and the reference model is used for radio frequency fingerprint identification. The first task is to identify the radio frequency fingerprint of the initial device; obtains first information and second information, wherein the first information is the information corresponding to the first task for updating the reference model, and the second information is the information corresponding to the second task for updating the reference model, and the second task is to update the data of the old device or introduce a new device; updates the above reference model according to the first information and the second information to obtain the first model. In this way, in the update of the radio frequency fingerprint model, the radio frequency fingerprint model is updated by the first information of the first task and the second information of the second task, so as to retain as much model information corresponding to the first task as possible, and at the same time, in the face of the emergence of new tasks, quickly learn different new samples to achieve effective update of the model.

[0118] Based on the above embodiments, Figure 2Schematic diagram of the process of updating the radio frequency fingerprint model provided in the embodiment of the present application Figure 2 ,like Figure 2 As shown, the method comprises the following steps:

[0119] Step 201: Establishing a radio frequency fingerprint recognition benchmark model corresponding to the initial task;

[0120] Step 202: Calculate the Fisher information matrix under the benchmark model;

[0121] Step 203: New equipment is introduced;

[0122] If the new task is the introduction of a new device, execute step 204 to step 205; otherwise, execute step 206 to step 208.

[0123] Step 204: Obtaining importance information and usage frequency of the new device;

[0124] The importance information and usage frequency of the new device are related to the usage scenario of the device, and this application does not make any specific limitations on this.

[0125] Step 205: define the objective function and regularization coefficient λ corresponding to the introduction of new equipment;

[0126] When a new device is introduced, the objective function is consistent with the objective function of the aforementioned task C, which will not be repeated here. The regularization coefficient λ and important information about the new device are related to the frequency of use.

[0127] Step 206: old device data update;

[0128] Step 207: Obtaining the time interval between new and old data collection;

[0129] The collection time of the training data of the initial task is obtained, and the collection time of the training data of the equipment that needs to be re-collected and updated is also obtained. Based on the above two collection times, the time interval between the collection of new and old data is obtained.

[0130] Step 208: define the objective function and regularization coefficient λ corresponding to the old device data update;

[0131] When the old device data is updated, the objective function is consistent with the objective function of the aforementioned task B, which will not be repeated here. The regularization coefficient λ and important information of the new device are related to the frequency of use.

[0132] Step 209: Update the baseline model to obtain a new model.

[0133] The initial task is to establish a RF fingerprint recognition benchmark model; calculate the Fisher information matrix under the benchmark model; when a new device is introduced, obtain the importance information and usage frequency of the new device, and define the corresponding objective function and regularization coefficient λ when the new device is introduced; when the old device data is updated, obtain the time interval between the new and old data collection, and define the corresponding objective function and regularization coefficient λ when the old device data is updated; update the benchmark model to obtain a new model. In this way, the present application introduces the EWC algorithm into the update of the RF fingerprint recognition model. By reasonably setting the hyperparameters, as much old model knowledge as possible is retained when the model is updated, and at the same time, multiple new tasks are coped with, so that it can quickly learn different new samples. Secondly, the time-related regularization parameter is introduced to make the model gradually forget the expired historical data, realize the effective update of the RF fingerprint recognition model, and effectively deal with the problems of new device identification and old device fingerprint aging. Use the EWC algorithm for incremental learning to avoid retraining, which takes a lot of time, while allowing the model to retain the knowledge of the old model in the process of learning the classification of new devices. When updating the model and data set, the model is updated through the relevant information of the initial task and the next task, and the regularization parameters of time correlation and regularization parameters related to the frequency and importance of new equipment are introduced to adjust the importance of the regularization term. The weight influence of historical tasks on model parameters is adjusted according to different tasks. In the face of the emergence of new tasks, different new samples are quickly learned to achieve effective updating of the RF fingerprint recognition model, and effectively deal with the problems of fingerprint aging of old equipment and the introduction of new equipment.

[0134] The preferred embodiments of the present application are described in detail above in conjunction with the accompanying drawings. However, the present application is not limited to the specific details in the above embodiments. Within the technical concept of the present application, the technical solution of the present application can be subjected to a variety of simple modifications, and these simple modifications all belong to the protection scope of the present application. For example, the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present application will not further explain various possible combinations. For another example, the various different embodiments of the present application can also be arbitrarily combined, as long as they do not violate the idea of ​​the present application, they should also be regarded as the contents disclosed in the present application. For another example, under the premise of no conflict, the various embodiments and / or the technical features in the various embodiments described in the present application can be arbitrarily combined with the prior art, and the technical solution obtained after the combination should also fall within the protection scope of the present application.

[0135] It should be understood that in the various method embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0136] Based on the same inventive concept as the above embodiments, Figure 3 is a schematic diagram of the structure of the device for updating the radio frequency fingerprint model provided in the embodiment of the present application, such as Figure 3 As shown, the device for updating the radio frequency fingerprint model includes:

[0137] The first processing unit 301 is used to establish a reference model corresponding to a first task, where the reference model is used for radio frequency fingerprint recognition, and the first task is to recognize the radio frequency fingerprint of an initial device;

[0138] A first acquisition unit 302 is used to obtain first information and second information, wherein the first information is information corresponding to the first task and used to update the above-mentioned benchmark model, and the second information is information corresponding to the second task and used to update the above-mentioned benchmark model, and the second task is to update the old device data or introduce the new device;

[0139] The first updating unit 303 is used to update the reference model according to the first information and the second information to obtain a first model.

[0140] In some implementations, the first processing unit 301 is further specifically used to: obtain first training data for the first task; and train the initial model according to the first training data to obtain a benchmark model.

[0141] In some implementations, the first acquisition unit 302 is further specifically used to: acquire the parameters of the aforementioned benchmark model, the aforementioned first training data, the data volume corresponding to the aforementioned first training data, and the Fisher information matrix of the aforementioned benchmark model.

[0142] In some embodiments, the first acquisition unit 302 is also specifically used to: acquire the first parameter, the second parameter, the first objective function and the second objective function; if the above-mentioned second task is to update the old device data, then determine the first objective function; and, acquire the collection time of the above-mentioned first training data and the collection time of the second training data, and determine the first parameter based on the collection time of the above-mentioned first training data and the collection time of the above-mentioned second training data; the second training data is the training data required for the above-mentioned old device data update; if the above-mentioned second task is to introduce a new device, then determine the second objective function; and, acquire the usage frequency of the new device and the importance information of the new device, and determine the second parameter based on the usage frequency of the new device and the importance information of the new device.

[0143] In some implementations, the first acquisition unit 302 is further specifically configured to: record the acquisition time of the aforementioned first training data.

[0144] In some implementations, the first acquisition unit 302 is further specifically configured to: re-acquire the second training data of the old device, and record the acquisition time of the second training data.

[0145] In some implementations, the first acquiring unit 302 is further specifically configured to: acquire third training data of the new device, and record the acquisition time of the third training data.

[0146] In some implementations, the first acquisition unit 302 is further specifically used to: obtain third information, where the third information is information corresponding to the third task and used to update the first model;

[0147] In some embodiments, the first update unit 303 is also specifically used for: if the second task is to update the data of an old device, then the baseline model is updated according to the aforementioned first parameter, the aforementioned first objective function, the aforementioned second training data and the aforementioned Fisher information matrix to obtain the first model; if the second task is to introduce a new device, then the baseline model is updated according to the aforementioned second parameter, the aforementioned second objective function, the third training data and the aforementioned Fisher information matrix to obtain the first model, and the third training data is the training data required for the introduction of the new device.

[0148] In some implementations, the first updating unit 303 is further specifically configured to: update the first model according to the first information, the second information, and the third information.

[0149] Those skilled in the art should understand that Figure 3 The implementation functions of each unit in the device for updating the radio frequency fingerprint model shown can be understood by referring to the relevant description of the aforementioned method. Figure 3 The functions of each unit in the device for updating the radio frequency fingerprint model shown can be implemented by a program running on a processor, or by a specific logic circuit.

[0150] Based on the above embodiments, the present application provides a processing device, which can be applied to Figures 1-2 In the method for updating the radio frequency fingerprint model provided in the corresponding embodiment, Figure 4 It is a schematic structural diagram of a processing device 400 provided in an embodiment of the present application. Figure 4 The processing device 400 shown includes a processor 401, which can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0151] Alternatively, if Figure 4 As shown, the processing device 400 may further include a memory 402. The processor 401 may call and run a computer program from the memory 402 to implement the method in the embodiment of the present application.

[0152] The memory 402 may be a separate device independent of the processor 401 , or may be integrated into the processor 401 .

[0153] Alternatively, if Figure 4 As shown, the processing device 400 may further include a transceiver 403, and the processor 401 may control the transceiver 403 to communicate with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices.

[0154] The transceiver 403 may include a transmitter and a receiver. The transceiver 403 may further include an antenna, and the number of the antennas may be one or more.

[0155] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by the hardware integrated logic circuit or software instructions in the processor. The above processor can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor are combined and performed. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, and other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0156] It can be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0157] It should be understood that the above-mentioned memory is exemplary but not restrictive. For example, the memory in the embodiments of the present application may also be static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM), etc. That is to say, the memory in the embodiments of the present application is intended to include but not limited to these and any other suitable types of memory.

[0158] An embodiment of the present application also provides a computer-readable storage medium for storing a computer program.

[0159] Optionally, the computer-readable storage medium can be applied to the processing device in the embodiments of the present application, and the computer program enables the computer to execute the corresponding processes implemented by the processing device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0160] An embodiment of the present application also provides a computer program product, including computer program instructions.

[0161] Optionally, the computer program product can be applied to the processing device in the embodiments of the present application, and the computer program instructions enable the computer to execute the corresponding processes implemented by the processing device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0162] The embodiment of the present application also provides a computer program.

[0163] Optionally, the computer program can be applied to the processing device in the embodiments of the present application. When the computer program runs on a computer, the computer executes the corresponding processes implemented by the processing device in the various methods of the embodiments of the present application. For the sake of brevity, they are not described here.

[0164] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0165] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0166] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0167] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0168] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0169] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0170] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for updating a radio frequency fingerprint model, It is characterized in that The method comprises: Establishing a reference model corresponding to a first task, wherein the reference model is used for radio frequency fingerprint recognition, wherein the first task is to recognize the radio frequency fingerprint of an initial device; Obtaining first information and second information, wherein the first information is information corresponding to a first task and used to update the reference model, and the second information is information corresponding to a second task and used to update the reference model, wherein the second task is to update data of an old device or to introduce a new device; The reference model is updated according to the first information and the second information to obtain a first model.

2. The method according to claim 1, It is characterized in that The establishing of a benchmark model corresponding to the first task includes: Acquire first training data for the first task; The initial model is trained according to the first training data to obtain the benchmark model.

3. The method according to claim 2, It is characterized in that The first information includes parameters of the benchmark model, the first training data, the amount of data corresponding to the first training data, and the Fisher information matrix of the benchmark model.

4. The method according to claim 3, It is characterized in that The second information includes: a first parameter, a second parameter, a first objective function and a second objective function; the method further includes: If the second task is to update the old device data, then determine the first objective function; and obtain the collection time of the first training data and the collection time of the second training data, and determine the first parameter according to the collection time of the first training data and the collection time of the second training data; the second training data is the training data required for updating the old device data; If the second task is to introduce a new device, then determine the second objective function; and obtain the usage frequency of the new device and the importance information of the new device, and determine the second parameter according to the usage frequency of the new device and the importance information of the new device.

5. The method according to claim 4, It is characterized in that After obtaining the first training data for the first task, the method further includes: The collection time of the first training data is recorded.

6. The method according to claim 4, It is characterized in that If the second task is to update the old device data, the method further includes: The second training data of the old device is collected again, and the collection time of the second training data is recorded.

7. The method according to claim 4, It is characterized in that If the second task is the introduction of a new device, the method further includes: Collect the third training data of the new device, and record the collection time of the third training data.

8. The method according to claim 4, It is characterized in that The updating of the reference model according to the first information and the second information to obtain a first model includes: If the second task is to update the old device data, then updating the benchmark model according to the first parameter, the first objective function, the second training data and the Fisher information matrix to obtain a first model; If the second task is the introduction of a new device, the benchmark model is updated according to the second parameter, the second objective function, the third training data and the Fisher information matrix to obtain the first model, and the third training data is the training data required for the introduction of the new device.

9. The method according to any one of claims 1 to 8, It is characterized in that After the reference model is updated according to the first information and the second information to obtain the first model, the method includes: Obtaining third information, where the third information is information corresponding to a third task and used to update the first model; The first model is updated according to the first information, the second information and the third information.

10. A device for updating a radio frequency fingerprint model, It is characterized in that The device comprises: A first processing unit, configured to establish a reference model corresponding to a first task, wherein the reference model is used for radio frequency fingerprint recognition, and the first task is to recognize the radio frequency fingerprint of an initial device; A first acquisition unit, configured to acquire first information and second information, wherein the first information is information corresponding to a first task and used to update the reference model, and the second information is information corresponding to a second task and used to update the reference model, wherein the second task is to update data of an old device or to introduce a new device; A first updating unit is used to update the reference model according to the first information and the second information to obtain a first model.

11. A processing device, It is characterized in that include: A processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory to execute the method as claimed in any one of claims 1 to 9.

12. A computer-readable storage medium, It is characterized in that Used to store a computer program, wherein the computer program causes a computer to execute the method according to any one of claims 1 to 9.

Images