LoRa device radio frequency fingerprint identification method and device based on conditional diffusion model

Abstract

The application provides a LoRa device radio frequency fingerprint identification method and device based on a conditional diffusion model, and the method comprises the following steps: acquiring radio frequency signals of different LoRa devices; pre-processing the radio frequency signals to obtain a training data set; constructing a conditional diffusion model, and training the conditional diffusion model by using the training data set; in the training process, introducing Rayleigh fading multiplicative interference and additive Gaussian noise to simulate the multipath fading characteristics of real LoRa device communication; generating a radio frequency fading signal by using the trained conditional diffusion model, and adding the generated radio frequency fading signal to the training data set to obtain an enhanced data set; training a pre-constructed radio frequency fingerprint identification model by using the enhanced data set; and identifying the radio frequency fading signal of a LoRa device to be identified by using the trained radio frequency fingerprint identification model, so as to obtain the category corresponding to the LoRa device to be identified; thereby improving the device identification accuracy under a multipath channel.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication security, and in particular to a LoRa device radio frequency fingerprint identification method based on a conditional diffusion model, a LoRa device radio frequency fingerprint identification device based on a conditional diffusion model, a computer readable storage medium and a computer device. BACKGROUND

[0002] In recent years, with the rapid development of Internet of Things technology, Long Range Radio (LoRa) technology has shown great application potential in the fields of smart cities and industrial Internet of Things. LoRa has become an important solution for long-distance communication due to its low power consumption and wide coverage. However, LoRa technology works in the unlicensed frequency band of 868MHz / 915MHz, and its open network architecture makes it vulnerable to malicious attacks such as information tampering and identity impersonation. Moreover, LoRa devices have limited computing resources, and traditional authentication schemes based on cryptography are difficult to deploy directly. Therefore, it is urgent to develop a lightweight access authentication scheme with high security and low computational complexity to ensure that only legally authenticated devices can access network communication.

[0003] Radio Frequency Fingerprint (RFF) is derived from inherent hardware differences in the manufacturing process of electronic components, which are reflected in the subtle characteristics of the transmitted signal, making it difficult to replicate and counterfeit. This provides a new solution for LoRa device authentication. However, in long-distance communication scenarios, signals propagate through line-of-sight or non-line-of-sight paths, experiencing multipath effects such as reflection and scattering, resulting in severe amplitude fluctuations and phase distortion of the received signal. Environmental noise further reduces the signal-to-noise ratio, making it even more difficult to accurately extract the original weak RFF characteristics, resulting in a significant decline in the recognition performance of the RFF identification method in the LoRa communication environment.

[0004] Robust RFF identification methods for LoRa devices generate signals under multipath fading channels based on channel models and add them to the training set of the identification model for training to improve the identification accuracy of the model under different multipath fading channels. For example, [A. Al-Shawabka, P. Pietraski, S. B. Pattar, F. Restuccia, and T. Melodia,“DeepLoRa: Fingerprinting LoRa devices at scale through deep learning and dataaugmentation,” in Proc. Int. Symp. Theory, Algorithmic Foundations, Protocol Design Mobile Netw. Mobile Comput. (MobiHoc), New York, USA, Jul. 2021, pp.251–260.] proposed a data augmentation method based on complex-valued Finite Impulse Response (FIR) filter. This method is based on the multipath channel model of the International Telecommunication Union Radio Communication Sector M.1225 standard. The radio frequency signal under good channel conditions is transformed by changing the filter tap coefficient, and the transformed signal is used as the training data of the identification model to improve the identification accuracy of the model under indoor and outdoor LoRa multipath channel environments. [G. Shen, J. Zhang, A. Marshall, and J. R. Cavallaro, “Towards scalable and channel-robust radio frequency fingerprint identification for LoRa,” IEEE Trans. Inf. Forensics Security, vol. 17, pp. 774–787, 2022.] proposed a data augmentation method based on random channel parameter selection. This method models the multipath fading channel effect based on the exponential Power Delay Profile (PDP). The radio frequency signal under good channel conditions is transformed by randomly selecting the root mean square delay spread and Rician factor, and is added to the training set of the identification model to improve the identification accuracy of LoRa devices under office multipath fading channel scenarios.Patent CN117136522A proposes a method, architecture, device and system related to data enhancement of radio frequency (RF) data for improved RF fingerprinting, which is based on the multi-path channel parameters in the PDP model to implement FIR filter taps and apply the filter on LoRa signal samples to simulate the real channel impulse response to transform the RF signal and improve the recognition accuracy under different multi-path channel fading scenarios on different days. However, these model-driven data enhancement methods are limited by the prior knowledge of a single channel model, and the recognition network may discard important robust features under complex multi-path channel conditions, resulting in a decrease in recognition performance.

[0005] In recent years, diffusion model as a new generation model can generate synthetic signals that conform to the characteristics of multipath fading channels in LoRa communication as an upstream model. Through data augmentation, it provides a large number of signal samples similar to the actual communication scenario of LoRa for training RFF identification model to improve the generalization ability of the model and ultimately improve the recognition performance of the deployed identification model in complex LoRa multipath fading environment. At present, diffusion model has made breakthrough progress in the generation task of wireless communication fields such as semantic communication, integrated sensing and radio frequency signal identification. For example, a radio frequency signal generation method based on time-frequency diffusion model [G. Chi, Z. Yang, C. Wu, J. Xu, Y. Gao, Y. Liu, and T. X. Han,“RF-Diffusion: Radio signal generation via time-frequency diffusion,” in Proc. 30th Annu. Int. Conf. Mobile Comput. Netw. (MobiCom), New York, USA, May. 2024, pp. 77–92.] adds time-domain Gaussian noise in the forward process to destroy the signal amplitude details, and uses conditional guidance in the reverse process to generate signals; An electromagnetic property sensing method based on conditional diffusion model [Y. Jiang, F. Gao, S. Jin, and T. J. Cui, “Electromagnetic property sensing based on diffusion model in ISAC system,” IEEE Trans. Wireless Commun., vol. 24, no. 3, pp. 2036–2051, Mar. 2025] introduces Gaussian noise in the forward process, and uses sensing channel estimation in the reverse process to guide the generation of the shape and electromagnetic properties of the target. However, existing research usually only considers the additive interference modeling in the forward process, and ignores the multiplicative interference influence brought by multipath fading effect, so the signal samples generated by diffusion model may be difficult to match the statistical characteristics of multipath fading channel, which may lead to the decline of the recognition performance of RFF identification model based on these signals in LoRa multipath fading channel communication environment. SUMMARY

[0006] The present application aims to at least solve one of the above technical problems in the art. To this end, one object of the present application is to propose a LoRa device radio frequency fingerprinting method based on a conditional diffusion model, which models the data weight of the conditional diffusion model as a channel fading effect and the noise weight as environmental noise, and accurately models the multipath channel fading characteristics in LoRa communication. A large number of synthetic signals that preserve the RFF features are generated by the conditional diffusion model, and the synthetic signals are added to the training of the radio frequency fingerprinting network, which significantly improves the device recognition accuracy under a multipath channel, provides a solution to the device authentication problem in LoRa long-distance communication, and has a broad application prospect.

[0007] A second object of the present application is to propose a computer-readable storage medium.

[0008] A third object of the present application is to propose a computer device.

[0009] A fourth object of the present application is to propose a LoRa device radio frequency fingerprinting device based on a conditional diffusion model.

[0010] To achieve the above object, the first aspect of the present application proposes a LoRa device radio frequency fingerprinting method based on a conditional diffusion model, which comprises the following steps: obtaining radio frequency signals of different LoRa devices; preprocessing the radio frequency signals to obtain a training data set; constructing a conditional diffusion model, inputting the radio frequency signals in the training data set into the conditional diffusion model for training to obtain a trained conditional diffusion model, wherein the Rician fading multiplicative interference and additive Gaussian noise are introduced in the training process to obtain a radio frequency fading signal, and the radio frequency fading signal and the Rician factor are input into the prediction network of the conditional diffusion model to obtain a network output value, and the prediction network is optimized by the mean square error loss function of the network output value and the radio frequency signal; generating a radio frequency fading signal using the trained conditional diffusion model, and adding the generated radio frequency fading signal to the training data set to obtain an enhanced data set; inputting the enhanced data set into a pre-constructed radio frequency fingerprinting model to obtain a trained radio frequency fingerprinting model, so as to use the trained radio frequency fingerprinting model to identify the radio frequency fading signal of a LoRa device to be identified, to obtain the category corresponding to the LoRa device to be identified; thereby improving the device recognition accuracy under a multipath channel.

[0011] In addition, the LoRa device radio frequency fingerprinting method based on the conditional diffusion model according to the above embodiments of the present application can also have the following additional technical features:

[0012] Optionally, the Rayleigh fading multiplicative interference and additive Gaussian noise are introduced in the training process to obtain the radio frequency fading signal, including: obtaining the total number of forward diffusion time steps and the number of Rayleigh factors, and obtaining the channel gain corresponding to all diffusion time steps of different Rayleigh factors by using the Rayleigh channel model; obtaining the batch size of training data, and obtaining batch training data by randomly sampling multiple radio frequency signals from the training data set; randomly sampling the diffusion time step, the Rayleigh factor and the Gaussian noise, inputting the batch training data into the forward fading diffusion process of the conditional diffusion model, calculating the cumulative signal strength and the cumulative noise strength, and applying different channel gains corresponding to the multiplicative interference to the radio frequency signal by the forward process Gaussian model, to obtain the radio frequency fading signal.

[0013] Optionally, the trained conditional diffusion model is used to generate the radio frequency fading signal, including: sampling the Gaussian noise signal, setting the denoising time step and the Rayleigh factor, and calculating the cumulative signal strength and the cumulative noise strength corresponding to the denoising time step; inputting the Gaussian noise signal and the Rayleigh factor into the trained prediction network to obtain the output value, and denoising all Gaussian noise signals by using non-Markov sampling to obtain the radio frequency fading signal.

[0014] Optionally, the trained radio frequency fingerprint identification model is used to identify the radio frequency fading signal of the to-be-identified LoRa device to obtain the category corresponding to the to-be-identified LoRa device, including: obtaining the radio frequency fading signal transmitted by the to-be-identified device through the multipath fading channel; inputting the radio frequency fading signal transmitted by the to-be-identified device into the trained radio frequency fingerprint identification model to extract the to-be-identified device fingerprint feature in the radio frequency fading signal, and taking the number corresponding to the maximum Softmax probability value output by the prediction network as the category of the to-be-identified device.

[0015] To achieve the above object, the second aspect of the present application provides a computer readable storage medium, which stores a LoRa device radio frequency fingerprint identification program based on a conditional diffusion model, and the LoRa device radio frequency fingerprint identification program based on the conditional diffusion model realizes the LoRa device radio frequency fingerprint identification method based on the conditional diffusion model when executed by a processor.

[0016] To achieve the above object, the third aspect of the present application provides a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and when the processor executes the program, the LoRa device radio frequency fingerprint identification method based on the conditional diffusion model is realized.

[0017] To achieve the above object, the fourth aspect of the present application provides a LoRa device radio frequency fingerprint identification device based on a conditional diffusion model, comprising: an acquisition module for acquiring radio frequency signals of different LoRa devices; a preprocessing module for preprocessing the radio frequency signals to obtain a training data set; a training module for constructing a conditional diffusion model, inputting the radio frequency signals in the training data set into the conditional diffusion model for training to obtain a trained conditional diffusion model, wherein the Rician fading multiplicative interference and additive Gaussian noise are introduced in the training process to obtain a radio frequency fading signal, and the radio frequency fading signal and the Rician factor are input into the prediction network of the conditional diffusion model to obtain a network output value, and the prediction network is optimized through the mean square error loss function of the network output value and the radio frequency signal; a data set enhancement module for generating a radio frequency fading signal using the trained conditional diffusion model, and adding the generated radio frequency fading signal to the training data set to obtain an enhanced data set; an identification module for inputting the enhanced data set into a pre-constructed radio frequency fingerprint identification model to obtain a trained radio frequency fingerprint identification model, so as to identify the radio frequency fading signal of the LoRa device to be identified using the trained radio frequency fingerprint identification model to obtain the category corresponding to the LoRa device to be identified.

[0018] In addition, the LoRa device radio frequency fingerprint identification device based on the conditional diffusion model according to the above-mentioned embodiments of the present application can have the following additional technical features:

[0019] Optionally, the training module is further configured to acquire the total number of forward diffusion time steps and the number of Rician factors, and obtain channel gains corresponding to all diffusion time steps of different Rician factors using a Rician channel model; acquire a training data batch size to randomly sample a plurality of radio frequency signals from the training data set to obtain batch training data; randomly sample diffusion time steps, Rician factors and Gaussian noise, input the batch training data into the forward fading diffusion process of the conditional diffusion model, calculate the cumulative signal strength and the cumulative noise strength, and apply multiplicative interference corresponding to different channel gains to the radio frequency signal through the forward process Gaussian model to obtain a radio frequency fading signal.

[0020] Optionally, the data set enhancement module is further configured to sample Gaussian noise signals, set a denoising time step and a Rician factor, calculate the cumulative signal strength and the cumulative noise strength corresponding to the denoising time step; input the Gaussian noise signal and the Rician factor into the trained prediction network to obtain an output value, and denoise all Gaussian noise signals using non-Markov sampling to obtain a radio frequency fading signal.

[0021] Optionally, the identification module is further configured to: acquire a radio frequency fading signal transmitted by the to-be-identified device through a multipath fading channel; input the radio frequency fading signal transmitted by the to-be-identified device into the trained radio frequency fingerprint identification model to extract a to-be-identified device fingerprint feature in the radio frequency fading signal, and take the number corresponding to the maximum Softmax probability value output by the prediction network as the category of the to-be-identified device. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A flowchart of a LoRa device radio frequency fingerprint identification method based on a conditional diffusion model according to an embodiment of the present application;

[0023] Figure 2 A framework diagram of a LoRa device radio frequency fingerprint identification method based on a conditional diffusion model according to an embodiment of the present application;

[0024] Figure 3 A visual time-domain waveform diagram of a synthesized LoRa radio frequency signal and a time-domain LoRa fading signal according to an embodiment of the present application;

[0025] Figure 4 An influence diagram of a data enhancement ratio on the performance of a LoRa device identification algorithm under different signal-to-noise ratios according to an embodiment of the present application;

[0026] Figure 5 A block diagram of a LoRa device radio frequency fingerprint identification device based on a conditional diffusion model according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] Embodiments of the present application will be described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and should not be understood as limiting the present application.

[0028] In order to better understand the above technical solutions, exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0029] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in combination with the accompanying drawings and specific embodiments.

[0030] REFERENCE Figure 1As shown, the LoRa device radio frequency fingerprinting method based on the conditional diffusion model of this invention includes the following steps:

[0031] S101 acquires radio frequency signals from different LoRa devices.

[0032] In other words, the radio frequency signals of different LoRa devices and their corresponding tags are obtained through signal acquisition.

[0033] S102, preprocess the radio frequency signal to obtain the training dataset.

[0034] It should be noted that existing signal detection, interception, and energy normalization processes can be used for preprocessing radio frequency signals, and this application does not impose any specific limitations on this.

[0035] S103. Construct a conditional diffusion model by inputting the radio frequency (RF) signal from the training dataset into the conditional diffusion model for training to obtain a trained conditional diffusion model. During the training process, Ricean fading multiplicative interference and additive Gaussian noise are introduced to obtain the RF fading signal. The RF fading signal and Ricean factor are input into the prediction network of the conditional diffusion model to obtain the network output value. The prediction network is optimized by the mean square error loss function of the network output value and the RF signal.

[0036] In other words, an RF channel conditional diffusion model (RF-CCDM) is established, which includes two stages: forward fading diffusion and reverse fading reconstruction.

[0037] As an example, Ricean fading multiplicative interference and additive Gaussian noise are introduced during the training process to obtain the radio frequency fading signal. This includes: obtaining the total forward diffusion time steps and the number of Ricean factors, and using a Ricean channel model to obtain the channel gain corresponding to all diffusion time steps for different Ricean factors; obtaining the batch size of the training data to randomly sample multiple radio frequency signals from the training dataset to obtain batch training data; randomly sampling the diffusion time steps, Ricean factors, and Gaussian noise, inputting the batch training data into the forward fading diffusion process of the conditional diffusion model, calculating the cumulative signal strength and cumulative noise strength, and applying multiplicative interference corresponding to different channel gains to the radio frequency signal through the forward process Gaussian model to obtain the radio frequency fading signal.

[0038] Specifically, in the forward fading diffusion stage, the total number of diffusion steps in the forward process is first determined. And the number of Rice factors Different Rice factors are obtained through the Rice channel model. Channel gain for all diffusion time steps Subsequently, the training batch size was determined. A sampling model based on uniform distribution is adopted. ,in This represents the dataset sampled by the sampling model, from a training dataset with L samples. The number of randomly selected sampling points is M, and the vector dimension is... radio frequency signals ,Right now Then, the same method was used. Sampling model, random sampling forward diffusion time steps Rice factor Gaussian model used Sampled Gaussian noise ,Right now The forward diffusion step is calculated using a cosine scheduler. From 1 to t Signal retention ratio And calculate the noise ratio Then calculate the first t Cumulative signal strength at forward diffusion time step and cumulative noise intensity .get Unit vector, and calculate the Gaussian model of the forward process. mean and variance Sampling model based on forward process Gaussian model For the original radio frequency signal By applying multiplicative interference corresponding to different channel gains, radio frequency fading signals are obtained. ;

[0039]

[0040] It should be noted that, compared with the traditional forward diffusion process, the modified forward process... When the signal strength coefficient becomes the cumulative signal strength The noise intensity coefficient becomes Changing these two coefficients allows the forward diffusion process to introduce different Rice factors. The corresponding channel fading characteristics are simulated to generate multiplicative interference under different LoRa communication multipath fading conditions, making the original radio frequency signal... Converted to radio frequency fading signal Finally, the radio frequency fading signal... and Rice factor Input diffusion model prediction network Obtain network output value The output value is compared with the original radio frequency signal. The mean squared error loss function is used to optimize the network.

[0041] S104. The trained conditional diffusion model is used to generate radio frequency fading signals, and the generated radio frequency fading signals are added to the training dataset to obtain the enhanced dataset.

[0042] As an example, a trained conditional diffusion model is used to generate a radio frequency fading signal, including: sampling a Gaussian noise signal, setting a denoising time step and a Rice factor, calculating the cumulative signal strength and cumulative noise strength corresponding to the denoising time step; inputting the Gaussian noise signal and Rice factor into a trained prediction network to obtain an output value, and using non-Markov sampling to denoise all Gaussian noise signals to obtain a radio frequency fading signal.

[0043] Specifically, in the reverse fading reconstruction phase, K Gaussian noise signals are sampled. Set the noise reduction time step and Rice factor The forward diffusion step is calculated using a cosine scheduler. From 1 to t Signal retention ratio And calculate the noise ratio Then calculate the first t and the T Cumulative signal strength at denoising time step , and cumulative noise intensity , Gaussian noise signal is obtained. and Rice factor Input to the training prediction network Output value Through non-Markov sampling Signal Denoising to faded signal ;

[0044]

[0045] It should be noted that, compared with the traditional reverse process sampling function, the modified sampling process... In the diffusion model network parameters are Predicting network input Rice factor added Guiding conditions, this guiding method makes the Gaussian noise signal to radio frequency fading signal The denoised trajectory can satisfy the Rice factor. The corresponding multipath fading conditions are used to generate fading signals that conform to the channel characteristics.

[0046] S105, the enhanced dataset is input into the pre-built radio frequency fingerprint recognition model to obtain the trained radio frequency fingerprint recognition model, so as to use the trained radio frequency fingerprint recognition model to identify the radio frequency fading signal of the LoRa device to be identified, so as to obtain the category corresponding to the LoRa device to be identified.

[0047] As an example, a trained radio frequency fingerprinting model is used to identify the radio frequency fading signal of the LoRa device to be identified in order to obtain the category corresponding to the LoRa device to be identified. This includes: acquiring the radio frequency fading signal transmitted by the device to be identified after propagation through a multipath fading channel; inputting the radio frequency fading signal transmitted by the device to be identified into the trained radio frequency fingerprinting model to extract the fingerprint features of the device to be identified in the radio frequency fading signal; and taking the number corresponding to the maximum Softmax probability value output by the prediction network as the category of the device to be identified.

[0048] In summary, the LoRa device RF fingerprinting method based on the conditional diffusion model according to embodiments of the present invention first collects RF signals from different LoRa devices, preprocesses them, and adds them to a training dataset. The conditional diffusion model is then trained using this training dataset. In the forward fading diffusion process of this conditional diffusion model, a Rice factor is introduced to model the multiplicative interference caused by the multipath channel conditions of LoRa communication. Furthermore, a Rice factor-guided condition is added during the reverse fading reconstruction process, generating a large number of LoRa device RF fading signals that conform to the multipath channel characteristics. During the LoRa device identification stage, the large number of RF fading signals generated by the RF channel conditional diffusion model are added to the original dataset for training the RF fingerprinting network. The trained RF fingerprinting network is then deployed on the receiving device to achieve identification of the device under LoRa multipath fading communication conditions.

[0049] Furthermore, to better understand the above technical means, a specific embodiment will be provided for detailed explanation, such as... Figure 2 As shown, the LoRa device RF fingerprinting method based on the conditional diffusion model includes a conditional diffusion model training phase, an RF fingerprinting network training phase, and a LoRa device identification phase. Figure 2 As shown, the method includes the following steps:

[0050] S1. During the diffusion model training phase, different LoRa device radio frequency signals are collected and preprocessed to generate a training set. An RF channel conditional diffusion model (RF-CCDM) is designed, and LoRa device radio frequency signals are randomly sampled from the training dataset to train the prediction network of the RF channel conditional diffusion model.

[0051] S2. During the training phase of the radio frequency fingerprinting network, LoRa fading signals are generated using the trained RF channel conditional diffusion model. These LoRa fading signals, along with the original dataset, serve as an augmented dataset for training the radio frequency fingerprinting network.

[0052] S3. During the LoRa device identification phase, the wireless signal transmitted by the device to be identified propagates through a multipath fading channel to reach the receiver. The pre-trained radio frequency fingerprinting network deployed on the receiver is responsible for extracting the transmitter fingerprint features from the wireless signal and using the number corresponding to the maximum Softmax probability value as the LoRa device tag.

[0053] Furthermore, the diffusion model training phase in step S1 specifically includes:

[0054] S11. First, determine the total forward diffusion time step. Number of Rice factors Different Rice factors are obtained through the Rice channel model. All diffusion time steps Corresponding channel gain .

[0055] S12, Then determine the batch size of the training data. From the sample size RF signals with 4000 random sampling points were selected from the LoRa device training set. Batch training data; random sampling diffusion time step Rice factor and Gaussian noise The batch training data is input into the forward fading diffusion process of the RF channel conditional diffusion model, and a cosine scheduler is used to calculate the forward diffusion steps. From 1 to t Signal retention ratio And calculate the noise ratio Then calculate the first t Cumulative signal strength at forward diffusion time step and cumulative noise intensity Through the forward process Gaussian model For the original radio frequency signal By applying multiplicative interference corresponding to different channel gains, radio frequency fading signals are obtained. .

[0056] S13. Finally, the radio frequency fading signal and Rice factor Input diffusion model prediction network Obtain network output value The output value is compared with the original radio frequency signal. The mean squared error loss function is used to optimize the network.

[0057] Furthermore, the radio frequency fingerprint recognition network training phase in step S2 specifically includes:

[0058] S21, Sampling Number of Gaussian noise signals Set the noise reduction time step and Rice factor The forward diffusion step is calculated using a cosine scheduler. From 1 to t Signal retention ratio And calculate the noise ratio Then calculate the first t and the T Cumulative signal strength at denoising time step and and cumulative noise intensity and Gaussian noise signal is obtained. and Rice factor Input to the training prediction network Output value Through non-Markov sampling All noise signals Denoising to faded signal .

[0059] S22, the synthesized The number of fading signals added to the number of samples The LoRa device training set was used to obtain a 140k augmented dataset, which was then used to train the radio frequency fingerprinting network.

[0060] As an example, to intuitively evaluate the channel fitting effect of the RF channel conditional diffusion model, the visualization results of the channel characteristic fitting between the LoRa signal samples generated by the RF channel conditional diffusion model and the time-domain fading signal in actual LoRa communication are shown below. Figure 3 As shown in the figure, the signal generated by the RF channel conditional diffusion model can fit the time-domain waveform of the LoRa fading signal well, and the channel fitting signal-to-noise ratio remains above 20 dB.

[0061] As an example, to evaluate the impact of different SNRs on the performance of the LoRa device identification algorithm, two typical radio frequency fingerprinting networks, DR-RFF and CGAN, were introduced, with the data augmentation ratio of the training set based on a sample size of 100k. The data augmentation effects under different SNRs are shown below. Figure 4As shown in the figure, with the increase of SNR, the average recognition accuracy of both the baseline and data-augmented models of the two recognition networks is significantly improved. Based on the optimal data augmentation ratio of 40%, compared to the baseline model, although the variance of the data-augmented model DR-RFF-augmentation only increased by 4.12%, its average performance was improved by 13.17% compared to the DR-RFF-baseline; and the variance of the data-augmented model CGAN-augmentation only increased by 4.98%, but its average recognition performance was improved by 36.51% compared to the CGAN-baseline.

[0062] In summary, this application introduces Ricean fading multiplicative interference and additive Gaussian noise in the forward diffusion stage to simulate the multipath fading characteristics of real LoRa communication. In the reverse fading reconstruction stage, the Ricean factor is used to guide the generation of high-quality RF signals that retain RF fingerprint features and channel characteristics. Finally, the original signal and the signal generated by the conditional diffusion model together constitute the training dataset to train the recognition model, effectively improving the recognition accuracy of LoRa devices under multipath fading channel conditions. Specifically, in fading signal generation, the signal generated by the RF channel conditional diffusion model can better fit the channel statistical characteristics of the actual LoRa RF fading signal, and the channel fitting signal-to-noise ratio remains above 20 dB; in RF fingerprint recognition, the preprocessed original LoRa signal and the high-speed sampled synthetic signal together constitute the training dataset, where adding 40% of the generated signal to the training set can improve the recognition performance of LoRa devices in multipath fading scenarios by 24.84%.

[0063] In addition, the present invention also proposes a computer-readable storage medium storing a LoRa device radio frequency fingerprinting program based on a conditional diffusion model, which, when executed by a processor, implements the LoRa device radio frequency fingerprinting method based on the conditional diffusion model as described above.

[0064] In addition, this invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the LoRa device radio frequency fingerprinting method based on the conditional diffusion model as described above.

[0065] Figure 5 This is a block diagram of a LoRa device radio frequency fingerprint recognition device based on a conditional diffusion model according to an embodiment of the present invention. Figure 5 As shown, the LoRa device radio frequency fingerprint recognition device includes: an acquisition module 10, a preprocessing module 20, a training module 30, a dataset augmentation module 40, and a recognition module 50;

[0066] The system comprises the following modules: Acquisition module 10 acquires radio frequency (RF) signals from different LoRa devices; Preprocessing module 20 preprocesses the RF signals to obtain a training dataset; Training module 30 constructs a conditional diffusion model by inputting the RF signals from the training dataset into the model for training, resulting in a trained model. During training, Ricean fading multiplicative interference and additive Gaussian noise are introduced to generate RF fading signals. The RF fading signals and Ricean factors are input into the prediction network of the conditional diffusion model to obtain the network output value. The prediction network is then optimized using the mean square error loss function of the network output value and the RF signal; Dataset augmentation module 40 generates RF fading signals using the trained conditional diffusion model and adds these signals to the training dataset to obtain an augmented dataset; and Identification module 50 inputs the augmented dataset into a pre-constructed RF fingerprint recognition model to obtain a trained RF fingerprint recognition model. This model is then used to identify the RF fading signals of the LoRa device to be identified, thus determining the corresponding category of the LoRa device.

[0067] As an example, the training module 30 is further configured to: obtain the total forward diffusion time steps and the number of Rice factors, and use the Rice channel model to obtain the channel gain corresponding to all diffusion time steps with different Rice factors; obtain the batch size of training data to randomly sample multiple radio frequency signals from the training dataset to obtain batch training data; randomly sample the diffusion time steps, Rice factors and Gaussian noise, input the batch training data into the forward fading diffusion process of the conditional diffusion model, calculate the cumulative signal strength and cumulative noise strength, and apply multiplicative interference corresponding to different channel gains to the radio frequency signal through the forward process Gaussian model to obtain the radio frequency fading signal.

[0068] As an example, the dataset augmentation module 40 is also used to sample Gaussian noise signals, set denoising time steps and Rice factors, calculate the cumulative signal strength and cumulative noise strength corresponding to the denoising time steps; input the Gaussian noise signals and Rice factors into the trained prediction network to obtain output values, and use non-Markov sampling to denoise all Gaussian noise signals to obtain radio frequency fading signals.

[0069] As an example, the identification module 50 is further configured to: acquire the radio frequency fading signal transmitted by the device to be identified through the multipath fading channel; input the radio frequency fading signal transmitted by the device to be identified into the trained radio frequency fingerprint recognition model to extract the fingerprint features of the device to be identified in the radio frequency fading signal; and take the number corresponding to the maximum Softmax probability value output by the prediction network as the category of the device to be identified.

[0070] It should be noted that the explanations and descriptions of the embodiments of the LoRa device radio frequency fingerprinting method based on the conditional diffusion model described above also apply to the LoRa device radio frequency fingerprinting device based on the conditional diffusion model in this embodiment, and will not be repeated here.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0072] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0075] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0076] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0077] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0078] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0079] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0080] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0081] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0082] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A LoRa device radio frequency fingerprinting method based on a conditional diffusion model, characterized in that, The method comprises the following steps: Obtaining radio frequency signals of different LoRa devices; Preprocessing the radio frequency signals to obtain a training data set; Building a conditional diffusion model, inputting the radio frequency signals in the training data set into the conditional diffusion model for training to obtain a trained conditional diffusion model, wherein, in the training process, the Rice fading multiplicative interference and the additive Gaussian noise are introduced to obtain a radio frequency fading signal, and the radio frequency fading signal and the Rice factor are input into a prediction network of the conditional diffusion model to obtain a network output value, and the prediction network is optimized through a mean square error loss function of the network output value and the radio frequency signal; Generating the radio frequency fading signal by using the trained conditional diffusion model, and adding the generated radio frequency fading signal to the training data set to obtain an enhanced data set; Inputting the enhanced data set into a pre-built radio frequency fingerprint identification model to obtain a trained radio frequency fingerprint identification model, so as to identify the radio frequency fading signal of a to-be-identified LoRa device by using the trained radio frequency fingerprint identification model to obtain a category corresponding to the to-be-identified LoRa device; In the training process, the Rice fading multiplicative interference and the additive Gaussian noise are introduced to obtain a radio frequency fading signal, comprising: Obtaining the total number of forward diffusion time steps and the number of Rice factors, and obtaining the channel gain corresponding to all diffusion time steps of different Rice factors by using a Rice channel model; Obtaining the batch size of the training data, randomly sampling a plurality of radio frequency signals from the training data set to obtain batch training data; Randomly sampling diffusion time steps, Rice factors and Gaussian noise, inputting the batch training data into the forward fading diffusion process of the conditional diffusion model, calculating the cumulative signal strength and the cumulative noise strength, and applying the multiplicative interference corresponding to different channel gains to the radio frequency signal through the forward process Gaussian model to obtain the radio frequency fading signal.

2. The LoRa device radio frequency fingerprinting method based on conditional diffusion model according to claim 1, wherein, Generating the radio frequency fading signal by using the trained conditional diffusion model, comprising: Sampling the Gaussian noise signal, setting the denoising time step and the Rice factor, and calculating the cumulative signal strength and the cumulative noise strength corresponding to the denoising time step; Inputting the Gaussian noise signal and the Rice factor into the trained prediction network to obtain an output value, and denoising all Gaussian noise signals by using non-Markov sampling to obtain the radio frequency fading signal. 3.The LoRa device radio frequency fingerprinting method based on conditional diffusion model of claim 1, wherein, Identifying the radio frequency fading signal of the to-be-identified LoRa device by using the trained radio frequency fingerprint identification model to obtain a category corresponding to the to-be-identified LoRa device, comprising: Obtaining the radio frequency fading signal transmitted by the to-be-identified device through the multipath fading channel; Inputting the radio frequency fading signal transmitted by the to-be-identified device into the trained radio frequency fingerprint identification model to extract the to-be-identified device fingerprint feature in the radio frequency fading signal, and taking the number corresponding to the maximum Softmax probability value output by the prediction network as the category of the to-be-identified device.

4. A computer-readable storage medium, characterized in that, The computer device has a LoRa device radio frequency fingerprint identification program based on a conditional diffusion model stored thereon, and the LoRa device radio frequency fingerprint identification program based on the conditional diffusion model is executed by the processor to implement the LoRa device radio frequency fingerprint identification method based on the conditional diffusion model in any one of claims 1-3.

5. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the LoRa device radio frequency fingerprint identification method based on the conditional diffusion model in any one of claims 1-3 when executing the computer program.

6. A LoRa device radio frequency fingerprint recognition device based on a conditional diffusion model, characterized in that, Comprise: An acquisition module for acquiring radio frequency signals of different LoRa devices; A preprocessing module for preprocessing the radio frequency signals to obtain a training data set; A training module for constructing a conditional diffusion model, inputting the radio frequency signals in the training data set into the conditional diffusion model for training to obtain a trained conditional diffusion model, wherein the Rician fading multiplicative interference and additive Gaussian noise are introduced in the training process to obtain a radio frequency fading signal, and the radio frequency fading signal and the Rician factor are input into the prediction network of the conditional diffusion model to obtain a network output value, and the prediction network is optimized through the mean square error loss function of the network output value and the radio frequency signal; A data set enhancement module for generating a radio frequency fading signal using the trained conditional diffusion model, and adding the generated radio frequency fading signal to the training data set to obtain an enhanced data set; An identification module for inputting the enhanced data set into a pre-constructed radio frequency fingerprint identification model to obtain a trained radio frequency fingerprint identification model, so as to identify the radio frequency fading signal of the LoRa device to be identified using the trained radio frequency fingerprint identification model to obtain the category corresponding to the LoRa device to be identified; Wherein, the training module is further used for, Obtaining the total time step of the forward diffusion and the number of Rician factors, and obtaining the channel gain corresponding to all diffusion time steps of different Rician factors using the Rician channel model; obtaining the training data batch size to randomly sample multiple radio frequency signals from the training data set to obtain batch training data; randomly sampling diffusion time steps, Rician factors and Gaussian noise, inputting the batch training data into the forward fading diffusion process of the conditional diffusion model, calculating the cumulative signal strength and the cumulative noise strength, and applying different channel gains corresponding to the multiplicative interference to the radio frequency signal through the forward process Gaussian model to obtain the radio frequency fading signal.

7. The LoRa device radio frequency fingerprinting apparatus based on a conditional diffusion model of claim 6, wherein, The data set enhancement module is further used for, Sampling Gaussian noise signals, setting the denoising time step and the Rician factor, calculating the cumulative signal strength and the cumulative noise strength corresponding to the denoising time step; inputting the Gaussian noise signal and the Rician factor into the trained prediction network to obtain the output value, and denoising all Gaussian noise signals using non-Markov sampling to obtain the radio frequency fading signal.

8. The LoRa device radio frequency fingerprinting apparatus based on a conditional diffusion model of claim 6, wherein, The identification module is further used for, Obtaining the radio frequency fading signal transmitted by the device to be identified through the multipath fading channel; inputting the radio frequency fading signal transmitted by the device to be identified into the trained radio frequency fingerprint identification model to extract the fingerprint features of the device to be identified in the radio frequency fading signal, and taking the number corresponding to the maximum Softmax probability value output by the prediction network as the category of the device to be identified.

Images