Lightweight industrial wireless physical layer security authentication method based on block chain

Abstract

The invention relates to the field of industrial wireless physical layer security, in particular to a lightweight industrial wireless physical layer security authentication method based on a block chain. According to the method, a multi-node cooperative authentication problem in a fixed scene is considered, and the method comprises two stages: in an offline training stage, a training sample of equipment is collected, and an optimal model is constructed and trained through a block chain; and in the online authentication stage, the equipment is authenticated according to the trained model. According to the method, channel state information between devices is collected, feature extraction is performed on the channel state information, then a local topological graph is constructed through a relative distance, an azimuth angle and an elevation angle between the model output devices, the local topological graph is uploaded to a block chain to trigger an intelligent contract to aggregate a global topological graph, and then the global topological graph is broadcasted to each node, so that collaborative error feedback optimization model training is realized. And finally, the trained model is used for positioning, and high-reliability security authentication of the industrial equipment is realized.

Description

Technical Field

[0001] This invention relates to the field of industrial wireless physical layer security, specifically a lightweight industrial wireless physical layer security authentication method based on blockchain. Background Technology

[0002] With the development of information technology and intelligent manufacturing, intelligent terminal devices are widely used in automated container factories, unmanned warehouses, and smart logistics. Compared with traditional industrial wired communication, wireless networks have the advantages of low cost, good scalability, and flexible deployment, which make them suitable for the ubiquitous sensing and collaborative reconfiguration needs of intelligent manufacturing. However, the open nature of wireless propagation media introduces many security threats.

[0003] Traditional authentication relies on keys, which depend on encryption algorithms, making them vulnerable to cracking and computationally intensive. In recent years, new authentication factors based on intrinsic entity characteristics have become increasingly prevalent. For industrial equipment, these intrinsic characteristics are the physical layer features of the device. Channel state information, on the other hand, focuses on the characteristics of the wireless channel, exhibiting dynamism and real-time performance, making it suitable for complex wireless environments.

[0004] Location-based channel fingerprinting authentication algorithms focus on leveraging the spatial differences in channel states to distinguish channel characteristics at different locations. These algorithms typically establish a mapping between location and channel features by sampling channel features at known locations. Request signals sent by legitimate nodes and attackers at different locations experience different channel fading at the authenticator, resulting in different channel states. Machine learning-based location classification methods are widely used to address dynamic changes in channel states. This method uses a machine learning model to establish a channel information-location mapping model at each deployment location of the device, thereby achieving location identification for authentication.

[0005] A new type of spoofing attack targeting CSI has recently emerged. Attackers manipulate CSI using physical layer devices or the radio frequency environment, causing the receiver to receive the same signal as the legitimate sender, thus simulating the location of the legitimate sender. Existing authentication methods are ineffective against this new attack. Since attackers cannot simultaneously simulate multiple links, collaborative authentication can mitigate this type of attack. Blockchain, a decentralized distributed ledger, possesses characteristics such as immutability and data transparency. It can not only enhance the security of collaborative authentication but also ensure the credibility and transparency of the authentication process, thereby improving the security of industrial wireless communication networks.

[0006] To better adapt to the high reliability and low power consumption characteristics of real-world industrial scenarios, a new security authentication method needs to be designed to address the security issues of industrial wireless networks. Summary of the Invention

[0007] This invention provides a lightweight industrial wireless physical layer security authentication method based on blockchain, which can better adapt to the reliability, security and low overhead of actual industrial scenarios.

[0008] The present invention adopts the following technical solution:

[0009] A lightweight industrial wireless physical layer security authentication method based on blockchain achieves physical layer security authentication through multi-node collaborative positioning and authentication strategies, including the following steps:

[0010] Offline training:

[0011] Local training phase: Collect channel state information from the device and extract channel state information features; train the localization model based on the extracted channel state information features, and construct a local topology map based on the output of the localization model;

[0012] Global optimization phase: Upload local topology map, construct global topology map through blockchain; optimize positioning model using collaborative error feedback;

[0013] Online verification:

[0014] The device is authenticated based on the trained positioning model.

[0015] The channel state information of the acquisition device is specifically as follows:

[0016] Sure The fixed deployment location of each device is, at coordinates [coordinates missing]. The device collects corresponding channel state information samples through wireless channel communication, denoted as: ,in, For equipment As the receiving end, device When used as a transmitter, the equipment and equipment The channel gain matrix between them is expressed as follows:

[0017] ;in Sampling time, for Multipath quantity at different times For the first Amplitude attenuation along the path, No. Phase offset of the path, For Dirichlet impulse function, For the delay of signal propagation, For the first The time delay of the path;

[0018] After acquiring channel state information, each device records its connection relationships with other devices. : ,in, Indicates device and equipment There is a communication connection between them. Indicates device The set of neighbors.

[0019] The extraction of channel state information features specifically includes:

[0020] Feature extraction is performed on the collected channel state information, and the amplitude, phase difference, path gain, and angle of arrival are combined to form a feature vector between the signal transmitted by transmitting device i and the corresponding receiving device j. :

[0021] ;

[0022] in, The amplitude represents the change in signal strength of the wireless signal during propagation; Phase difference represents the spatial characteristics of a wireless signal between different subcarrier frequencies or different antennas; Path gain represents the power attenuation caused by the distance the wireless signal travels. Angle of arrival indicates the direction in which a wireless signal reaches the receiving antenna array.

[0023] The angle of arrival was estimated using the MUSIC algorithm, as follows:

[0024] Calculate the covariance matrix of the channel state information received by multiple antennas. For the covariance matrix Perform eigenvalue decomposition to divide it into signal subspaces. and noise subspace ;

[0025]

[0026] in, and It is the corresponding eigenvalue matrix;

[0027] Next, an angle-based steering vector is constructed and interacted with the noise subspace of eigenvalue decomposition to finally obtain the arrival angle. Specifically, each candidate angle Constructing the guiding vector :

[0028] ,

[0029] in, It is the antenna spacing. It is the signal wavelength. This refers to the number of receiving antennas;

[0030] Constructing the MUSIC spectrum:

[0031] ,

[0032] scanning Find the peak position of the MUSIC spectrum by taking the value of the angle of arrival. ,in Corresponding to the transmitting device and receiving equipment .

[0033] The step of training the localization model based on the extracted channel state information features specifically involves:

[0034] Each device will have multiple feature vectors Composition of feature matrix and adjacency matrix As input to the localization model, the model is trained, and the output is the relative distance between devices. Azimuth and elevation angle .

[0035] The construction of a local topology map based on the output of the localization model specifically involves:

[0036] Each device Based on its own coordinates and predicted target equipment Relative position parameters, including distance Azimuth and elevation angle Calculate the predicted coordinates of the target device:

[0037] ,

[0038] ,

[0039] ;

[0040] According to the equipment Construct a local topology graph by combining the connection relationships between devices. :

[0041]

[0042] in, This represents the predicted coordinates of the j-th device.

[0043] The process of uploading a local topology map and constructing a global topology map through blockchain is as follows:

[0044] Each device uploads its constructed local topology map to the blockchain. Upon receiving the map, the blockchain triggers a smart contract to read the device's prediction results from the local topology map and constructs a global topology map using a weighted average method. :

[0045]

[0046] in, Indicates the weighted average calculation device Global coordinates;

[0047] , , ,

[0048] in, It is equipment For equipment The predicted weights, ,

[0049] ,

[0050] in, Indicates the prediction error. M Indicates the ability to predict equipment i The number of transmitting devices, that is, the number of devices i that have established a transmit / receive relationship with device j; Indicates the target device Predicted coordinates Indicates the target device The actual coordinates.

[0051] The aforementioned method for optimizing the positioning model using collaborative error feedback specifically includes:

[0052] Blockchain will create a global topology graph Broadcast to all devices so that each device receives the global coordinates; each device Based on the target device in the global topology diagram global coordinates and predicted coordinates Calculation error:

[0053]

[0054] With the goal of minimizing the error, a loss function is defined, and the localization model parameters are updated through backpropagation.

[0055] 9. A lightweight industrial wireless physical layer security authentication method based on blockchain as described in claim 1, characterized in that the step of authenticating the device identity based on a trained positioning model is as follows:

[0056] The channel state information of the device is acquired in real time, and channel state information features are extracted; a feature matrix is ​​constructed based on the extracted channel state information features. ,Will Adjacency Matrix As input to the trained model, predict the target device. The relative distance, azimuth, and elevation angles are calculated; a local topology map is constructed based on the output of the positioning model.

[0057] Upload a partial topology map and construct a global topology map using the blockchain; the blockchain then aggregates the partial topology maps uploaded by all devices through smart contracts to generate the global topology map. And broadcast it to all devices; each device Compare the target devices in the global topology map Coordinates vs. predicted coordinates, calculation error:

[0058]

[0059] If error If the device exceeds a preset threshold, it is considered an attacker; otherwise, it is considered a legitimate user.

[0060] The present invention has the following beneficial effects and advantages:

[0061] 1. This invention combines blockchain technology to ensure the transparency and credibility of the authentication process, guarantee the integrity and authenticity of data, and enhance the overall trustworthiness of the system.

[0062] 2. This invention combines amplitude, phase difference, path gain, and angle of arrival into a feature vector for collaborative authentication based on multi-device CSI, effectively identifying attackers and accurately judging spoofed signals and identifying disguised attackers in response to new CSI spoofing attacks.

[0063] 3. This invention constructs a local topology map based on the output of the positioning model and builds a global topology map through blockchain. It adopts collaborative error feedback to optimize the positioning model, which can further accurately locate the attacker's position based on the identification of the attacker, realize the traceability of the attack source, and effectively improve the security of industrial wireless network communication. Attached Figure Description

[0064] Figure 1 Schematic diagram of the security authentication architecture of this invention;

[0065] Figure 2A schematic diagram of the spatial relationship between devices in an embodiment of the present invention;

[0066] Figure 3 This embodiment of the invention illustrates a schematic diagram of the device calculating feedback error by comparing the global topology map with the local topology map. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to practical examples.

[0068] like Figure 1 As shown, this invention proposes a lightweight industrial wireless physical layer security authentication method based on blockchain. Its main steps are divided into offline training and online authentication: During offline training, channel state information between devices is first collected, features are extracted, and a neighbor matrix of device connections is obtained. The features and the adjacency matrix are used as model input, and the relative distances, azimuth angles, and elevation angles between devices are output. A local topology map is constructed and uploaded to the blockchain. The blockchain triggers a smart contract to aggregate a global topology map and broadcast it to all devices, enabling collaborative error feedback to optimize model training and improve positioning accuracy. During online authentication, each device uses the trained model for positioning. After constructing a local topology map, it is uploaded to the blockchain. The blockchain aggregates and generates a global topology map, and then performs position comparison to achieve device authentication.

[0069] 1. Offline training phase

[0070] During the offline training phase, channel state information between devices is first collected and features are extracted. A neighbor matrix is ​​then constructed as model input, and the relative distances, azimuth angles, and elevation angles between devices are output. Based on these outputs, a local topology map is constructed and uploaded to the blockchain. This is then aggregated to generate a global topology map, which is broadcast to all devices. Collaborative error feedback optimizes the model, improving device location accuracy.

[0071] 1.1 Training Samples for Acquisition Equipment

[0072] This method considers a typical fixed scenario for industrial wireless devices, which includes... There are 10 devices, each fixedly deployed in a 3D space, with coordinates as follows: The devices communicate with each other via wireless channels. Each communication session does not include any device-related information. and equipment ( These correspond to the transmitting and receiving devices, respectively. The receiving device can estimate the characteristics of the wireless channel by comparing the transmitted and received signals. Specifically, during signal propagation, the signal is affected by various effects such as reflection, refraction, and scattering from obstacles in the environment. The superposition of these effects collectively determines the channel characteristics, also known as channel state information. The data representation of channel state information is a channel gain matrix.

[0073] Each device ) and equipment The channel gain matrix between them is given by , where the channel gain matrix can be represented by . express:

[0074] ,

[0075] ,

[0076] in Sampling time, for Multipath quantity at different times For the first Amplitude attenuation along the path, No. Phase offset of the path, For Dirichlet impulse function, For the first The time delay of the path.

[0077] After acquiring channel state information, each device records the connection relationships with other devices. These connection relationships can be represented by an adjacency matrix. express:

[0078] ,

[0079] in Indicates device and equipment There is a communication connection between them. Indicates device The set of neighbors.

[0080] 1.2 Channel State Information Feature Extraction

[0081] Channel State Information (CSI) reflects the propagation process of a signal between transmitting and receiving devices, serving as a characteristic describing the device's location. Since the wireless channel state between devices is unstable, but the statistical characteristics of the channel are relatively stable, each device will use the acquired CSI data... Stable channel statistical features are extracted, and local channel characteristics and topology are learned through graph neural networks (GNNs).

[0082] (1) Amplitude

[0083] Amplitude represents the change in signal strength caused by physical phenomena such as multipath effects and fading during the propagation of a wireless signal. Signal strength is related to the propagation distance between the transmitter and receiver. Its characteristic estimation formula is:

[0084] ,

[0085] in It is the real part of CSI. It is the imaginary part of CSI.

[0086] (2) Phase difference

[0087] Phase directly reflects the phase change of a wireless signal during propagation caused by path length and direction. However, CSI phase includes unknown offsets introduced by device hardware, which vary with device, channel, and environment, making them difficult to eliminate accurately. Phase difference, on the other hand, is a relative value. Calculating the difference between two phases effectively eliminates the influence of hardware offsets while preserving channel geometry. Phase difference can also capture the spatial characteristics of a signal between different subcarrier frequencies or antennas, especially angle-related variations. Its characteristic estimation formula is:

[0088] ,

[0089] ,

[0090] in and Subcarriers and The phase.

[0091] (3) Path gain

[0092] Path gain is caused by the radiative diffusion of transmitted power and the propagation characteristics of the channel. Path gain is inversely proportional to the distance between devices, and its characteristic estimation formula is as follows:

[0093]

[0094] (4) Angle of arrival

[0095] The angle of arrival (Angle of Arrival) is the direction of a wireless signal's propagation from the transmitter to the receiver along the receiving antenna array. It reflects the spatial geometry of the signal at the receiver and is directly related to the relative angles between devices. The MUSIC algorithm uses the covariance matrix of the received signal to decompose the signal space into a signal subspace and a noise subspace. Utilizing the orthogonality of the noise subspace, the MUSIC algorithm can estimate the Angle of Arrival. Its characteristic estimation formula is:

[0096] 1) Calculate the covariance matrix of the signal using CSI data received from multiple antennas. Specifically, the covariance matrix is ​​related to the channel gain matrix. Related: ,in It is the conjugate transpose of the channel matrix.

[0097] 2) For the covariance matrix Perform eigenvalue decomposition:

[0098] ,

[0099] in These are the eigenvectors of the signal subspace. It is the eigenvector of the noise subspace.

[0100] 3) For each candidate angle Constructing the guiding vector :

[0101] ,

[0102] in It is the antenna spacing. It is the signal wavelength. Constructing the music spectrum:

[0103] ,

[0104] scanning Find the peak positions of the music spectrum using the values ​​of [values], and these angles are the angles of arrival. ,in Corresponding to the transmitting device and receiving equipment .

[0105] The above combination of amplitude, phase difference, path gain, and angle of arrival is the transmitting device. The feature vector between the transmitted signal and the corresponding receiving device j is denoted as... Therefore, the characteristic matrix is ​​defined. ,in It represents the set of amplitude, phase difference, path gain, and angle of arrival.

[0106] 1.3 Training of the Authentication Model

[0107] Each device will have a feature matrix and adjacency matrix As input to the model, it enables the GNN model to learn the set relationships with neighboring devices and output the relative distance, azimuth, and elevation angles with neighboring devices.

[0108] To train this model, a loss function needs to be designed. The first training stage aims to minimize the prediction errors of distance and angle between devices, and the loss function is defined as:

[0109] ,

[0110] ,

[0111] ,

[0112] .

[0113] in, This indicates the error between the predicted distance and the actual distance between the transceiver devices. and These represent the errors between the predicted and actual azimuth and elevation angles between the transmitting and receiving devices, respectively. It is worth noting that, for example... Figure 2 As shown, the azimuth and elevation angles can uniquely determine the angle of arrival. feature.

[0114] Using the designed loss function, the training set data is iteratively optimized to minimize the loss function. We use a gradient descent-based optimization algorithm combined with the Adam optimizer to update the parameters and dynamically adjust the learning rate of each parameter to avoid getting trapped in local optima.

[0115] 1.4 Constructing a local topology graph

[0116] like Figure 2 The diagram shows the spatial relationships between the devices. Each device... Combine its own coordinates and model output target device Calculate the coordinates based on the relative distance, azimuth, and elevation angle:

[0117] ,

[0118] ,

[0119] .

[0120] By combining the connection relationships between devices, a local topology graph is constructed. Represented as, where the equipment :

[0121] .

[0122] 1.5 Generate a global topology graph

[0123] Each device uploads its local topology map, built in the first phase, to the blockchain. The blockchain triggers a smart contract to aggregate the local topology maps of all devices and generate a global topology map.

[0124] .

[0125] During aggregation, it is necessary to combine the predictions of all devices for the target device and use a weighted average to calculate the device's performance. Global coordinates:

[0126] ,

[0127] ,

[0128] ,

[0129] in It is equipment For equipment The prediction weights.

[0130] Use equipment For equipment The prediction error is the reciprocal of the weight. The smaller the error, the more reliable the prediction of the device, and the higher its weight. The formula for calculating the weight is: where... These represent the actual device coordinates:

[0131] ,

[0132] ,

[0133] Where M represents the ability to predict devices The number of transmitting devices, that is, the number of devices i that have established a transmit / receive relationship with device j:

[0134] .

[0135] 1.6 Training of the Cooperative Error Feedback Optimization Model

[0136] The blockchain broadcasts the generated global topology map to all devices. Each device receives the global topology map and extracts the global coordinates of the target device. By comparing with the local topology map, the feedback error is calculated, such as... Figure 3 As shown:

[0137]

[0138] Each device uses error feedback as its optimization objective and updates its model parameters through backpropagation. Through multiple iterations, each device gradually learns the coordinate information in the global topology map, and its prediction results gradually approximate the global coordinates, thereby enhancing authentication accuracy.

[0139] 2. Online authentication stage

[0140] 2.1 Initialization

[0141] Each device Initialize each GNN model and load the best model from the offline training phase. Initialize the blockchain for storing, aggregating, and broadcasting the global topology graph.

[0142] 2.2 Certification Phase

[0143] Each device receives channel state information in real time. Extracting features and generating feature matrices Compare it with the adjacency matrix As model input, the output predicts the target device. The relative distance, azimuth, and elevation angle.

[0144] Each device Combine with its own coordinates Calculate target device coordinates A local topology map is constructed. Each device uploads its local topology map to the blockchain, triggering a smart contract that aggregates the predictions of all devices for the target device to generate a global topology map.

[0145] By analyzing the differences between the global topology map and the prediction results of each device, attackers can be detected.

[0146] ,

[0147] If a certain device If the threshold is exceeded, the attacker is identified.

[0148] 2.3 Attacker Location

[0149] When each device compares its local topology map with the global topology map, it can not only authenticate the attacker but also pinpoint the attacker's location coordinates. Since the global topology map contains the coordinate information of all devices, once the attacker's identity is determined, their coordinates in the global topology map can be directly retrieved, revealing the attacker's true location.

Claims

1.A blockchain-based lightweight industrial wireless physical layer security authentication method, characterized in that, Security authentication of the physical layer is realized through a multi-node cooperative positioning and authentication strategy, including the following steps: Offline training: Local training stage: collect channel state information of the device, extract channel state information features; train a positioning model according to the extracted channel state information features, and construct a local topology graph based on the positioning model output; Global optimization stage: upload the local topology graph, construct a global topology graph through a blockchain; optimize the positioning model using cooperative error feedback; Online authentication: According to the trained positioning model, the identity of the device is authenticated. 2.The lightweight industrial wireless physical layer security authentication method based on blockchain of claim 1, wherein, The channel state information of the device is collected, specifically as follows: determining a fixed deployment location of the device, coordinates of ; the device collects corresponding channel state information samples through wireless channel communication, denoted as: , wherein, is the channel state information matrix of the device as a receiving end, the device as a transmitting end, the device , and the channel gain matrix between the device , and the expression form is: ; wherein is the sampling time, is the number of multipaths at time is the amplitude attenuation of the th path, is the phase offset of the th path, is the Dirichlet impulse function, is the delay of the signal propagation, is the time delay of the th path; Each device acquires channel state information and records connection relationships with other devices : wherein, denotes that there is a communication connection between device and device , denotes the neighbor set of device . 3.The lightweight industrial wireless physical layer security authentication method based on blockchain of claim 1, wherein, The channel state information features are extracted, specifically as follows: Feature extraction is performed on the collected channel state information, and the amplitude, phase difference, path gain, and angle of arrival are combined as a feature vector between the signal transmitted by the sending device i to the corresponding receiving device j : ； wherein, is the amplitude, representing the signal strength variation of the wireless signal in the propagation process; is the phase difference, representing the spatial characteristics of the wireless signal between different subcarrier frequencies or different antennas; is the path gain, representing the power attenuation caused by the transmission distance of the wireless signal; is the angle of arrival, representing the direction of the wireless signal when it arrives at the receiving antenna array. 4.The lightweight industrial wireless physical layer security authentication method based on blockchain of claim 1, wherein, The angle of arrival is estimated using the MUSIC algorithm, specifically as follows: Channel state information received by the multiple antennas, calculating a channel state information covariance matrix for the multiple antennas , performing eigen decomposition on the covariance matrix , separating it into a signal subspace and a noise subspace ; ； wherein and are the corresponding eigenvalue matrices; Then, the angle-based steering vector is constructed and acts on the noise subspace of eigen-decomposition, and finally the angle of arrival is obtained ; specifically, each candidate angle , the steering vector is constructed: ， wherein, is the antenna spacing, is the signal wavelength, is the number of receive antennas; Constructing a MUSIC spectrum: ， Scan The peak position of the MUSIC spectrum is found, i.e. the angle of arrival where corresponding to the transmitting device and the receiving device respectively. 5.The lightweight industrial wireless physical layer security authentication method based on blockchain of claim 1, wherein, The positioning model is trained according to the extracted channel state information features, specifically as follows: Each device will have multiple feature vectors The feature matrix is composed , and the adjacency matrix As input to the localization model, the localization model is trained, outputting the relative distances between devices , azimuth , and elevation . 6.The lightweight industrial wireless physical layer security authentication method based on blockchain of claim 1, wherein, The local topology graph is constructed based on the positioning model output, specifically as follows: Each device According to its own coordinates And the predicted relative position parameters, including distance , azimuth And elevation , of the target device , the predicted coordinates of the target device are calculated: ， ， ； According to the device A local topology graph is constructed in combination with the connection relationship between devices : ； wherein, represents the predicted coordinates of the jth device. 7.The lightweight industrial wireless physical layer security authentication method based on blockchain of claim 1, wherein, The local topology graph is uploaded to construct a global topology graph through a blockchain, specifically as follows: Each device uploads the constructed local topology graph to the blockchain, and the blockchain receives the local topology graph and triggers the smart contract to read the prediction results of the devices from the local topology graph and construct a global topology graph using a weighted average method : ； wherein, represents a weighted average computing device global coordinates; ， ， ， wherein, is a device a prediction weight for a device , , ， wherein, represents a prediction error, M represents a number of transmitting devices capable of predicting a device i , that is, a number of devices i that establish a transceiving relationship with a device j; represents a predicted coordinate of a target device , represents a real coordinate of a target device . 8.The lightweight industrial wireless physical layer security authentication method based on blockchain of claim 1, wherein, The positioning model is optimized using cooperative error feedback, specifically as follows: Blockchain will create a global topology graph Broadcast to all devices so that each device receives the global coordinates; each device Based on the target device in the global topology diagram global coordinates and predicted coordinates Calculation error: ； Minimizing the error as the goal, define the loss function, and update the positioning model parameters through backpropagation. 9.The lightweight industrial wireless physical layer security authentication method based on blockchain of claim 1, wherein, According to the trained positioning model, the identity of the device is authenticated, specifically as follows: The channel state information of the device is acquired in real time, and channel state information features are extracted; a feature matrix is ​​constructed based on the extracted channel state information features. ,Will Adjacency Matrix As input to the trained model, predict the target device. The relative distance, azimuth, and elevation angles are calculated; a local topology map is constructed based on the output of the positioning model. Upload local topology graph, build global topology graph through blockchain; the blockchain aggregates all local topology graphs uploaded by devices through smart contract to generate a global topology graph , and broadcast it to all devices; each device compares the target device in the global topology graph with the predicted coordinates, and calculates the error: ； If the error exceeds a preset threshold, the device is considered an attacker; otherwise, the device is a legitimate one.

Images