Wireless charging authentication method based on ubiquitous electromagnetic signals

Abstract

The application uses an alternating electromagnetic signal as a fingerprint to completely passively authenticate a wireless charging device, and due to the omnipresence of the electromagnetic signal, more reliable bidirectional authentication in the system can be achieved. The application is very convenient for users, and only needs to be passively collected and detected using a commercial magnetic sensor, and does not depend on a wireless charging authentication method based on omnipresent electromagnetic signals, comprising: collecting an electromagnetic signal; performing privacy protection processing; preprocessing to obtain a signal sample for a certain smart device; performing feature extraction on the signal sample to obtain a series of feature sets constituted by each sample as a device fingerprint; and inputting the device fingerprint into a classifier for training, so that when a device needs to access a wireless charging system, whether to allow access and whether there is an intrusion behavior can be judged through detailed comparison of the fingerprint. Relying on additional manual operation.

Description

Technical Field

[0001] This invention relates to the technical field of authentication, and more particularly to a wireless charging authentication method based on ubiquitous electromagnetic signals. Background Technology

[0002] In emerging IoT scenarios, a large number of battery-powered devices (such as smartphones, electric toothbrushes, and home appliances) face an urgent need for charging anytime, anywhere. Wireless charging, due to its convenience and ease of use, is increasingly being deployed in public infrastructure such as airports, subway stations, cafeterias, and electric vehicle charging stations, experiencing rapid growth. Despite the significant benefits of wireless charging, malicious attackers can launch device spoofing attacks. In such attacks, attackers can impersonate victims, threatening the safety of users and wireless charging systems (e.g., wireless power banks, public charging pads). Regarding threats to users, attackers can impersonate victims during billing, causing financial losses. As for attacks on the entire charging system, attackers can attach adversary coils to malicious devices, disrupting the wireless charging system and causing overheating or even explosions during charging. Therefore, device spoofing attacks pose a significant threat to the deployment of wireless charging systems.

[0003] The primary cause of device spoofing attacks is the lack of device authentication mechanisms in current wireless charging systems. Device fingerprinting is a typical candidate authentication scheme, which creates a fingerprint (i.e., a unique identifier) ​​for each device to defend against spoofing attacks. Existing methods cannot be passively and conveniently integrated into the wireless charging process. Therefore, proposing a user-friendly method that implements device fingerprinting while ensuring normal wireless charging operation is crucial.

[0004] To overcome the challenges of existing solutions in wireless charging, this invention provides a passive device fingerprinting system that utilizes commercially available sensors to collect electromagnetic signals as a unique fingerprint without interrupting the normal charging process. The entire process requires no user intervention. Furthermore, it can serve as an alternative to hardware-based two-factor authentication, compatible with existing authentication schemes. The fundamental idea behind this invention is based on the phenomenon that wireless charging signals can serve as a unique factor for device fingerprinting. More specifically, when a device is wirelessly charging, the variable wireless signal (i.e., magnetic signal) possesses unique characteristics related to the inherent hardware properties of the device. Therefore, by non-contactly collecting and analyzing wireless signals, a unique device fingerprint can be passively constructed without requiring any special operation from the device owner or the device itself.

[0005] At the same time, this invention differs from previous work in that, since the electromagnetic signals of wireless charging are strongly correlated with the device, data privacy protection is required.

[0006] Therefore, this invention should have the following advantages:

[0007] (1) The present invention should passively extract device fingerprints, which is user-friendly and convenient.

[0008] (2) The authentication method of this invention needs to be more effective and robust. In particular, the data transmission process needs to be more secure and reliable.

[0009] (3) This invention should propose more diverse and mixed features to construct fingerprints and optimize experimental results.

[0010] However, developing a convenient, effective, and robust device fingerprint recognition solution based on this phenomenon remains quite challenging. This invention faced the following research challenges during its design:

[0011] (1) How to conveniently collect useful data?

[0012] (2) How to effectively capture the unique and inherent characteristics of the device during charging?

[0013] (3) How to verify its robustness in real-world environments? Summary of the Invention

[0014] To address the security threat posed by device spoofing attacks during wireless charging, and to reduce or eliminate risks and harms, this invention provides a wireless charging authentication method based on electromagnetic signals. First, for Challenge 1, when a new device is placed at a charging station, this invention utilizes an array of four commercially available miniature magnetic sensors to collect alternating magnetic signals. For convenience, these sensors do not require modification of their internal circuitry. Second, for Challenge 2, to effectively capture features, this invention implements preprocessing and noise filtering schemes to improve the quality of the acquired data. A novel feature extraction method is also proposed, using a deep neural network to extract features from the device's electromagnetic signals. Furthermore, to further ensure the security of data throughout the entire process, this invention designs a privacy protection module, using differential privacy to measure security, and enhances system security by adding Laplace noise and designing a novel reliable security aggregation scheme. Finally, for Challenge 3, this invention proposes validating the proposed wireless authentication system on widely adopted real-world commercial wireless charging systems, considering and evaluating various factors (e.g., battery level, location, background applications, training dataset size).

[0015] This invention addresses device spoofing attacks by defining an electromagnetic authentication system for wireless charging security. It collects, analyzes, and processes widely available wireless charging electromagnetic signals, binds and manages devices with fingerprints extracted from these signals, and ultimately verifies malicious behavior by comparing these fingerprints.

[0016] This invention employs oversampling and spectrum analysis strategies to ensure the availability and feasibility of the proposed detection scheme in the context of vehicle networking.

[0017] The technical solution of the present invention is as follows:

[0018] A wireless charging authentication method based on ubiquitous electromagnetic signals, characterized by including:

[0019] S1. Use a sensor array to collect electromagnetic signals during the wireless charging process of smart devices;

[0020] S2. Perform privacy protection processing on the electromagnetic signal to obtain an encrypted electromagnetic signal;

[0021] S3. Preprocess the encrypted electromagnetic signal to obtain a signal sample for a specific smart device;

[0022] S4. Extract features from the signal samples to obtain a series of feature sets composed of each sample, which serve as the device fingerprint;

[0023] S5. Input the device fingerprint into the classifier for training. When a device needs to access the wireless charging system, it can determine whether access is allowed and whether there is any intrusion behavior by comparing the fingerprint in detail.

[0024] Furthermore, the sensor array includes four magnetic field sensors. Let the sampling rate of the i-th magnetic field sensor S be Fs, and the magnetic field signal Mi collected be P sampling points when the duration is T, i.e., P = Fs * T.

[0025] Furthermore, step S2 performs privacy protection processing on the electromagnetic signal to obtain an encrypted electromagnetic signal. The specific steps include:

[0026] S2.1 Add noise conforming to the Laplace distribution: f(x|u,b)=(1 / 2b)*exp((-1 / b)*|xu|)

[0027] Where u and b represent the mean and variance of the selected distribution, respectively, exp is the noise of the exponential distribution and the Laplace distribution, and the model parameter obfuscation based on distributed differential privacy is implemented at the edge nodes.

[0028] S2.2 uses the generated key sk k For data based on the discrete logarithm difficulty assumption Protection is implemented, where i refers to each magnetic sensor node, t refers to the round of data update and aggregation, and the final result is the encrypted secure ciphertext transmitted during the process.

[0029] Where H(t) represents the hash mapping relationship, sk k and n k This refers to the set of key pairs for the k-th magnetic sensor node and the number of connected nodes. This represents the data transmitted between sensors k and j in round t.

[0030] Further, step S3, preprocessing the encrypted electromagnetic signal to obtain a signal sample for a specific smart device, includes the following steps:

[0031] For each valid data collected by the sensor, the pre-emphasis operation is performed as follows:

[0032] E[n]=Mi[n]-∈Mi[n-1], n≥2,

[0033] Where E[n] is the data after the pre-emphasis process, and the pre-emphasis parameter relationship is set to 0.9;

[0034] The signal curve is smoothed using a moving average filter with a filter window size of 10.

[0035] Further, in step S4, feature extraction is performed on the signal samples to obtain a series of feature sets composed of each sample, which serve as the device fingerprint. Specific steps include:

[0036] For the data from the i-th sensor, FE[1:P,i] contains the sampling point P;

[0037] The filter size of the convolutional layer is designed to be 5*5, and the data is reduced in dimensionality by the pooling layer to obtain high-dimensional device fingerprint features.

[0038] Furthermore, in step S5, the device fingerprint is input into a classifier for training. When a device needs to access the wireless charging system, a detailed comparison of the fingerprint can be performed to determine whether access is permitted and whether intrusion occurs. The specific steps are as follows:

[0039] S5.1 Building a Classifier: Collect data from various charging devices, then extract features and label the devices to construct a feature dataset:

[0040] F=[X0[1:k],…,Xj[1:k]],

[0041] For a specific device, F represents the dataset with merged features, and X represents a feature in a certain dimension.

[0042] S5.2 Input the training data into the classifier;

[0043] The classifier generated by S5.3 is used for device verification. After the classifier is obtained through training, for a given device, the classifier outputs the predicted identity. If the predicted label is the one it claims, the device is considered benign and the given device is charged; otherwise, the charging request is rejected.

[0044] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0045] (1) More convenient and faster passive detection to realize passive wireless charging fingerprint construction.

[0046] (2) Utilizing deep neural networks to extract more effective features can be applied to wireless charging scenarios. Enhanced data security is achieved through the design of a privacy protection module.

[0047] (3) Better cost and effect, lower cost can be obtained by using commercial equipment, and lower latency can be obtained for the equipment.

[0048] (4) Unlike the unidirectional fingerprint construction of patent CN 112712046 A, this invention enhances the privacy protection module to protect the security of fingerprint data of wireless charging devices. Instead of using internal hardware signals and bus signals, it achieves device authentication entirely passively through the collection of ubiquitous electromagnetic signals, which is more convenient and faster.

[0049] (5) By utilizing alternating electromagnetic signals as fingerprints, wireless charging devices can be authenticated entirely passively. Furthermore, due to the ubiquity of electromagnetic signals, more reliable two-way authentication within the system can be achieved. This invention is very convenient for users, requiring only passive collection and detection using commercially available magnetic sensors, without relying on additional manual operations. Attached Figure Description

[0050] Figure 1 Flowchart of the wireless charging authentication method based on ubiquitous electromagnetic signals of the present invention Detailed Implementation

[0051] The following provides a detailed description of examples of the present invention: This embodiment is implemented based on the technical solution of the present invention, and provides detailed implementation methods and specific operation processes. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

[0052] A wireless charging authentication system based on electromagnetic signals extracts electromagnetic signals from ubiquitous space, then performs operations such as framing, filtering, and pre-emphasis on the collected electromagnetic signals, and obtains the fingerprint of a specific device by selecting features, and finally verifies it on a classifier.

[0053] Furthermore, it can be carried out in five steps: data acquisition, privacy protection module, data preprocessing, fingerprint construction based on deep neural network, and classification verification.

[0054] In the data acquisition phase, in order to construct a more robust and stable fingerprint, this invention uses a designed magnetic sensor array for data acquisition. For example, by using four sensors distributed around and inside the circular primary charging coil, multi-dimensional information can be acquired more effectively.

[0055] Furthermore, regarding the privacy protection module, for the acquired data (collected in the t-th round of data exchange and on the i-th magnetic sensor), this invention employs differential privacy to measure and enhance security.

[0056] First, by adding an appropriate amount of noise conforming to the Laplace distribution: f(x|u,b)=(1 / 2b)*exp((-1 / b)*|xu|)

[0057] Where u and b represent the mean and variance of the selected distribution, respectively, and exp is the exponential distribution. Appropriate noise can ensure availability while preventing the leakage of user device privacy after theft.

[0058] Subsequently, this invention achieves reliable data transmission and aggregation through a cryptographic scheme. Specifically, it is a secure aggregation for local gradient updates. This is achieved using the generated key sk. k For data based on the discrete logarithm difficulty assumption Protection is implemented, where i refers to each magnetic sensor node, t refers to the round of data update and aggregation, and the encrypted secure ciphertext transmitted during the process can be obtained.

[0059] In the above formula, H(t) represents the hash mapping relationship, and sk k and n k This refers to the set of key pairs for the k-th magnetic sensor node and the number of connected nodes. This represents the data transmitted between sensors k and j in round t. Ultimately, reliable encrypted ciphertext will be used for data transmission and sharing.

[0060] Furthermore, in the data preprocessing stage, this invention enhances and optimizes the acquired signal through preprocessing, smoothing filtering, and other methods, reducing interference and influence of the external environment on the electromagnetic signal. Simultaneously, this invention uses an oversampling technique developed based on a wireless charging system to recover a finer-grained electromagnetic signal.

[0061] Specifically, due to hardware limitations, the sampling rate of commercial magnetic sensors is limited to 100Hz, making it difficult to capture minute signal changes and extract features of the device under test during charging. Therefore, we implement oversampling rate reconstruction (SRR) to reconstruct the signal using more samples from other time periods involved in a segment of the signal frame.

[0062] To further demonstrate the superiority of this system's magnetic sensor-based data acquisition, a corresponding theoretical analysis was conducted. Among the signals acquired by the four deployed magnetic field sensors, the most stable magnetic field signal was selected, and an additional information dimension (the average of the four sensors) was added to prepare for more accurate feature extraction in the future.

[0063] Furthermore, in the fingerprint construction stage based on deep neural networks, this invention designs a novel deep neural network to extract effective device fingerprint features. Specifically, it mainly includes 3 convolutional layers and 3 pooling layers. To enhance the experimental results, this invention uses small-sized filters, designing the filter size of the convolutional layers to be 5*5. Next, the pooling layers are used to perform dimensionality reduction on the data, which can reduce the overall computational load and improve the system's response speed. Assuming that the final output of a traditional neural network is different, this invention inputs the denoised data FE[1:P,i] containing P sampling points into the deep neural network to obtain high-dimensional device fingerprint features.

[0064] To further supplement multi-dimensional features, this invention uses feature extraction libraries, such as LibXtract, to extract supplementary features and improve the acquired feature quality. Finally, this invention merges the above features to obtain the final device fingerprint.

[0065] In the classification and verification phase, this invention selects multiple classifiers from a given pool of charging devices to determine the device's identity. Specifically, this system selects four classifiers (SVM-Linear, SVM-rbf, SVM-poly, and Naive Bayes). The classifiers are trained using the previously collected feature data, and device verification is performed. For a given device, the classifier predicts its identity based on the provided fingerprint output and compares it. If the predicted label matches its claim, this invention considers the device benign and charges the given device. Otherwise, the charging request is rejected, and a device spoofing attack is reported.

[0066] It can be deployed and used in wireless charging systems, including the following steps:

[0067] Step 1: Data acquisition and preprocessing. For wireless charging scenarios, this invention deploys a magnetic field sensor array to acquire ubiquitous signals during the wireless charging process and performs preprocessing to obtain high-quality electromagnetic signals for a specific smart device.

[0068] Step two involves fingerprint construction. This invention utilizes a deep neural network to optimize features. By reducing and merging the features, the final device fingerprint is obtained.

[0069] Step 3, classification verification. This invention selects four classifiers (i.e., Support Vector Machine (SVM)-Linear, SVM-rbf, SVM-poly, and Naive Bayes). Since the extracted fingerprint features are clear and distinct, no neural network was used for related training. Using a simple classifier model can effectively perform classification verification. Finally, the identification of malicious counterfeit devices is achieved.

[0070] The technical solution provided by the present invention will be further described in detail below with reference to the accompanying drawings.

[0071] First, data acquisition and preprocessing are performed. For the array sensor designed in this invention, the data acquisition is performed on the magnetic field sensor Si. When the sampling rate is Fs for the i-th magnetic field sensor, the duration is T and the sampling point is P = Fs * T.

[0072] This invention introduces the concept of secret sharing to further protect the privacy of ubiquitous sensing data through cryptographic means. Considering the overall network situation, a strategy for local gradient updates is proposed, consisting of two parts: key generation and encrypted local gradient updates.

[0073] 1) Key Generation: The secret sharing scheme of this invention is based on protection of the discrete logarithmic difficulty assumption. To generate shared secrets, this scheme employs a trusted third-party entity. First, a cyclic group G of prime number p is determined, with p and g serving as generators. Then, for any two wireless sensing nodes across all magnetic sensors, denoted as c... u and c v We start from Z p Generating pairs of secret sk uv and SK vu , satisfy sk uv +sk vu = 0 mod p. At this point, the secret vector s k For c k Yes: s k ={sk kv :v=0,…,h and v≠k}.

[0074] 2) Encrypted local gradient update: For each iteration of the magnetic sensor node, first calculate... All local gradient updates are within the range [10^-4; 1]. The ciphertext for updating the gradients is obtained cryptographically.

[0075] In the above formula, H(t) represents the hash mapping relationship, and sk k and n k This refers to the set of key pairs for the k-th magnetic sensor node and the number of connected nodes. This represents the data transmitted between sensors k and j in round t. Ultimately, reliable encrypted ciphertext will be used for data transmission and sharing.

[0076] Next, the collected data is preprocessed. First, a pre-emphasis operation is performed. For each valid data point collected by the sensor, the operation is as follows: E[n] = Mi[n] - ∈ Mi[n-1], n ≥ 2, where E[n] is the data after the pre-emphasis process, and the parameter is set to 0.9. This invention uses a moving average filter to smooth the signal curve, employing a five-point sampling method with a corresponding filter window size of 10. The optimized data is then transmitted to the feature extraction module to generate the device's unique fingerprint.

[0077] Network structure: It consists of 3 convolutional layers and 3 pooling layers. In the convolutional layers, small-sized filters (5x5) are used for convolution to obtain multi-dimensional feature data. Based on this, the pooling layers reduce the dimensionality of the data. Simultaneously, normalization is performed on the obtained fingerprint feature data.

[0078] To further supplement multi-dimensional features, this invention uses feature extraction libraries, such as LibXtract, to extract supplementary features and improve the obtained feature quality. Finally, this invention reduces and merges the aforementioned features to obtain the final device fingerprint.

[0079] Ultimately, a high-dimensional device fingerprint can be obtained, and this fingerprint is lightweight, making it suitable for wireless charging tasks.

[0080] After classifier pre-training and registration, when faced with device spoofing attacks, the classifier can compare the device's access fingerprint with the fingerprints in the database and output 1 or 0. 1 represents a benign sample device itself, which is allowed to access; 0 represents a malicious device input, which will be denied access.

[0081] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A wireless charging authentication method based on ubiquitous electromagnetic signals, characterized in that: include: S1. Use a sensor array to collect electromagnetic signals during the wireless charging process of smart devices; S2. Perform privacy protection processing on the electromagnetic signal to obtain an encrypted electromagnetic signal; S3. Preprocess the encrypted electromagnetic signal to obtain a signal sample for a specific smart device; S4. Extract features from the signal samples to obtain a series of feature sets composed of each sample, which serve as the device fingerprint; S5. Input the device fingerprint into the classifier for training. When a device needs to access the wireless charging system, determine whether access is allowed and whether there is any intrusion behavior by comparing the fingerprint in detail. Step S2 performs privacy protection processing on the electromagnetic signal to obtain an encrypted electromagnetic signal. Specific steps include: S2.1 Add noise conforming to the Laplace distribution: f(x|u,b)=(1 / 2b)*exp((-1 / b)*|xu|) Where u and b represent the mean and variance of the selected distribution, respectively, exp is the noise of the exponential distribution and the Laplace distribution, and the model parameter obfuscation based on distributed differential privacy is implemented at the edge nodes. S2.2 using the generated key For data based on the discrete logarithm difficulty assumption Protection is implemented, where i refers to each magnetic sensor node, t refers to the round of data update and aggregation, and the final result is the encrypted secure ciphertext transmitted during the process. , Where H(t) represents the hash mapping relationship, and This refers to the set of key pairs for the k-th magnetic sensor node and the number of connected nodes. This represents the data transmitted between sensors k and j in round t.

2. The wireless charging authentication method based on ubiquitous electromagnetic signals according to claim 1, characterized in that, The sensor array includes four magnetic field sensors. Let the sampling rate of the i-th magnetic field sensor S be Fs, and the magnetic field signal Mi collected be P sampling points when the duration is T, i.e., P = Fs * T.

3. The wireless charging authentication method based on ubiquitous electromagnetic signals according to claim 1, characterized in that, Step S3. Preprocessing the encrypted electromagnetic signal to obtain a signal sample for a specific smart device, specifically including the following steps: For each valid data collected by the sensor, the pre-emphasis operation is performed as follows: E[n] = Mi[n]−ϵMi[n−1], n≥2, Where E[n] is the data after the pre-emphasis process, Mi is the acquired magnetic field signal, and the pre-emphasis parameter relationship is set to 0.9; The signal curve is smoothed using a moving average filter with a filter window size of 10.

4. The wireless charging authentication method based on ubiquitous electromagnetic signals according to claim 1, characterized in that, Step S4. Extracting features from the signal samples to obtain a series of feature sets for each sample, which serve as the device fingerprint. Specific steps include: For the data from the i-th sensor, FE[1:P,i] contains the sampling point P; The filter size of the convolutional layer is designed to be 5*5, and the data is reduced in dimensionality by the pooling layer to obtain high-dimensional device fingerprint features.

5. The wireless charging authentication method based on ubiquitous electromagnetic signals according to claim 1, characterized in that: Step S5. Input the device fingerprint into the classifier for training. When a device needs to access the wireless charging system, determine whether access is allowed and whether there is any intrusion by comparing the fingerprint in detail. The specific steps are as follows: S5.1 Building a Classifier: Collect data from various charging devices, then extract features and label the devices to construct a feature dataset: F = [X0[1: k],…,Xj[1: k]], For a specific device, F represents the dataset with merged features, and X represents a feature in a certain dimension. S5.2 Input the training data into the classifier; The classifier generated by S5.3 is used for device verification. After the classifier is obtained through training, for a given device, the classifier outputs the predicted identity. If the predicted label is the one it claims, the device is considered benign and the given device is charged; otherwise, the charging request is rejected.

Images