Identity authentication and anomaly detection system for TMR sensor array

Abstract

The application discloses an identity authentication and anomaly detection system for a TMR sensor array, comprising a control module, N signal conditioning and acquisition modules, a feature extraction module, a PUF feature processing and self-checking module, a behavior anomaly detection module, and a decision fusion and communication module. The system realizes time division multiplexing of the identity authentication function and the anomaly detection function on the same hardware link by uniformly scheduling the N signal conditioning and acquisition modules and controlling the working state of the identity authentication and anomaly detection system composed of the control module, the feature extraction module, the PUF feature processing and self-checking module, the behavior anomaly detection module, and the decision fusion and communication module. The system has the advantages of small overall occupied area, low power consumption, and synergistic fusion of physical feature recognition and anomaly detection, and can realize continuous verification from "device identity trust" to "running state trust", and has strong defense capability against composite attacks.

Description

Technical Field

[0001] This invention relates to an identity authentication and anomaly detection system, and more particularly to an identity authentication and anomaly detection system for a TMR sensor array. Background Technology

[0002] Tunnel magnetoresistance (TMR) sensors have become a core technology for precision magnetic field sensing due to their high sensitivity and low power consumption. In many critical applications, a single TMR sensor cannot meet the requirements for spatial analysis of complex magnetic field environments; therefore, sensor arrays composed of multiple TMR sensors have become an inevitable technological development. These TMR sensor arrays typically arrange multiple independent sensors in a two-dimensional matrix and acquire data synchronously through multiple channels to accurately depict high-dimensional information such as the spatial distribution, gradient, or vector of the magnetic field. Each TMR sensor outputs a pair of differential signals under the action of an excitation voltage, and an array of N TMR sensors outputs N pairs of differential signals. For example, in high-end CNC machine tools and multi-joint robots, TMR sensor arrays are integrated into the magnetic encoder of a servo motor for high-precision analysis of rotation angles and positions; in smart grids, multiple TMR sensor arrays construct magnetic field distribution models to accurately calculate load currents and detect faults; and in high-speed railway wheel and axle flaw detection systems, TMR sensor arrays are used to capture leakage magnetic field signals generated by tiny cracks in the wheelsets when trains pass.

[0003] In these security-critical application scenarios, TMR sensor arrays are deeply embedded in networked control systems, and the reliability of their data directly affects the security of the entire control system. Therefore, two core security requirements arise: First, to ensure the reliability of the data source, the physical characteristics of the TMR sensor array must be identified and verified to prevent nodes from being forged, replaced, or hijacked. Second, to ensure the accuracy and reliability of magnetic field sensing results, the output signal of the TMR sensor array must be monitored in real time to identify abnormal changes in the external magnetic field environment.

[0004] Currently, researchers have designed physical feature recognition systems and anomaly detection systems for TMR sensor arrays. Physical feature recognition systems primarily utilize the non-repeatable microscopic physical differences formed during the manufacturing process of TMR sensors to construct a Physically Unclonable Function (PUF). These random features are then extracted using dedicated high-precision interface circuits, generating a unique "digital fingerprint" for each TMR sensor, thus achieving unique identification and secure access authentication. However, the feature recognition process in these systems is typically performed only once during the initial device connection or power-on authentication phase. During operation, there is a lack of periodic feature verification mechanisms, making it impossible to promptly detect physical layer anomalies, drift, or replacement risks after long-term use or attacks. Anomaly detection systems primarily monitor the operation of the TMR sensor array. They utilize statistical analysis, spectral feature extraction, or deep learning-based models to model and analyze the real-time output signal, operating status, and external environmental interference of the TMR sensor array, identifying potential performance drift, component failures, or malicious attacks.

[0005] The aforementioned physical feature recognition system and anomaly detection system are designed independently, resulting in a large overall footprint when deployed within the TMR sensor array. Furthermore, the authentication system typically relies on high-precision analog interface circuits, leading to high computational load and a large number of model parameters, resulting in high power consumption. The anomaly detection system, on the other hand, relies on high-performance processors or cloud-based models, which either consume significant power or experience high latency, resulting in poor real-time performance. Additionally, the physical feature recognition system and anomaly detection system belong to different security domains, with separate data paths and decision-making logic. Attackers could exploit this logical isolation to forge TMR sensor responses or intermediate layer data, rendering the TMR sensor array inadequate against combined attacks. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a physical feature recognition and anomaly detection system for TMR sensor arrays that integrates physical feature recognition and anomaly detection functions. When deployed in a TMR sensor array, this system has a small overall footprint and low power consumption. Furthermore, the synergistic integration of physical feature recognition and anomaly detection enables continuous verification from "trustworthy device identity" to "trustworthy operating status," and provides strong defense against complex attacks.

[0007] The technical solution adopted by this invention to solve the above-mentioned technical problems is as follows: an identity authentication and anomaly detection system for a TMR sensor array, wherein the TMR sensor array includes multiple TMR sensors, and the number of TMR sensors in the TMR sensor array is denoted as N. The identity authentication and anomaly detection system includes a control module, N signal conditioning and acquisition modules, a feature extraction module, a PUF feature processing and self-testing module, a behavior anomaly detection module, and a decision fusion and communication module; the N signal conditioning and acquisition modules are connected one-to-one with the N TMR sensors. The control module provides excitation signals to each TMR sensor and coordinates the timing of N signal conditioning and acquisition modules, feature extraction module, PUF feature processing and self-testing module, behavior anomaly detection module, and decision fusion and communication module, enabling the identity authentication and anomaly detection system to perform identity authentication and anomaly detection functions. The identity authentication and anomaly detection system has two operating modes: normal operation mode and self-testing mode. The default operating mode of the identity authentication and anomaly detection system is normal operation mode. The control module periodically controls the identity authentication and anomaly detection system to enter self-testing mode at preset time intervals T2, or controls the system to enter self-testing mode upon receiving an external forced self-testing command. When the identity authentication and anomaly detection system is in normal operation mode and has not entered self-testing mode, the control module maintains a preset constant excitation voltage applied to each TMR sensor, causing each TMR sensor to maintain its corresponding differential output. The system uses analog signals to control the feature extraction module and the PUF feature processing and self-test module to be in standby mode, disable signal reception, and control N signal conditioning and acquisition modules, behavior anomaly detection module, and decision fusion and communication module to perform monitoring work periodically according to a preset normal monitoring cycle T1, generating an operating status signal output representing the overall safety status of the sensor array in each cycle. When the identity authentication and anomaly detection system is in normal operating mode and is switched to self-test mode, the control module controls the feature extraction module and the PUF feature processing and self-test module to enter working state, enable signal reception, and output a preset constant excitation voltage to each TMR sensor in each excitation cycle according to a preset excitation cycle, so that each TMR sensor outputs a corresponding differential analog signal until the end of the Kth excitation cycle; where K is a positive integer, ranging from 32 to 64, and the amplitude of the constant excitation voltage output in each of the K excitation cycles is different, ranging from 0 to 3.3V; In each excitation cycle, the control module controls N signal conditioning and acquisition modules, behavior anomaly detection module, and decision fusion and communication module to perform L self-check monitoring cycles, where L is a positive integer ranging from 5 to 20, and the self-check monitoring cycle is shorter than the normal monitoring cycle; and in each excitation cycle, the control module controls the feature extraction module and the PUF feature processing and self-check module to perform an identity information acquisition operation once, until the identity information acquisition operation of the Kth excitation cycle is completed, then the control module controls the PUF feature processing and self-check module to perform an identity authentication operation once, and finally the control module controls the decision fusion and communication module to generate an operating status signal characterizing the overall safety status of the sensor array.

[0008] Compared with existing technologies, the advantages of this invention lie in its use of a control module, N signal conditioning and acquisition modules, a feature extraction module, a PUF feature processing and self-testing module, a behavior anomaly detection module, and a decision fusion and communication module to construct an identity authentication and anomaly detection system. The N signal conditioning and acquisition modules, the behavior anomaly detection module, and the decision fusion and communication module implement anomaly detection functionality, while the N signal conditioning and acquisition modules, the feature extraction module, and the PUF feature processing and self-testing module implement identity authentication functionality. Through unified scheduling of the N signal conditioning and acquisition modules, the control module controls the working status of the feature extraction module, the PUF feature processing and self-testing module, the behavior anomaly detection module, and the decision fusion and communication module, thus achieving time-division multiplexing of identity authentication and anomaly detection functions on the same hardware link. This eliminates the need for a separate high-precision analog acquisition circuit for PUF feature extraction, thereby reducing hardware resource consumption and board area. The high-power feature extraction module is disabled in normal operation mode and only enabled in self-test mode, effectively reducing overall system power consumption. Furthermore, this invention, through precise timing control of the control module, enables... During the PUF self-test with K-group excitation, the behavior anomaly detection module remains operational, achieving parallel operation of self-test and monitoring and avoiding monitoring gaps. In terms of physical feature recognition, this invention utilizes multi-amplitude excitation to extract the nonlinear response features of the TMR sensor and suppresses temperature drift and common-mode interference through differential combination, forming a highly stable, high-entropy physical fingerprint to improve the reliability of identity authentication. Regarding anomaly detection, this invention employs a parallel feature processing structure in the time and frequency domains, combined with a conditional variational autoencoder for feature reconstruction, enabling the detection of multiple types of anomalies and the identification of unknown anomalies. Furthermore, by fusing PUF identity authentication results with anomaly detection results, this invention establishes a unified security decision mechanism, simultaneously assessing the physical integrity and operational status of the device, improving the system's security protection capabilities against multiple types of composite threats. Therefore, this invention integrates physical feature recognition and anomaly detection functions. When deployed in a TMR sensor array, it occupies a small area and consumes less power. The synergistic fusion of physical feature recognition and anomaly detection enables continuous verification from "trustworthy device identity" to "trustworthy operational status," providing strong defense against composite attacks.

[0009] Furthermore, the monitoring process is as follows: the control module first controls each signal conditioning and acquisition module to perform a sampling and conditioning operation once, and then controls the behavior anomaly detection module and the decision fusion and communication module to perform an anomaly monitoring operation once; wherein, each signal conditioning and acquisition module performs a sampling and conditioning operation once specifically: first, it samples the differential analog signal output by its corresponding TMR sensor once, then performs differential amplification and filtering on the differential analog signal to obtain one analog signal, then converts the analog signal into a digital signal, and outputs it to the behavior anomaly detection module and the feature extraction module respectively; the behavior anomaly detection module and the decision fusion and communication module... The anomaly detection module performs an anomaly monitoring operation as follows: The behavior anomaly detection module performs time-domain transformation and frequency-domain transformation on the N digital signals output from the N signal conditioning and acquisition modules, generating N time-domain feature vectors and N frequency-domain feature vectors. Then, it performs feature fusion and reconstruction operations on the N time-domain signal features and N frequency-domain signal features to obtain a composite reconstruction error value. The composite reconstruction error value is then compared with a preset threshold to generate an anomaly diagnosis result signal characterizing the behavior state of the TMR sensor array and output it to the decision fusion and communication module. The decision fusion and communication module generates a corresponding operating status signal output based on the anomaly diagnosis result signal.

[0010] Furthermore, the identity information acquisition process specifically involves: the feature extraction module receiving each digital signal output by each signal conditioning and acquisition module during the current excitation cycle; and after receiving L digital signals output by each signal conditioning and acquisition module, arranging these L digital signals in chronological order to form a digital signal sequence of length L as the original response sequence, resulting in N original response sequences. Then, point-to-point difference operations are performed on any two original response sequences to obtain the difference response sequence between each pair of original response sequences. At this point, a total of N original response sequences are obtained. We obtain N differential response sequences of length L; then, we calculate the mean of each original response sequence as a channel mean feature, resulting in N channel mean features. We then extract five differential statistical features from each differential response sequence. Finally, we concatenate the N channel mean features with the five differential statistical features from all differential response sequences in a preset order to generate a sequence of dimension N+. A set of feature vectors of size ×5 is output to the PUF feature processing and self-testing module; the PUF feature processing and self-testing module performs feature mapping and encoding processing on the received set of feature vectors to generate corresponding fingerprint response data; the identity authentication process is as follows: the PUF feature processing and self-testing module first aggregates the generated K fingerprint response data to obtain a device fingerprint sequence, then compares the device fingerprint sequence with a pre-stored reference fingerprint sequence and calculates the difference between the two; when the difference is less than or equal to a preset self-testing threshold, a self-testing result signal characterizing the stability of the physical characteristics of the TMR sensor array is generated; when the difference is greater than the preset self-testing threshold, a self-testing result signal characterizing the drift of the physical characteristics of the TMR sensor array is generated; and the self-testing result signal is output to the decision fusion and communication module.

[0011] Furthermore, the status update process specifically involves: the decision fusion and communication module updating the identity authentication status parameters of the TMR sensor array based on the self-test result signal it currently receives; and logically fusing the abnormal diagnosis result signal output by the behavior anomaly detection module at the current moment with the updated identity authentication status parameters to generate an operating status signal characterizing the overall security status of the sensor array, and uploading it to the host computer or cloud monitoring platform.

[0012] Furthermore, each signal conditioning and acquisition module includes a differential amplifier circuit, a filter circuit, and an analog-to-digital converter circuit. The differential amplifier circuit has a positive input terminal, a negative input terminal, and an output terminal; the filter circuit has an input terminal and an output terminal; and the analog-to-digital converter circuit has an analog input terminal, a control input terminal, and a digital output terminal. The positive and negative input terminals of the differential amplifier circuit are used to acquire and input the differential analog signal output by the TMR sensor, and the output terminal is connected to the input terminal of the filter circuit. The output terminal of the filter circuit is connected to the analog input terminal of the analog-to-digital converter circuit. The control input terminal of the analog-to-digital converter circuit is connected to the control module, and the digital output terminal is connected to the feature extraction module and the behavior anomaly detection module, respectively.

[0013] Furthermore, the feature extraction module includes a data buffer unit, a differential calculation unit, and a feature vector generation unit. The data buffer unit has N input terminals, an enable signal input terminal, and an output terminal. The differential calculation unit has an input terminal and an output terminal, and the feature vector generation unit has a first input terminal, a second input terminal, and an output terminal. The N input terminals of the data buffer unit are respectively connected to the output terminals of N signal conditioning and acquisition modules, the enable signal input terminal is connected to the control module, and the output terminal is respectively connected to the input terminal of the differential calculation unit and the second input terminal of the feature vector generation unit. The output terminal of the differential calculation unit is connected to the first input terminal of the feature vector generation unit. The enable signal input terminal of the data buffer unit receives a self-test enable signal from the control module.

[0014] Furthermore, the PUF feature processing and self-testing module includes a deep autoencoder and a fingerprint generation unit; the deep autoencoder includes an input mapping layer and an encoding layer; the input mapping layer has an input terminal, an enable signal input terminal, and an output terminal, and the encoding layer has an input terminal and an output terminal; the fingerprint generation unit has an input terminal and an output terminal; the input terminal of the input mapping layer is connected to the output terminal of the feature extraction module, its enable signal input terminal is connected to the control module, and its output terminal is connected to the input terminal of the encoding layer; the output terminal of the encoding layer serves as the output terminal of the deep autoencoder and is connected to the input terminal of the fingerprint generation unit; the output terminal of the fingerprint generation unit is connected to the decision fusion and communication module; the enable signal input terminal of the input mapping layer receives a self-test enable signal from the control module.

[0015] Furthermore, the behavior anomaly detection module includes a multi-domain feature extraction unit, a conditional variational autoencoder, and an anomaly determination unit; the conditional variational autoencoder includes a time-domain coding sub-network, a frequency-domain coding sub-network, a conditional fusion layer, a latent space mapping layer, and a decoding and reconstruction layer; the multi-domain feature extraction unit has N input terminals, a first output terminal, and a second output terminal; the time-domain coding sub-network has input terminals and output terminals; the frequency-domain coding sub-network has input terminals and output terminals; the conditional fusion layer has a first input terminal, a second input terminal, a tag signal input terminal, and an output terminal; the latent space mapping layer has input terminals and an output terminal; the decoding and reconstruction layer has input terminals and an output terminal; the anomaly determination unit has a first input terminal, a second input terminal, a third input terminal, and an output terminal; the N input terminals of the multi-domain feature extraction unit are respectively connected to the digital output terminals of N signal conditioning and acquisition modules; The first output of the multi-domain feature extraction unit is connected to the input of the temporal coding sub-network and the first input of the anomaly detection unit, respectively; the second output of the multi-domain feature extraction unit is connected to the input of the frequency coding sub-network and the second input of the anomaly detection unit, respectively; the output of the temporal coding sub-network is connected to the first input of the conditional fusion layer; the output of the frequency coding sub-network is connected to the second input of the conditional fusion layer; the label signal input of the conditional fusion layer is connected to the control module for accessing the running status label, and its output is connected to the input of the latent spatial mapping layer; the output of the latent spatial mapping layer is connected to the input of the decoding and reconstruction layer; the output of the decoding and reconstruction layer is connected to the third input of the anomaly detection unit; and the output of the anomaly detection unit is connected to the decision fusion and communication module.

[0016] Furthermore, the decision fusion and communication module includes a decision fusion unit and a communication circuit; the decision fusion unit has a first input terminal, a second input terminal, and an output terminal; the communication circuit has a data input terminal, a data output terminal, an instruction input terminal, and an instruction output terminal; the first input terminal of the decision fusion unit is connected to the output terminal of the fingerprint generation unit, the second input terminal is connected to the output terminal of the anomaly determination unit, and its output terminal is connected to the data input terminal of the communication circuit; the instruction output terminal of the communication circuit is connected to the control module.

[0017] Furthermore, the control module includes a microcontroller and a digital-to-analog converter circuit; the microcontroller has an excitation control terminal, a sampling control terminal, an enable signal output terminal, a tag signal output terminal, and an instruction input terminal; the excitation control terminal of the microcontroller is connected to the input terminal of the digital-to-analog converter circuit, and the output terminal of the digital-to-analog converter circuit is connected to the power input terminal of the TMR sensor array; the sampling control terminal of the microcontroller is connected to the control input terminals of the analog-to-digital converter circuits in the N signal conditioning and acquisition modules; the enable signal output terminal of the microcontroller is connected to the enable signal input terminal of the data buffer unit and the enable signal input terminal of the input mapping layer, respectively; the tag signal output terminal of the microcontroller is connected to the tag signal input terminal of the conditional fusion layer; and the instruction input terminal of the microcontroller is connected to the instruction output terminal of the communication circuit. Attached Figure Description

[0018] Figure 1 This is a structural diagram of the identity authentication and anomaly detection system for a TMR sensor array according to the present invention; Figure 2 This is a structural diagram of the signal conditioning and acquisition module for the identity authentication and anomaly detection system of the TMR sensor array of the present invention; Figure 3 This is a structural diagram of the feature extraction module for the TMR sensor array authentication and anomaly detection system of the present invention; Figure 4 This is a structural diagram of the PUF feature processing and self-testing module for the identity authentication and anomaly detection system of TMR sensor array of the present invention; Figure 5 This is a structural diagram of the behavior anomaly detection module for the identity authentication and anomaly detection system of the TMR sensor array of the present invention; Figure 6 This is a structural diagram of the decision fusion and communication module for the identity authentication and anomaly detection system of the TMR sensor array of the present invention; Figure 7 This is a structural diagram of the control module for the TMR sensor array authentication and anomaly detection system of the present invention; Figure 8 This is a heatmap of the inter-group Hamming distance for the identity authentication and anomaly detection system for a TMR sensor array according to the present invention. Figure 9 This is a diagram showing the inter-group Hamming distance distribution of the identity authentication and anomaly detection system for TMR sensor arrays according to the present invention. Figure 10 This is a PCA distribution diagram of the temporal features of the identity authentication and anomaly detection system for TMR sensor arrays according to the present invention; Figure 11This is a frequency domain feature PCA distribution diagram of the identity authentication and anomaly detection system for TMR sensor arrays according to the present invention. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0020] Example 1: As Figure 1As shown, an authentication and anomaly detection system for a TMR sensor array is disclosed. The TMR sensor array includes multiple TMR sensors, denoted as N. The system comprises a control module, N signal conditioning and acquisition modules, a feature extraction module, a PUF feature processing and self-testing module, a behavior anomaly detection module, and a decision fusion and communication module. Each of the N signal conditioning and acquisition modules is connected to one of the N TMR sensors. The control module provides excitation signals to each TMR sensor and coordinates the timing of N signal conditioning and acquisition modules, feature extraction module, PUF feature processing and self-testing module, behavior anomaly detection module, and decision fusion and communication module, enabling the identity authentication and anomaly detection system to perform identity authentication and anomaly detection functions. The identity authentication and anomaly detection system has two operating modes: normal operation mode and self-testing mode. The default operating mode is normal operation mode. The control module periodically controls the identity authentication and anomaly detection system to enter self-testing mode at preset time intervals T2, or controls it to enter self-testing mode upon receiving an external forced self-testing command. When the identity authentication and anomaly detection system is in normal operation mode and not in self-testing mode, the control module maintains a preset constant excitation voltage applied to each TMR sensor, ensuring that each TMR sensor outputs the corresponding differential analog signal. The control module keeps the feature extraction module and PUF feature processing and self-test module in standby mode, disables signal reception, and controls N signal conditioning and acquisition modules, behavior anomaly detection module, and decision fusion and communication module to perform monitoring work periodically according to the preset normal monitoring cycle T1, generating an operating status signal output characterizing the overall safety status of the sensor array in each cycle. When the identity authentication and anomaly detection system is in normal operating mode and is switched to self-test mode, the control module controls the feature extraction module and PUF feature processing and self-test module to enter working state, enables signal reception, and outputs a preset constant excitation voltage to each TMR sensor in each excitation cycle according to the preset excitation cycle, so that each TMR sensor outputs a corresponding differential analog signal until the end of the Kth excitation cycle; where K is a positive integer, ranging from 32 to 64, and the amplitude of the constant excitation voltage output in each of the K excitation cycles is different, ranging from 0 to 3.3V; In each excitation cycle, the control module controls N signal conditioning and acquisition modules, behavior anomaly detection module, and decision fusion and communication module to perform L self-test monitoring cycles, where L is a positive integer ranging from 5 to 20, and the self-test monitoring cycle is less than the normal monitoring cycle; In each excitation cycle, the control module controls the feature extraction module and PUF feature processing and self-test module to perform an identity information acquisition operation once, until the identity information acquisition operation of the Kth excitation cycle is completed. Then, the control module controls the PUF feature processing and self-test module to perform an identity authentication operation once. Finally, the control module controls the decision fusion and communication module to generate an operating status signal characterizing the overall safety status of the sensor array.

[0021] In this embodiment, the monitoring process is as follows: the control module first controls each signal conditioning and acquisition module to perform a sampling and conditioning operation once, and then controls the behavior anomaly detection module and the decision fusion and communication module to perform an anomaly detection operation once. Specifically, each signal conditioning and acquisition module performs a sampling and conditioning operation once by: first sampling the differential analog signal output by its corresponding TMR sensor; then performing differential amplification and filtering on the differential analog signal to obtain one analog signal; then converting the analog signal into a digital signal and outputting it to the behavior anomaly detection module and the feature extraction module respectively; the behavior anomaly detection module and the decision fusion and communication module... The specific steps of an anomaly detection operation are as follows: The behavior anomaly detection module performs time-domain transformation and frequency-domain transformation on the N digital signals output from the N signal conditioning and acquisition modules, generating N time-domain feature vectors and N frequency-domain feature vectors. Then, it performs feature fusion and reconstruction operations on the N time-domain signal features and N frequency-domain signal features to obtain a composite reconstruction error value. The composite reconstruction error value is then compared with a preset threshold to generate an anomaly diagnosis result signal characterizing the behavior state of the TMR sensor array and output it to the decision fusion and communication module. The decision fusion and communication module generates a corresponding operating status signal output based on the anomaly diagnosis result signal.

[0022] In this embodiment, the identity information acquisition process is as follows: The feature extraction module receives each digital signal output by the signal conditioning and acquisition module for each current excitation cycle. After receiving L digital signals output by each signal conditioning and acquisition module, the L digital signals are arranged in chronological order to form a digital signal sequence of length L as the original response sequence, resulting in N original response sequences. Then, point-to-point difference operations are performed on any two original response sequences to obtain the difference response sequence between each pair of original response sequences. At this point, a total of N original response sequences are obtained. We obtain N differential response sequences of length L; then, we calculate the mean of each original response sequence as a channel mean feature, resulting in N channel mean features. We then extract five differential statistical features from each differential response sequence. Finally, we concatenate the N channel mean features with the five differential statistical features from all differential response sequences in a preset order to generate a sequence of dimension N+. A set of feature vectors of ×5 is output to the PUF feature processing and self-testing module; the PUF feature processing and self-testing module performs feature mapping and encoding processing on the received set of feature vectors to generate corresponding fingerprint response data; the identity authentication process is as follows: the PUF feature processing and self-testing module first aggregates the generated K fingerprint response data to obtain the device fingerprint sequence, and then compares the device fingerprint sequence with the pre-stored reference fingerprint sequence to calculate the difference between the two; when the difference is less than or equal to the preset self-testing threshold, a self-testing result signal representing the stability of the physical characteristics of the TMR sensor array is generated; when the difference is greater than the preset self-testing threshold, a self-testing result signal representing the drift of the physical characteristics of the TMR sensor array is generated; and the self-testing result signal is output to the decision fusion and communication module.

[0023] In this embodiment, the status update process is as follows: the decision fusion and communication module updates the identity authentication status parameters of the TMR sensor array based on the self-test result signal it currently receives; and logically fuses the abnormal diagnosis result signal output by the behavior anomaly detection module at the current moment with the updated identity authentication status parameters to generate an operating status signal that characterizes the overall safety status of the sensor array, and uploads it to the host computer or cloud monitoring platform.

[0024] In this embodiment, N signal conditioning and acquisition modules, a behavior anomaly detection module, and a decision fusion and communication module implement the anomaly detection function, while N signal conditioning and acquisition modules, a feature extraction module, and a PUF feature processing and self-testing module implement the identity authentication function. Through unified scheduling of the N signal conditioning and acquisition modules, the control module controls the working status of the identity authentication and anomaly detection system comprised of the feature extraction module, PUF feature processing and self-testing module, behavior anomaly detection module, and decision fusion and communication module. This achieves time-division multiplexing of the identity authentication and anomaly detection functions on the same hardware link, eliminating the need for a separate high-precision analog acquisition circuit for PUF feature extraction, thereby reducing hardware resource consumption and board area. The high-power feature extraction module is disabled in normal operation mode and enabled only in self-test mode, effectively reducing the overall system power consumption. Meanwhile, this invention, through precise timing control of the control module, ensures that the behavior anomaly detection module remains operational during the PUF self-test process involving K sets of excitations, achieving parallel operation of self-testing and monitoring and avoiding monitoring gaps. In terms of physical feature recognition, this invention utilizes multi-amplitude excitation to extract the nonlinear response features of the TMR sensor and suppresses temperature drift and common-mode interference through differential combination, forming a highly stable, high-entropy physical fingerprint to improve the reliability of identity authentication. Regarding anomaly detection, this invention employs a parallel feature processing structure in the time and frequency domains, combined with a conditional variational autoencoder for feature reconstruction, enabling the detection of multiple types of anomalies and the ability to identify unknown anomalies. Furthermore, by fusing PUF identity authentication results with anomaly detection results, this invention establishes a unified security decision mechanism, simultaneously assessing the physical integrity and operational status of the device, thereby improving the system's security protection capabilities against multiple types of complex threats.

[0025] Example 2: This example is basically the same as Example 1, except that: Figure 2 As shown in this embodiment, each signal conditioning and acquisition module includes a differential amplifier circuit, a filter circuit, and an analog-to-digital converter circuit. The differential amplifier circuit has a positive input terminal, a negative input terminal, and an output terminal; the filter circuit has an input terminal and an output terminal; and the analog-to-digital converter circuit has an analog input terminal, a control input terminal, and a digital output terminal. The positive and negative input terminals of the differential amplifier circuit are used to acquire and input the differential analog signal output by the TMR sensor, and the output terminal is connected to the input terminal of the filter circuit. The output terminal of the filter circuit is connected to the analog input terminal of the analog-to-digital converter circuit. The control input terminal of the analog-to-digital converter circuit is connected to the control module, and the digital output terminal is connected to the feature extraction module and the behavior anomaly detection module, respectively.

[0026] In this embodiment, the differential amplifier circuit is used to differentially amplify the input differential analog signal to obtain a differential amplified signal output to the filter circuit; the filter circuit is used to perform low-pass filtering on the differential amplified signal to obtain an analog signal output to the analog-to-digital converter circuit; the analog-to-digital converter circuit is used, under the control of the control module, to sample and digitize the analog signal output by the filter circuit to generate a corresponding digital signal output through its digital output terminal.

[0027] Example 3: This example is basically the same as Example 2, except that: Figure 3 As shown, in this embodiment, the feature extraction module includes a data buffer unit, a differential calculation unit, and a feature vector generation unit. The data buffer unit has N input terminals, an enable signal input terminal, and an output terminal. The differential calculation unit has an input terminal and an output terminal, and the feature vector generation unit has a first input terminal, a second input terminal, and an output terminal. The N input terminals of the data buffer unit are respectively connected to the output terminals of N signal conditioning and acquisition modules, the enable signal input terminal is connected to the control module, and the output terminal is respectively connected to the input terminal of the differential calculation unit and the second input terminal of the feature vector generation unit. The output terminal of the differential calculation unit is connected to the first input terminal of the feature vector generation unit. The enable signal input terminal of the data buffer unit receives a self-test enable signal from the control module.

[0028] In this embodiment, the data buffer unit receives a self-test enable signal from the control module via its enable signal input terminal. This self-test enable signal is a periodic signal, and in each cycle, it has two states: valid and invalid. When the self-test enable signal is valid, the feature extraction module enters the working state; when the self-test enable signal is invalid, the feature extraction module enters the standby state. During the valid period of each cycle of the self-test enable signal, the data buffer unit receives L digital signals output from each signal conditioning and acquisition module in each excitation cycle, and arranges the L digital signals of each excitation cycle in sequential order to form an original response sequence. After obtaining N original response sequences for each excitation cycle, these N original response sequences are output to the differential calculation unit and the feature vector generation unit, respectively. The differential calculation unit performs element-wise subtraction on any two of the N original response sequences in each excitation cycle to generate... A series of differential response sequences of length L are generated and output to the feature vector generation unit. The feature vector generation unit calculates the arithmetic mean of each original response sequence in each excitation cycle as the channel mean feature, resulting in N channel mean features. For each differential response sequence in each excitation cycle, five differential statistical features are extracted: differential mean, differential standard deviation, differential median absolute deviation, differential skewness, and differential kurtosis. Then, the N channel mean features of each excitation cycle are combined with... The differential statistical features of the differential response sequences are concatenated in a preset order to generate a sequence with dimension N+. A set of feature vectors of ×5 is generated and output to the PUF feature processing and self-testing module.

[0029] Example 4: This example is basically the same as Example 3, except that: Figure 4 As shown, in this embodiment, the PUF feature processing and self-testing module includes a deep autoencoder and a fingerprint generation unit; the deep autoencoder includes an input mapping layer and an encoding layer; the input mapping layer has an input terminal, an enable signal input terminal, and an output terminal, and the encoding layer has an input terminal and an output terminal; the fingerprint generation unit has an input terminal and an output terminal; the input terminal of the input mapping layer is connected to the output terminal of the feature extraction module, its enable signal input terminal is connected to the control module, and its output terminal is connected to the input terminal of the encoding layer; the output terminal of the encoding layer serves as the output terminal of the deep autoencoder and is connected to the input terminal of the fingerprint generation unit; the output terminal of the fingerprint generation unit is connected to the decision fusion and communication module; the enable signal input terminal of the input mapping layer receives a self-test enable signal from the control module.

[0030] In this embodiment, the self-test enable signal is a periodic signal. In each period, the self-test enable signal has two states: valid and invalid. When the self-test enable signal is valid, the PUF feature processing and self-test module enters the working state; when the self-test enable signal is invalid, the PUF feature processing and self-test module enters the standby state. During the valid period of the self-test enable signal, the input mapping layer receives K sets of feature vectors output by the feature extraction module and performs dimension normalization and preliminary nonlinear mapping on each set of feature vectors to obtain mapped features output to the encoding layer. The encoding layer performs multi-layer nonlinear transformations on the mapped features, thereby extracting discriminative and stable latent feature vectors from each set of feature vectors and outputting them to the fingerprint generation unit for fingerprint generation. The unit receives K sets of latent feature vectors, performs threshold binarization on each set of latent feature vectors to obtain fingerprint response segments, aggregates the K sets of fingerprint response segments to generate a fixed-length device fingerprint sequence, and then compares the device fingerprint sequence bit by bit with the reference fingerprint sequence stored in the system registration stage to calculate the Hamming distance between the two. When the Hamming distance is less than or equal to a preset self-test threshold, it is determined that the physical features of the TMR sensor array are stable, and a self-test result signal representing the stability of the physical features is generated. When the Hamming distance is greater than the preset self-test threshold, it is determined that the physical features of the TMR sensor array have drifted, a self-test result signal representing the drift of the physical features is generated, and the self-test result signal is output to the decision fusion and communication module.

[0031] Compared to existing PUF generation methods based on direct signal quantization or simple linear feature mapping, the PUF feature processing and self-testing module in this embodiment adopts a deep autoencoder-based architecture, achieving machine learning enhancement of the PUF mechanism. Leveraging the powerful nonlinear feature representation capabilities of deep autoencoders, this PUF feature processing and self-testing module first preprocesses the features through an input mapping layer, and then performs multi-layer nonlinear transformations using an encoding layer to map and transform the low-dimensional initial feature vector into a high-dimensional latent feature vector. This nonlinear mapping mechanism from low to high dimensions not only enriches the expressive power of the feature space and significantly improves the uniqueness and discriminability of fingerprints, but more importantly, it establishes an extremely complex nonlinear correspondence between physical features and digital fingerprints. If an attacker attempts to fit this mapping relationship using machine learning algorithms, they must invert the nonlinear network parameters in an extremely high-dimensional space. This significantly increases the amount of training data and computational cost required for modeling attacks, thereby significantly improving the anti-cloning and anti-modeling attack capabilities of the identity authentication and anomaly detection system for TMR sensor arrays in this invention. Furthermore, through the aggregation processing mechanism of K sets of fingerprint response fragments, sporadic errors in single sampling are further eliminated, ensuring the high reliability of the final device fingerprint sequence.

[0032] Example 5: This example is basically the same as Example 4, except that: Figure 5As shown, in this embodiment, the behavior anomaly detection module includes a multi-domain feature extraction unit, a conditional variational autoencoder, and an anomaly determination unit. The conditional variational autoencoder includes a time-domain coding sub-network, a frequency-domain coding sub-network, a conditional fusion layer, a latent space mapping layer, and a decoding and reconstruction layer. The multi-domain feature extraction unit has N input terminals, a first output terminal, and a second output terminal. The time-domain coding sub-network has input terminals and an output terminal. The frequency-domain coding sub-network has input terminals and an output terminal. The conditional fusion layer has a first input terminal, a second input terminal, a tag signal input terminal, and an output terminal. The latent space mapping layer has input terminals and an output terminal. The decoding and reconstruction layer has input terminals and an output terminal. The anomaly determination unit has a first input terminal, a second input terminal, a third input terminal, and an output terminal. The N input terminals of the multi-domain feature extraction unit are respectively connected to N signal conditioning and acquisition modules. The digital output of the block is connected; the first output of the multi-domain feature extraction unit is connected to the input of the time-domain coding sub-network and the first input of the anomaly detection unit, respectively; the second output of the multi-domain feature extraction unit is connected to the input of the frequency-domain coding sub-network and the second input of the anomaly detection unit, respectively; the output of the time-domain coding sub-network is connected to the first input of the conditional fusion layer; the output of the frequency-domain coding sub-network is connected to the second input of the conditional fusion layer; the tag signal input of the conditional fusion layer is connected to the control module for accessing the running status tag, and its output is connected to the input of the latent spatial mapping layer; the output of the latent spatial mapping layer is connected to the input of the decoding and reconstruction layer; the output of the decoding and reconstruction layer is connected to the third input of the anomaly detection unit; the output of the anomaly detection unit is connected to the decision fusion and communication module.

[0033] In this embodiment, the multi-domain feature extraction unit performs time-domain convolution calculations on the N input digital signals to extract time-domain feature vectors and fast Fourier transforms to extract frequency-domain feature vectors. Its first output terminal outputs a time-domain feature vector, and its second output terminal outputs a frequency-domain feature vector. The time-domain coding sub-network is implemented using a deep neural network to encode the time-domain feature vectors to generate time-domain embedded feature vectors. The frequency-domain coding sub-network is also implemented using a deep neural network to encode the frequency-domain feature vectors to generate frequency-domain embedded feature vectors. The conditional fusion layer is used to combine the time-domain embedded feature vectors and the frequency-domain feature vectors. Embedded feature vectors and system operating status labels are jointly mapped to form latent distribution parameters under conditional prior; the latent space mapping layer is used to sample the latent distribution parameters under the reparameterization mechanism to generate latent variable vectors; the decoding and reconstruction layer is used to generate time-domain reconstruction features and frequency-domain reconstruction features with the latent variable vectors as input; the anomaly determination unit is configured to execute decision logic based on composite reconstruction error, which is different from the conventional method of simple threshold determination relying only on a single signal amplitude or a single dimension feature in the prior art; the anomaly determination unit calculates the composite reconstruction error E using formula (1): (1) Among them, Y t Y is the time-domain reconstructed feature received at the third input of the anomaly detection unit. f X is the frequency domain reconstruction feature received at the third input of the anomaly detection unit. t X is the time-domain feature vector received at the first input of the anomaly detection unit. f The frequency domain feature vector received at the second input terminal of the anomaly detection unit, ||·|| 2 The vector represents the squared Euclidean distance (or mean square error); 'a' is the time-domain weight, and 'b' is the frequency-domain weight. Both 'a' and 'b' range from greater than 0 to less than or equal to 1, used to balance the contribution of time-domain and frequency-domain features to the composite error. When the authentication and anomaly detection system is in an unknown environment or needs to accommodate multiple anomaly types simultaneously, the feature vector is usually normalized, and 'a' = 'b' = 0.5. In this case, the authentication and anomaly detection system has equal sensitivity to time-domain waveform distortion and frequency-domain energy anomalies, suitable for most general scenarios. When the application scenario mainly faces anomalies such as the proximity of a strong magnet, instantaneous pulse interference, or sensor magnetic saturation that cause drastic changes in the time-domain waveform, the weight of the time-domain weight 'a' can be increased. When the application scenario mainly faces anomalies such as electromagnetic injection attacks, power frequency interference, or mechanical periodic vibrations that mainly manifest as changes in spectral energy distribution, the weight of the frequency-domain weight 'b' needs to be increased.

[0034] The anomaly determination unit compares the calculated composite reconstruction error E with the system's preset threshold (determined according to the actual application environment); when E is less than or equal to the preset threshold, it generates an anomaly diagnosis result signal indicating that the behavior of the TMR sensor array is normal; when E is greater than the preset threshold, it generates an anomaly diagnosis result signal indicating that the behavior of the TMR sensor array is abnormal, and outputs the anomaly diagnosis result signal to the decision fusion and communication module.

[0035] In this embodiment, the behavior anomaly detection module introduces system operating status labels as prior conditions into the conditional fusion layer, enabling the conditional variational autoencoder to possess operating condition awareness and automatically adapt to signal fluctuations caused by excitation voltage changes in self-test mode. This mechanism successfully resolves the technical contradiction that introducing identity self-testing into identity authentication and anomaly detection systems often disrupts signal references, leading to anomaly detection failure. This allows the identity authentication and anomaly detection system to accurately isolate normal fluctuations caused by excitation voltage changes even during drastic self-test excitation voltage scans, maintaining high-precision monitoring of subtle external anomalies. Simultaneously, utilizing a parallel feature extraction mechanism of time-domain and frequency-domain coding sub-networks, the conditional variational autoencoder can deeply fuse the waveform texture and spectral energy distribution of the signal in the latent space. Compared to single-domain features, this multi-domain reconstruction mechanism significantly improves the sensitivity to subtle frequency anomalies or transient impacts masked by noise. The anomaly determination unit does not rely solely on the deviation of a single dimension, but instead calculates the composite reconstruction error between the time-domain feature vector, the frequency-domain feature vector, and the reconstructed features. This composite determination mechanism based on multi-source feature comparison achieves cross-verification of the signal state, effectively identifying covert attacks or faults that only pretend to be normal in a single domain. This significantly reduces the false alarm rate of the identity authentication and anomaly detection system for TMR sensor arrays in this invention, ensuring the comprehensiveness and accuracy of the anomaly diagnosis results.

[0036] Example 6: This example is basically the same as Example 5, except that: Figure 6 As shown, in this embodiment, the decision fusion and communication module includes a decision fusion unit and a communication circuit; the decision fusion unit has a first input terminal, a second input terminal, and an output terminal; the communication circuit has a data input terminal, a data output terminal, an instruction input terminal, and an instruction output terminal; the first input terminal of the decision fusion unit is connected to the output terminal of the fingerprint generation unit, the second input terminal is connected to the output terminal of the anomaly determination unit, and its output terminal is connected to the data input terminal of the communication circuit; the instruction output terminal of the communication circuit is connected to the control module.

[0037] In this embodiment, the decision fusion unit is used to receive the self-test result signal output by the fingerprint generation unit and update and maintain the identity authentication status parameters using the self-test result signal; on the other hand, it is used to receive the anomaly diagnosis result signal output by the anomaly determination unit and perform weighted logic fusion on the identity authentication status parameters and the anomaly diagnosis result signal to generate an operating status signal characterizing the overall security status of the sensor array, and output it to the communication circuit; the communication circuit is used to send the operating status signal to the external upper control system or cloud platform through its data output terminal; at the same time, it receives control commands from the outside through the command input terminal and transmits them to the control module through its command output terminal, so that the control module can trigger the self-test mode or adjust the system parameters according to the control commands.

[0038] Example 7: This example is basically the same as Example 6, except that: Figure 7 As shown, in this embodiment, the control module includes a microcontroller and a digital-to-analog converter circuit. The microcontroller has an excitation control terminal, a sampling control terminal, an enable signal output terminal, a tag signal output terminal, and an instruction input terminal. The excitation control terminal of the microcontroller is connected to the input terminal of the digital-to-analog converter circuit, and the output terminal of the digital-to-analog converter circuit is connected to the power input terminal of the TMR sensor array. The sampling control terminal of the microcontroller is connected to the control input terminals of the analog-to-digital converter circuits in the N signal conditioning and acquisition modules. The enable signal output terminal of the microcontroller is connected to the enable signal input terminal of the data buffer unit and the enable signal input terminal of the input mapping layer, respectively. The tag signal output terminal of the microcontroller is connected to the tag signal input terminal of the conditional fusion layer. The instruction input terminal of the microcontroller is connected to the instruction output terminal of the communication circuit.

[0039] In this embodiment, the specific control logic of the control module is as follows: the microcontroller is internally equipped with timing control logic to monitor whether a preset time interval T2 has been reached, and monitors whether an external forced self-test command has been received through the command input terminal; when in normal operating mode: the microcontroller sends a first control command through the excitation control terminal to control the digital-to-analog converter circuit to output a single preset constant excitation voltage to the TMR sensor array; at the same time, it sends a system operating status tag representing the normal operating state to the conditional fusion layer through the tag signal output terminal, and controls the analog-to-digital converter circuits of N signal conditioning and acquisition modules to work according to the normal monitoring cycle in normal operating mode through the sampling control terminal; and sets the self-test enable signal output by the enable signal output terminal to an invalid state; when the preset time interval T2 is reached or an external forced self-test command is received... When an external forced self-test command is received, and the self-test mode is switched and imported: the microcontroller sends a second control command through the excitation control terminal to control the digital-to-analog converter circuit to generate a stepped excitation signal, which sequentially outputs K sets of constant excitation voltages with different amplitudes to the TMR sensor array, with an amplitude range of 0 to 3.3V; during the output of each set of constant excitation voltages, the microcontroller sends a valid self-test enable signal through its enable signal output terminal, and at the same time controls the analog-to-digital converter circuits of N signal conditioning and acquisition modules to perform L consecutive samplings according to the self-test sampling cycle through the sampling control terminal; at the same time, the microcontroller sends a system operation status tag representing the self-test status to the conditional fusion layer through its tag signal output terminal, so as to instruct the behavior anomaly detection module to perform anomaly monitoring under the current superimposed self-test excitation signal.

[0040] To further verify the actual performance, long-term stability, and beneficial effects of the identity authentication and anomaly detection system for TMR sensor arrays of this invention, a dedicated experimental verification platform was constructed for testing and data analysis. The verification platform uses a high-performance microcontroller as the core of its control module. This microcontroller is configured to perform PUF lifecycle management, specifically responsible for excitation voltage generation in self-test mode, data synchronization acquisition, and anomaly detection model inference in operating mode. At the sensing front end, three independent TMR sensor array devices were constructed as test samples. The differential analog signal output from each TMR sensor was processed by a signal conditioning and acquisition module. This module uses a precision operational amplifier to construct a multi-channel filtering and amplification circuit. The processed signal is then input to the microcontroller's built-in 16-bit high-precision ADC for synchronous sampling. To eliminate uncontrollable environmental interference and simulate specific magnetic field disturbances, data acquisition was conducted in a permalloy magnetic shielding chamber. The shielded room contains a Helmholtz coil, which is powered by a regulated DC power supply. The excitation current is finely adjusted by a series high-precision adjustable resistor, thereby generating a precisely adjustable static background magnetic field inside the shield to simulate the environmental magnetic field disturbances that the TMR sensor may encounter under actual working conditions.

[0041] In self-test mode, the analog-to-digital converter circuit of the control module generates a stepped-wave excitation sequence containing 64 discrete voltage values ​​(i.e., K=64) to acquire data from three different TMR sensor array devices. For the acquired response data, a feature extraction module extracts differential statistical features, and PUF feature processing and self-test module generate a 128-bit device fingerprint sequence. The inter-group Hamming distance heatmap of the device fingerprint sequence under different excitation conditions for the TMR sensor array authentication and anomaly detection system of this invention is shown below. Figure 8 As shown; the distribution of Hamming distances between groups is shown in the figure. Figure 9 As shown. Figure 8 Used to assess the uniqueness of a device fingerprint sequence. Figure 8 In the graph, both the horizontal and vertical axes represent the DAC excitation values, and the color intensity represents the Hamming distance. Analysis Figure 8 As can be seen, the diagonal region is dark blue, representing autocorrelation under the same excitation conditions; while the off-diagonal region exhibits a rich warm hue, indicating that the generated device fingerprint sequence changes significantly when the excitation voltage changes. This proves that the nonlinear features extracted by this invention using K groups of different excitation voltages have extremely low correlation, effectively expanding the "excitation-response" feature space, making it difficult for attackers to predict the response under other excitations using limited observation data. Figure 9 The distribution of normalized Hamming distances between a large number of different sample pairs was statistically analyzed. Figure 9In the diagram, the blue bars represent the density distribution of the actual statistical data, and the red curve represents the fitted normal distribution curve. Analysis Figure 9 It is known that the inter-group Hamming distance of the device fingerprint sequences generated by this invention follows a normal distribution. Its average value is 0.5035, and its standard deviation is 0.1125. The average value is extremely close to the ideal theoretical value of 0.5, which means that half of the bits are different between any two different device fingerprint sequences. This proves that the physical fingerprints (device fingerprint sequences) extracted by this invention have extremely high randomness and uniqueness, and the probability of physical fingerprint collisions between different devices is extremely low.

[0042] To further verify the ability of the feature vectors extracted by the multi-domain feature extraction unit of this invention to distinguish different operating states, a visualization analysis was performed on the test sample set. The time-domain and frequency-domain PCA distribution diagrams of the identity authentication and anomaly detection system for TMR sensor arrays of this invention are shown below. Figure 10 and Figure 11 As shown. Figure 10 and Figure 11 This paper presents a visualization of the time-domain and frequency-domain feature vector distributions for a representative time period sample. Three typical operating conditions were selected as samples: normal state (dark blue dots), spike pulse interference (dark green dots), and instantaneous magnetic saturation (light green dots). For these samples, high-dimensional time-domain and frequency-domain feature vectors were extracted using a multi-domain feature extraction unit, and principal component analysis was employed to map the high-dimensional features onto a two-dimensional coordinate system for visualization. Figure 10 The diagram illustrates the distribution of temporal convolutional features. Normal category sample points are highly clustered in the central region, forming a tight cluster, indicating that the temporal waveform features of the TMR sensor exhibit extremely high consistency and stability during normal operation. In contrast, sample points from spike pulses and transient saturation categories are scattered in the peripheral region of the normal cluster, showing a significant distance from the normal samples. Figure 11 The distribution of frequency domain features is shown in the diagram. Similarly, normal samples are tightly clustered in specific regions of low-dimensional space, while abnormal samples exhibit a specific divergent pattern; this means that frequency domain features can keenly capture the differences in the spectral energy distribution of abnormal signals.

[0043] In summary, the identity authentication and anomaly detection system for TMR sensor arrays of the present invention integrates physical feature recognition and anomaly detection functions. Through unified scheduling of N signal conditioning and acquisition modules, and through the control module controlling the working status of the feature extraction module, PUF feature processing and self-testing module, behavior anomaly detection module, and decision fusion and communication module, the identity authentication and anomaly detection system is constructed. This achieves time-division multiplexing of identity authentication and anomaly detection functions on the same hardware link. When deployed in a TMR sensor array, it has a small overall footprint and low power consumption. Furthermore, the collaborative fusion of physical feature recognition and anomaly detection enables continuous verification from "trustworthy device identity" to "trustworthy operating status," and provides strong defense against complex attacks.

Claims

1. An authentication and anomaly detection system for a TMR sensor array, wherein the TMR sensor array comprises multiple TMR sensors, and the number of TMR sensors in the TMR sensor array is denoted as N, characterized in that: The identity authentication and anomaly detection system includes a control module, N signal conditioning and acquisition modules, a feature extraction module, a PUF feature processing and self-testing module, a behavior anomaly detection module, and a decision fusion and communication module. The N signal conditioning and acquisition modules are connected one-to-one with N TMR sensors. The control module provides excitation signals to each TMR sensor and coordinates the timing of the N signal conditioning and acquisition modules, feature extraction module, PUF feature processing and self-testing module, behavior anomaly detection module, and decision fusion and communication module, enabling the identity authentication and anomaly detection system to perform identity authentication and anomaly detection functions. The identity authentication and anomaly detection system has two operating modes: normal operation mode and self-testing mode. The default operating mode of the identity authentication and anomaly detection system is normal operation mode. The control module periodically controls the identity authentication and anomaly detection system to enter self-testing mode at preset time intervals T2, or controls the identity authentication and anomaly detection system to enter self-testing mode when receiving an external forced self-testing command. When the identity authentication and anomaly detection system is in normal operation mode and has not entered self-testing mode... During operation, the control module maintains a preset constant excitation voltage applied to each TMR sensor, ensuring each TMR sensor outputs a corresponding differential analog signal. It controls the feature extraction module and the PUF feature processing and self-test module to be in standby mode, disabling signal reception. It also controls N signal conditioning and acquisition modules, the behavior anomaly detection module, and the decision fusion and communication module to periodically perform monitoring according to a preset normal monitoring cycle T1, generating an operating status signal representing the overall safety status of the sensor array in each cycle. When the identity authentication and anomaly detection system is in normal operating mode and has been switched to self-test mode, the control module controls the feature extraction module and the PUF feature processing and self-test module to enter working mode, enabling signal reception. According to a preset excitation cycle, it outputs a preset constant excitation voltage to each TMR sensor in each excitation cycle, causing each TMR sensor to output a corresponding differential analog signal, until the end of the Kth excitation cycle. Here, K is a positive integer ranging from 32 to 64, and the amplitude of the constant excitation voltage output in each of the K excitation cycles is different, ranging from 0 to 3.3V; In each excitation cycle, the control module controls N signal conditioning and acquisition modules, behavior anomaly detection module, and decision fusion and communication module to perform L self-check monitoring cycles, where L is a positive integer ranging from 5 to 20, and the self-check monitoring cycle is shorter than the normal monitoring cycle; and in each excitation cycle, the control module controls the feature extraction module and the PUF feature processing and self-check module to perform an identity information acquisition operation once, until the identity information acquisition operation of the Kth excitation cycle is completed, then the control module controls the PUF feature processing and self-check module to perform an identity authentication operation once, and finally the control module controls the decision fusion and communication module to generate an operating status signal characterizing the overall safety status of the sensor array.

2. The authentication and anomaly detection system for a TMR sensor array according to claim 1, characterized in that, The monitoring process is as follows: the control module first controls each signal conditioning and acquisition module to perform a sampling and conditioning operation once, and then controls the behavior anomaly detection module and the decision fusion and communication module to perform an anomaly detection operation once. Specifically, each signal conditioning and acquisition module performs a sampling and conditioning operation by: first sampling the differential analog signal output by its corresponding TMR sensor; then performing differential amplification and filtering on the differential analog signal to obtain one analog signal; then converting the analog signal into a digital signal and outputting it to the behavior anomaly detection module and the feature extraction module respectively; the behavior anomaly detection module and the decision fusion and communication module... The specific steps of an anomaly detection operation are as follows: The behavior anomaly detection module performs time-domain transformation and frequency-domain transformation on the N digital signals output from the N signal conditioning and acquisition modules, generating N time-domain feature vectors and N frequency-domain feature vectors. Then, it performs feature fusion and reconstruction operations on the N time-domain signal features and N frequency-domain signal features to obtain a composite reconstruction error value. The composite reconstruction error value is then compared with a preset threshold to generate an anomaly diagnosis result signal characterizing the behavior state of the TMR sensor array and output it to the decision fusion and communication module. The decision fusion and communication module generates a corresponding operating status signal output based on the anomaly diagnosis result signal.

3. The authentication and anomaly detection system for a TMR sensor array according to claim 2, characterized in that, The identity information acquisition process specifically involves: the feature extraction module receiving each digital signal output by each signal conditioning and acquisition module during the current excitation cycle; and after receiving L digital signals output by each signal conditioning and acquisition module, arranging these L digital signals in chronological order to form a digital signal sequence of length L as the original response sequence, resulting in N original response sequences. Then, point-to-point difference operations are performed on any two original response sequences to obtain the difference response sequence between each pair of original response sequences. At this point, a total of N original response sequences are obtained. We obtain N differential response sequences of length L; then, we calculate the mean of each original response sequence as a channel mean feature, resulting in N channel mean features. We then extract five differential statistical features from each differential response sequence. Finally, we concatenate the N channel mean features with the five differential statistical features from all differential response sequences in a preset order to generate a sequence of dimension N+. A set of feature vectors of size ×5 is output to the PUF feature processing and self-testing module; the PUF feature processing and self-testing module performs feature mapping and encoding processing on the received set of feature vectors to generate corresponding fingerprint response data; the identity authentication process is as follows: the PUF feature processing and self-testing module first aggregates the generated K fingerprint response data to obtain a device fingerprint sequence, then compares the device fingerprint sequence with a pre-stored reference fingerprint sequence and calculates the difference between the two; when the difference is less than or equal to a preset self-testing threshold, a self-testing result signal characterizing the stability of the physical characteristics of the TMR sensor array is generated; when the difference is greater than the preset self-testing threshold, a self-testing result signal characterizing the drift of the physical characteristics of the TMR sensor array is generated; and the self-testing result signal is output to the decision fusion and communication module.

4. The authentication and anomaly detection system for a TMR sensor array according to claim 3, characterized in that, The state update process specifically involves: the decision fusion and communication module updating the identity authentication status parameters of the TMR sensor array based on the self-test result signal it currently receives; and logically fusing the abnormal diagnosis result signal output by the behavior anomaly detection module at the current moment with the updated identity authentication status parameters to generate an operating status signal characterizing the overall safety status of the sensor array, and uploading it to the host computer or cloud monitoring platform.

5. The authentication and anomaly detection system for a TMR sensor array according to claim 2, characterized in that, Each signal conditioning and acquisition module includes a differential amplifier circuit, a filter circuit, and an analog-to-digital converter circuit. The differential amplifier circuit has a positive input terminal, a negative input terminal, and an output terminal; the filter circuit has an input terminal and an output terminal; and the analog-to-digital converter circuit has an analog input terminal, a control input terminal, and a digital output terminal. The positive and negative input terminals of the differential amplifier circuit are used to acquire and input the differential analog signal output by the TMR sensor, and the output terminal is connected to the input terminal of the filter circuit. The output terminal of the filter circuit is connected to the analog input terminal of the analog-to-digital converter circuit. The control input terminal of the analog-to-digital converter circuit is connected to the control module, and the digital output terminal is connected to the feature extraction module and the behavior anomaly detection module, respectively.

6. The authentication and anomaly detection system for a TMR sensor array according to claim 5, characterized in that, The feature extraction module includes a data buffer unit, a differential calculation unit, and a feature vector generation unit. The data buffer unit has N input terminals, an enable signal input terminal, and an output terminal. The differential calculation unit has an input terminal and an output terminal. The feature vector generation unit has a first input terminal, a second input terminal, and an output terminal. The N input terminals of the data buffer unit are respectively connected to the output terminals of N signal conditioning and acquisition modules. The enable signal input terminal is connected to the control module. The output terminal is connected to the input terminal of the differential calculation unit and the second input terminal of the feature vector generation unit. The output terminal of the differential calculation unit is connected to the first input terminal of the feature vector generation unit. The enable signal input terminal of the data buffer unit receives a self-test enable signal from the control module.

7. The authentication and anomaly detection system for a TMR sensor array according to claim 6, characterized in that: The PUF feature processing and self-testing module includes a deep autoencoder and a fingerprint generation unit. The deep autoencoder includes an input mapping layer and an encoding layer. The input mapping layer has an input terminal, an enable signal input terminal, and an output terminal. The encoding layer has an input terminal and an output terminal. The fingerprint generation unit has an input terminal and an output terminal. The input terminal of the input mapping layer is connected to the output terminal of the feature extraction module, its enable signal input terminal is connected to the control module, and its output terminal is connected to the input terminal of the encoding layer. The output terminal of the encoding layer serves as the output terminal of the deep autoencoder and is connected to the input terminal of the fingerprint generation unit. The output terminal of the fingerprint generation unit is connected to the decision fusion and communication module. The enable signal input terminal of the input mapping layer receives a self-test enable signal from the control module.

8. The authentication and anomaly detection system for a TMR sensor array according to claim 7, characterized in that, The behavior anomaly detection module includes a multi-domain feature extraction unit, a conditional variational autoencoder, and an anomaly determination unit. The conditional variational autoencoder includes a temporal coding sub-network, a frequency coding sub-network, a conditional fusion layer, a latent space mapping layer, and a decoding and reconstruction layer. The multi-domain feature extraction unit has N input terminals, a first output terminal, and a second output terminal. The temporal coding sub-network has input terminals and output terminals. The frequency coding sub-network has input terminals and output terminals. The conditional fusion layer has a first input terminal, a second input terminal, a label signal input terminal, and an output terminal. The latent spatial mapping layer has an input terminal and an output terminal; the decoding and reconstruction layer has an input terminal and an output terminal; the anomaly detection unit has a first input terminal, a second input terminal, a third input terminal, and an output terminal; the N input terminals of the multi-domain feature extraction unit are respectively connected to the digital output terminals of the N signal conditioning and acquisition modules; the first output terminal of the multi-domain feature extraction unit is respectively connected to the input terminal of the time-domain coding sub-network and the first input terminal of the anomaly detection unit; the second output terminal of the multi-domain feature extraction unit is respectively connected to the input terminal of the frequency-domain coding sub-network and the second input terminal of the anomaly detection unit; the output terminal of the time-domain coding sub-network is connected to the first input terminal of the conditional fusion layer; the output terminal of the frequency-domain coding sub-network is connected to the second input terminal of the conditional fusion layer; the tag signal input terminal of the conditional fusion layer is connected to the control module for accessing the running status tag, and its output terminal is connected to the input terminal of the latent spatial mapping layer; the output terminal of the latent spatial mapping layer is connected to the input terminal of the decoding and reconstruction layer; the output terminal of the decoding and reconstruction layer is connected to the third input terminal of the anomaly detection unit; the output terminal of the anomaly detection unit is connected to the decision fusion and communication module.

9. The authentication and anomaly detection system for a TMR sensor array according to claim 8, characterized in that: The decision fusion and communication module includes a decision fusion unit and a communication circuit; the decision fusion unit has a first input terminal, a second input terminal, and an output terminal. The communication circuit has a data input terminal, a data output terminal, an instruction input terminal, and an instruction output terminal; the first input terminal of the decision fusion unit is connected to the output terminal of the fingerprint generation unit, the second input terminal is connected to the output terminal of the anomaly determination unit, and its output terminal is connected to the data input terminal of the communication circuit; the instruction output terminal of the communication circuit is connected to the control module.

10. The authentication and anomaly detection system for a TMR sensor array according to claim 9, characterized in that: The control module includes a microcontroller and a digital-to-analog converter (DAC). The microcontroller has an excitation control terminal, a sampling control terminal, an enable signal output terminal, a tag signal output terminal, and an instruction input terminal. The excitation control terminal of the microcontroller is connected to the input terminal of the DAC, and the output terminal of the DAC is connected to the power input terminal of the TMR sensor array. The sampling control terminal of the microcontroller is connected to the control input terminals of the analog-to-digital converters in the N signal conditioning and acquisition modules. The enable signal output terminal of the microcontroller is connected to the enable signal input terminal of the data buffer unit and the enable signal input terminal of the input mapping layer, respectively. The tag signal output terminal of the microcontroller is connected to the tag signal input terminal of the conditional fusion layer. The instruction input terminal of the microcontroller is connected to the instruction output terminal of the communication circuit.

Images