DEVICE IDENTIFICATION AND DETECTION METHOD BASED ON ELECTROMAGNETIC EMISSION SIGNALS EMITTED BY DEVICES

Abstract

The invention relates to a method for identifying and detecting electronic devices by examining the electromagnetic emission signals they emit during operation.

Description

1 TARIFF ELECTROMAGNETIC EMISSION SIGNAL EMITTED BY DEVICES DEVICE IDENTIFICATION AND TRANSACTION METHOD Technical Area 5 The invention concerns the electromagnetic emissions released by electronic devices during operation. a signal to create a device identity and to detect that identity It is related to the method. Previous Technique With the proliferation of electronic devices in various fields, efficient and reliable devices are becoming increasingly common. The need for identification techniques has become increasingly important. Traditional device identification methods typically use serial numbers or MAC addresses. It relies on easily spoofed identifiers, such as addresses. This situation, the ability to spoof identifiers and allow unknown devices to infiltrate critical systems It makes it possible to reach them. 15 Unknown devices imitating other devices and being present in critical environments. This is a very significant threat to system security. To overcome this situation... Devices are given unique artificial identities, and these identities are controlled. However, the artificial assignment of these identities also implies that these are artificial identities. 20 to prevent the system from being infiltrated by being taken over by other devices It is not possible to pass. The electromagnetic emission signal that electronic devices emit during operation. There are ongoing efforts to identify devices. 2 The device is described in document US2021406861A1, which is included in the known state of the art. emissions released by electronic devices during their operation for identity verification The method being discussed utilizes an electromagnetic emission signal. The method utilizes a deep learning algorithm. According to document JP2020154351A, which is included in the known state of the art, device 5 emissions released by electronic devices during their operation for identity verification This refers to a method that uses an electromagnetic emission signal. "Electromagnetic radiation-based IC device" is included in the state of the art. identification and verification using deep learning (Hong-xin Zhang et all, 2020)” The article titled "Device Identification During the Operation of Electronic Devices" describes 10... a method that uses the electromagnetic emission signal it emits It is mentioned. When the solutions mentioned in these documents are examined, the emissions from the devices Identifying devices using electromagnetic emission signals It appears that this was done using deep learning algorithms. However, 15 The electromagnetic signals emitted by devices are not constant. For example, a device goes into sleep mode. While operating in mode, a different electromagnetic emission signal, at full capacity It may emit a different electromagnetic emission signal while operating. Therefore, through the electromagnetic emission signal emitted by the devices During the identification process, the operating modes (states) of the devices and the 20 in those modes are considered. the differences that can arise between the electromagnetic emission signals they can emit Differences should also be taken into account. Purposes of the Invention The purpose of this invention is to reduce the emissions produced by electronic devices during operation. Creating a device identity using an electromagnetic emission signal and this identity 25 It is the implementation of a method for detection. 3 Another purpose of this invention is to analyze the electromagnetic emission signal emitted by the devices. Device identification is performed via the system, and the modes (states) in which the devices are operating are also indicated. and among the electromagnetic emission signals they can emit in these modes a method of identification that takes into account possible differences It is the realization of. 5 The invention describes a method that targets the emissions produced by electronic devices during their operation. Identification is performed using an electromagnetic emission signal. Electronic electromagnetic signals emitted by devices are analyzed using deep learning methods. By classifying them, the safety of the devices is ensured with this model. Every electronic device... Thanks to the tolerance values ​​of the components inside the device, it is an exact match. Even if they are produced on the same production line, the components have different specifications, therefore the devices differ. They can be differentiated, thus creating unique identities. In the method described in the invention, the raw electromagnetic signal received by near-field probes... deep learning model to learn complex patterns and features from data The model was created using electromagnetic signals from various devices. By training with a generated dataset, a robust and accurate device identification system can be created. The aim was to improve it. When measurements are taken in the area close to electronic boards. Signals can often be more complex because electromagnetic waves Reflections, distortions, and interactions are more pronounced because they are close to the objects. This complexity arises from electromagnetic activity occurring near a device. This allows for a detailed examination of the interactions. Thus electromagnetic near field probe of electronic boards or electronic devices by receiving signals and training them with deep learning, they become unique. Identities can be created. Electronic systems used in critical environments and systems. This identification process is carried out to prevent the devices from being counterfeited. 25 It is of considerable importance. In addition, the modes (states) in which the devices operate while all these processes are being carried out, and this the electromagnetic emission signals that can be emitted in different modes 4 The differences were also taken into account, and model training was conducted accordingly. For this process... The output layer of the deep learning model also includes information about the operating modes of the devices. It is receiving. Detailed Description of the Invention The operation of the electronic devices created to achieve the purpose of this invention is 5 Device identification is determined by the electromagnetic emission signal it emits during operation. The methods for identifying this person are shown in the attached figures. This shape; Figure 1: Schematic of the measuring device to be used in the method described in the invention. It is the appearance. 10 Figure 2: Schematic of the interior of the measuring device to be used in the method described in the invention. It is the appearance. Figure 3: Flowchart of the data collection and data set creation process in the method described in the invention. It is a diagram. Figure 4: The process of creating the deep learning model in the method in question (15). It is a flowchart. Figure 5: Identification of the device identity created in the method described in the invention. It is a flowchart of the process. The parts shown in the figure are individually numbered, and the corresponding numbers correspond to these numbers. given below. 20 1. Measuring device 2. Signal section 3. Near field probe 4. Analog-to-digital converter 5. Embedded GPU 6. Touch screen 7. Battery The subject of the invention is method 5. - Measurements taken from various points on the devices to be measured and in close proximity for each operating mode. Measurement with field probes (3), - The measurements taken are first sent to the preamplifier to amplify the signal, and from there to the analog amplifier. sending to the digital converter (4), - Converting analog measurement values ​​into digital data and transmitting them to the embedded GPU (5) 10 and the measuring device associated with this data, the operating mode of the device and the embedded measurement point. Creating a dataset by labeling on the GPU (5), - By providing this dataset as input to a deep learning model, we can perform deep learning. training the model, - Device identities and operating modes can be detected thanks to the trained model. 15 to be able to, - Real-time execution and measurement of the deep learning model. Data collected from these points determines the identity and current operating mode of the devices. security checks are carried out thanks to its detection. It includes the following steps. 20 Figure 1 shows a schematic representation of the measuring device (1). Signals The signal section (2), which is the part where it is received, processed and displayed, and the design There is a battery (7) that provides its energy. Detailed information about the signal section (2) 6 When viewed, near field probes (3), analog-to-digital converter (4), embedded GPU (5) and It is shown in Figure 2 that it consists of a touch screen (6). Near field probes (3) detect the operation of an electronic circuit board or electronic device. It captures electromagnetic signals in detail. The systems used here... The number of near field probes (3) was considered to be 5, but there is more than one near field probe. It is possible to use the signals received by the analog-to-digital converter (4) to the device. It is measured separately at specific points. For this, a touchscreen is used. (6) Point selection is very important. Measurements on specific points of the device These points are recorded on the embedded GPU (5). These points are recorded manually. These can be determined and recorded during the measurement, or a 10 placed in front of the measuring instrument can be used. It is possible to automate this with the help of a camera. Separate cameras at designated points. Separate measurements are taken and recorded along with the device names. In addition, the operating mode of the device being measured is also recorded. The device being monitored, measurement points, measurement results, and device modes, along with the embedded GPU, are all included. (5) A dataset is created within. All 15 to be identified in a system These processes are carried out for the devices to first create a suitable dataset. The created dataset... After the dataset, training is performed with the deep learning model in the embedded GPU (5). Here, if raw data is to be used for training the deep learning model, LSTM, GRU, and For 2D images converted to spectrograms when using methods such as TCN. Methods such as CNN can be used. Apart from these models, different depth 20 Learning techniques are also subjected to / not subjected to different pre-processing steps. It is available for testing credentials generated using a deep learning model. deep learning model for embedded GPU (5) in real time It is being run. An electronic board or device registered in the deep learning model. The following steps are followed to test it. 25 1. Select the measurement point via the touchscreen (6). (Built-in camera) In this case, the camera identifies direct points on the image. 7 If not, then it consists of 5 points: a midpoint and 4 corner points. (This can be done, but it can also be obtained from more specific points for the study) 2. Take the measuring device (1) to the measurement point and start the measurement. 3. When creating the deep learning model, for all points taken from the device Apply this. 5 4. All the collected data goes to the deep learning model and the result... is shown on the touchscreen (6). 5. The measurement accuracy is very low or the device is brought to the touch screen (6) If it's incorrect, the system detects that the device is imitating something. All necessary steps (data set collection, training, and testing) have been completed using the embedded GPU 10. (5) It is being done within and there is no connection to the server. However, locally with a lower-level embedded Linux in applications that don't need to run as such. This is possible by working with a server connection. Near field probes (3) consist of magnetic and electric field probes. Example In practice, it is stated as 5, but there may be fewer or more of these 15. It can be used from probes. Analog-to-digital converter (4), from near field probes (3) Converting the received analog signals to digital via UART, SPI, I2C, PCI-E etc. transferring to a digital device (here embedded GPU (5)) via protocols Here, the level of signals received from the near field probes (3) is provided. To amplify, the signals are first sent to the pre-amp and then to the embedded GPU (5) 20 It is being transmitted. The embedded GPU (5) processes the signals, which is the most important part of the operation. This is the section where identification and verification are performed. Here, device data is processed. To assemble the set, we first need to obtain the location information from the user, then... the analog-to-digital converter (4) measures the signals and point information and the device 25 8 It performs the process of labeling the information. For educational purposes, it is received from the user. It creates a deep learning model using the information it possesses. Thanks to the GPU (5), it is possible to perform parallel processing during training. It allows for faster model generation compared to traditional calculation methods. Test 5 was conducted to ensure the deep learning model created could correctly identify the identities. In this section, information is again obtained from the user and measurements are taken from the measurement points. The data is collected. The obtained values ​​are fed into deep learning processes, and the output is given to the user. is sent. Interaction with the user is via the touch screen (6). is being done. In addition, the modes (states) in which the devices operate while all these processes are being carried out, and these 10 the electromagnetic emission signals that can be emitted in different modes The differences were also taken into account, and model training was conducted accordingly. For this process... The output layer of the deep learning model also includes information about the operating modes of the devices. It is receiving.

Claims

9 REQUESTS 1. The invention relates to the electromagnetic radiation emitted by electronic devices during operation. Creating and identifying the device using the emission signal. is a method aimed at achieving this, - Measurements will be taken from various points on the devices to be measured, and 5 for each operating mode. Measurement with near field probes (3), - The measurements taken are first sent to the preamplifier to amplify the signal, and from there to... sending to analog-to-digital converter (4), - Converting analog measurement values ​​into digital data and sending them to the embedded GPU (5) transmission and the measuring device for this data, the operating mode of the device and the measurement 10 a dataset by labeling the point on the embedded GPU (5) creation, - By providing this dataset as input to a deep learning model, we can perform deep learning. training the model, - Device identities and operating modes can be detected thanks to the trained model. 15 to be able to, - Real-time execution and measurement of the deep learning model. Data collected from these points determines the identity and current operation of the devices. Security check is performed by detecting the mode. It is characterized by including steps. 20

Images