Phase-Shift Configuration Method for Improving Physical Layer Security in Reconfigurable Intelligent Surfaces-Based Mobile Networks

Abstract

The present invention relates to a method for adjusting the angle of an intelligent reflective panel (RIS) to improve physical layer security performance in a mobile network, wherein the angle of the intelligent reflective panel (RIS) is adjusted to improve physical layer security performance in a mobile network in which nodes are moving. A method for adjusting the angle of a reflective panel to improve physical layer security performance in a mobile network according to the present invention comprises a source node (S), a destination node (D), an eavesdropper (E), and a UAV (Unmanned Aerial Vehicle) (U) equipped with an intelligent reflective panel (RIS) having R reflective panels, wherein a signal transmitted from the source node (S) is reflected by the intelligent reflective panel (RIS) of the UAV (U) and transmitted to the destination node (D), and the UAV (U) adjusts the angle of the reflective panel equipped in the intelligent reflective panel (RIS) to improve physical layer security performance, thereby preventing the signal reflected from the UAV (U) from being intercepted by the eavesdropper (E).

Description

Technology Field

[0001] The present invention relates to a method for adjusting the angle of an intelligent reflective panel to improve physical layer security performance in a mobile network, and more specifically, to a method for adjusting the angle of an intelligent reflective panel to improve physical layer security performance in a mobile network where nodes are moving. Background Technology

[0003] The next-generation network for 6G aims to provide infinite connectivity through the integration of terrestrial networks (TN) and non-terrestrial networks (NTN). Terrestrial networks (TN) primarily consist of ground-based infrastructure such as base stations and user equipment (UE). In contrast, non-terrestrial networks (NTN) integrate aerial platforms such as satellites, High Altitude Platforms (HAPs), and Unmanned Aerial Vehicles (UAVs) to provide more comprehensive coverage and enhanced connectivity in remote or unserviced areas.

[0004] One of the ongoing challenges in both terrestrial (TN) and non-terrestrial (NTN) networks is managing line-of-sight (LOS) and non-line-of-sight (NLOS) propagation issues. Under line-of-sight (LOS) conditions, signals travel directly from transmitter to receiver, generally creating a stronger and more stable communication link. However, in non-line-of-sight (NLOS) environments where signals are obscured by buildings, terrain, or other obstacles, signal quality can be severely degraded. Furthermore, channel conditions in these networks become more variable and complex due to the mobility of nodes, such as moving vehicles, UAVs (U), and users.

[0005] Physical Layer Security (PLS) is a critical concern for both terrestrial (TN) and non-terrestrial (NTN) networks, aiming to protect data from eavesdropping and unauthorized access by leveraging the physical characteristics of the communication medium. In other words, Physical Layer Security aims to ensure that while legitimate users can obtain messages from normally received signals, eavesdroppers (E) cannot obtain messages from intercepted signals. In this type of Physical Layer Security, security performance is measured by the difference between the main channel and the intercepted channel.

[0006] Meanwhile, Reconfigurable Intelligent Surfaces (RIS) have recently been introduced to mobile networks. RIS is a collection of metamaterials (reflective panels) that reflect signals, and each reflective panel can adjust its angle using a motor. Through this, signals are controlled to support high spectral efficiency and energy efficiency. However, the introduction of these intelligent reflective panels (RIS) poses a new challenge to physical layer security, because while RIS can improve signal propagation, it can also unintentionally reflect signals in a direction that can be intercepted by an eavesdropper (E). Prior art literature

[0008] Korean Patent Publication No. 10-2697063 (Aug. 20, 2024) Korean Patent Publication No. 10-2315147 (October 19, 2021) The problem to be solved

[0009] The present invention is proposed to solve physical layer security problems in mobile networks where intelligent reflective panels (RIS) exist. The objective of the present invention is to provide a method for adjusting the angle of an intelligent reflective panel to improve physical layer security performance, which can improve physical layer security performance by adjusting the angle of the intelligent reflective panel (RIS) existing in the mobile network to maximize security channel capacity. means of solving the problem

[0011] A method for adjusting the angle of an intelligent reflective panel (RIS) to improve physical layer security performance in a mobile network according to the present invention for achieving the above objective comprises, in a mobile network including a source node (S), a destination node (D), an eavesdropper (E), and a UAV (Unmanned Aerial Vehicle) (U) equipped with an intelligent reflective panel (RIS) having R reflective panels, a signal transmitted from the source node (S) is reflected by the intelligent reflective panel (RIS) of the UAV (U) and transmitted to the destination node (D), wherein the UAV (U) adjusts the angle of the reflective panel equipped in the intelligent reflective panel (RIS) to improve physical layer security performance, thereby preventing the signal reflected from the UAV (U) from being intercepted by the eavesdropper (E).

[0012] Here, the angle of the reflector panel equipped in the intelligent reflector panel (RIS) of the UAV (U) is adjusted to an angle that maximizes the real part of the Signal-to-Noise Ratio (SNR) of the main channel, which is the path through which a signal transmitted from the source node (S) is reflected from the intelligent reflector panel (RIS) of the UAV (U) and transmitted to the destination node (D).

[0013] In addition, the angle of the reflective panel equipped in the intelligent reflective panel (RIS) of the above UAV (U) is the security channel capacity ( ), (Here, is the main channel capacity, It can be adjusted to an angle that maximizes the (indicating the eavesdropping channel capacity).

[0014] Meanwhile, the above security channel capacity ( ) is predicted through transfer learning equipped with a source model that performs learning by having an input layer, multiple hidden layers, and an output layer; and a target model that receives parameters learned from the source model, performs learning with new environmental data, and outputs a result, wherein the parameters of the frozen layer and the fine-tune layer of the target model are copied from the source model, the input layer and the frozen layer are set with fixed parameters and do not participate in backpropagation during target model training, and the new layer and the fine-tune layer are trained with new environmental data to predict the secure channel capacity ( It can output predicted values. Effects of the invention

[0016] According to the present invention, by adjusting the angle of an intelligent reflective panel (RIS) present in a mobile network to maximize the real coefficient of the main channel efficiency and the security channel capacity, physical layer security performance can be improved in a mobile network where an eavesdropper (E) is present and nodes are moving.

[0017] In addition, the present invention has the effect of measuring the physical layer security performance of a network without implementing an actual network or going through a mathematical proof process by using a learned artificial intelligence model to measure the main channel capacity and the eavesdropping channel capacity in various node movement models. Brief explanation of the drawing

[0019] FIG. 1 is a conceptual diagram of a mobile network having an intelligent reflective panel according to the present invention. FIG. 2 is an example of a movement path in which a legitimate user moves to an RPG movement model and an eavesdropper (E) moves to an RPG movement model according to the present invention. FIG. 3 is an example of a movement path in which, according to the present invention, a legitimate user follows an RPG movement model and an eavesdropper (E) follows an RWM model. FIG. 4 is an example of a movement pattern in which an eavesdropper (E) moves differently from legitimate users according to a GMM model in accordance with the present invention. FIG. 5 is an example of a search algorithm executed to adjust the angle of a reflection panel to maximize the real part of the SNR of the main channel according to the present invention. FIG. 6 is an example of an algorithm executed to adjust the angle of a reflective panel using the OSP method according to the present invention. FIG. 7 is a conceptual diagram of transfer learning according to the present invention, FIG. 8 is an example of an algorithm executed for transfer learning according to the present invention. FIG. 9 is an example of a simulation parameter for verifying a learning result according to the present invention. Figure 10 is an example of the change in accuracy according to the number of training iterations by performing training according to the simulation parameters of Figure 9. FIG. 11 is an example of the result of comparing the average security channel capacity when a UAV (U) has four reflective panels according to the present invention and an eavesdropper (E) moves as an RPG model. FIG. 12 is an example of a security channel capacity when an eavesdropper (E) moves according to an RPG movement model and the OSP method is applied to the RIS according to the present invention. FIG. 13 is an example of a secure channel capacity when an eavesdropper (E) moves according to an RWP model and the OSP method is applied to the RIS according to the present invention. FIG. 14 is an example of a secure channel capacity when an eavesdropper (E) moves according to a GMM model and the OSP method is applied to the RIS according to the present invention. Figure 15 shows the effect of the number of learning layers on the inference result when performing transfer learning according to the present invention. Specific details for implementing the invention

[0020] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings.

[0022] Figure 1 shows a conceptual diagram of a mobile network having an intelligent reflective panel according to an embodiment of the present invention.

[0023] As illustrated in FIG. 1, in a mobile network having an intelligent reflective panel according to the present invention, there is a source node (S), a destination node (D), an eavesdropper (E), and a UAV (Unmanned Aerial Vehicle; U) equipped with an intelligent reflective panel (RIS). The intelligent reflective panel (RIS) equipped on the UAV (U) is provided with R reflective panels.

[0024] Here, the source node (S) cannot communicate directly with the destination node (D) or the eavesdropper (E) due to obstacles. Therefore, the source node (S) and the destination node (D) can transmit signals through the intelligent reflective panel (RIS) mounted on the UAV (U), and the eavesdropper (E) can only eavesdrop on the signals reflected through the intelligent reflective panel (RIS).

[0025] The intelligent reflective panel (RIS) mounted on the above UAV (U) has R reflective panels, and the angles of each reflective panel can be represented in the form of a matrix as shown in the following mathematical formula 1.

[0026]

[0027] Here, represents the Amplitude Reflection Coefficient of the r-th reflection panel, and represents the angle of the r-th reflective panel. In this invention, it is assumed that all reflective panels have the same amplification and reflection efficiency because they are all of the same type. Additionally, the range of the angles of the reflective panels is am.

[0028] Therefore, the signal transmitted from the source node (S), reflected through the intelligent reflection panel (RIS), and received at the destination node (D) can be represented as Equation 2.

[0029]

[0030] Here, is the transmission power of the source node (S), Large-scale fading between the UAV (U) and the destination node (D), is the channel efficiency (Compelx Channel Coefficient) between the r-th reflector panel mounted on the UAV (U) and the destination node (D), Large-scale Fading between source node (S) and UAV (U), is the channel efficiency between the source node (S) and the r-th reflector panel mounted on the UAV (U), is the message to be conveyed, represents channel noise. Channel noise is Additive White Gaussian Noise (AWGN), with a mean of 0 and a variance of . am.

[0031] Large-scale fading between two nodes in the above mathematical equation 2 can be expressed as the following mathematical equation 3.

[0032]

[0033] Here, is the frequency band of the carrier wave, is the speed of light, is the Path Loss Exponent, represents the standard deviation of Shadow Fading. is the distance between two nodes and can be expressed as in mathematical equation 4.

[0034]

[0036] In the above mathematical formula 2, channel efficiency ( , When expressed in terms of the polar plane, it is as follows in Equations 5 and 6.

[0037]

[0038]

[0039] Here, and represents the channel efficiency size of each link, and and represents an angle.

[0040] Therefore, the signal-to-noise ratio (SNR) model of the signal received at the destination node (D) is as follows in Equation 7.

[0041]

[0042] Here, It refers to the average signal-to-noise ratio.

[0043] Therefore, the channel capacity of the main channel, which is the path from the source node (S) to the destination node (D) via the UAV (U), is given by the following mathematical formula 8.

[0044]

[0046] Similar to the main channel, the signal overheard by the eavesdropper (E) can be represented as shown in the following mathematical formula 9.

[0047]

[0048] Here, Large-Scale Fading between UAV(U) and eavesdropper(E), is the complex channel coefficient between the r-th reflective panel of the UAV (U) and the eavesdropper (E). Large-Scale Fading between source node (S) and UAV (U), is the complex channel coefficient between the r-th reflection panel of the UAV (U) and the source node (S). represents the noise of the eavesdropping channel. Since the eavesdropping channel noise is in the same environment as the main channel, it follows additional white Gaussian noise, with a mean of 0 and a variance of am.

[0049] Therefore, the SNR model of the eavesdropping channel, which is the path from the source node (S) to the eavesdropper (E) via the UAV (U), is as follows in Equation 10.

[0050]

[0051] Here, It refers to the average SNR at the eavesdropper (E). In the present invention, in order to maintain consistency It is assumed that...

[0052] Therefore, the channel capacity of the eavesdropping channel is equal to the following mathematical formula 11.

[0053]

[0055] In this invention, the performance of the system is evaluated using Instantaneous Secrecy Capacity (ISC) and Average Secrecy Rate (ASR). Instantaneous Secrecy Capacity ( ) and Average Security Channel Capacity (ASR) are given by the following Equations 12 and 13.

[0056]

[0057]

[0059] Mobility Models:

[0060] In this invention, to improve physical layer security performance even when nodes are moving in a mobile network, the angle of an intelligent reflective panel (RIS) is adjusted, and the physical layer security performance changing according to the angle adjustment of the intelligent reflective panel (RIS) is measured using an artificial intelligence model. To this end, in this invention, the movement of each node moving in a mobile network is defined as follows.

[0061] 1) Legitimate nodes : Source node (S), destination node (D), UAV (U) equipped with intelligent reflective panel (RIS)

[0062] All of these nodes move similarly. This can be represented by the RPG (Reference Point Group) Mobility Model. Therefore, if the movements of legitimate users at time t are plotted on the x / y plane, they can be expressed as shown in the following Equations 14 and 15.

[0063]

[0064] Here, is the position at time (t-1), represents the speed of the node, and indicates the direction of movement of the node.

[0065]

[0066] Here, is the position at time (t-1), represents the speed of the node, and indicates the direction of movement of the node.

[0067] And the group's orientation and velocity follow the Gauss-Markov Mobility Model. Therefore, the group's orientation ( ) and speed( ) is equal to the following mathematical formulas 16 and 17.

[0068]

[0069] Here, is memory level, It follows a Gaussian distribution at the rate of legitimate users. The mean of that distribution is , variance is Follows.

[0070]

[0071] Here, is memory level, It follows a Gaussian distribution in the direction of legitimate users. The mean of the distribution is , variance is Follows.

[0072] Therefore, the speed and direction of each node within the group can be expressed as Equations 18 and 19.

[0073]

[0074]

[0075] Here, is the velocity variation of each node, represents the directional deviation of each node. And these deviations follow a uniform distribution.

[0076] Consequently, when the legitimate nodes follow the RPG movement model, the positions of each node are as shown in the following mathematical formulas 20 and 21.

[0077]

[0078]

[0080] 2) eavesdropper (E)

[0081] The movement model of the eavesdropper (E) has different effects. Therefore, to determine the effects of ① when the eavesdropper (E) moves together with legitimate users and ② when the eavesdropper (E) moves differently from legitimate users, the movement of the eavesdropper (E) is defined as follows.

[0082] ① The movement of the eavesdropper (E) RPG If following the movement model:

[0083] In this case, since the eavesdropper (E) moves together with the legitimate users, it follows the RPG movement model, which is the movement of the legitimate users explained earlier. Therefore, the speed of the eavesdropper (E) ( ) and direction( ) can be expressed by the following mathematical formulas 22 and 23.

[0084]

[0085] Here, is the maximum movement speed of the eavesdropper (E), The group leader's movement speed, represents the deviation in movement speed.

[0086]

[0087] Here, is the group leader's direction of movement, represents the deviation in direction.

[0088] Consequently, if the eavesdropper (E) follows the RPG movement model, the position (x-coordinate and y-coordinate) of the eavesdropper (E) can be expressed as Equation 24 and Equation 25.

[0089]

[0090]

[0091] Figure 2 shows an example of a movement path in which a legitimate user moves to an RPG movement model, and an eavesdropper (E) also moves to an RPG movement model.

[0093] ② Eavesdropper (E) of movement RWM (Random Walk Mobility) In the case of a model :

[0094] When an eavesdropper (E) follows their own movement model instead of the RPG movement model, which is the movement model of a legitimate user, the RWM model randomly selects a speed and direction from a uniform distribution. The uniform distribution for selecting the speed and direction is given by the following mathematical equations 26 and 27.

[0095]

[0096] Here, is the minimum value of the uniform velocity distribution, represents the maximum value of the uniform velocity distribution.

[0097]

[0098] Here, an angle smaller than 0 degrees and 2π (360 degrees) is arbitrarily selected.

[0099] Therefore, if the eavesdropper (E) follows the RWM model, the position (x-coordinate, y-coordinate) of the eavesdropper (E) can be expressed as shown in the following mathematical equations 28 and 29.

[0100]

[0101]

[0102] Figure 3 shows an example of a movement path where a legitimate user follows the RPG movement model and an eavesdropper (E) follows the RWM model.

[0104] ③ Eavesdropper(E) of movement GMM (Gauss- Markov If following the Mobility) model:

[0105] When an eavesdropper (E) follows their own movement model instead of following the RPG movement model, which is the movement model of a legitimate user, and follows the Gauss-Markov Mobility (GMM) model, which determines future movement speed and direction based on current speed and direction, the next movement speed and direction can be represented by Equation 30 and Equation 31.

[0106]

[0107]

[0108] Here, represents the memory level for the movement of the eavesdropper (E) according to Markov characteristics. is a Gaussian distribution based on the movement speed of the eavesdropper (E), and the mean is , variance is am.

[0109] Consequently, the x / y coordinates of the eavesdropper (E) can be expressed by the following mathematical formulas 32 and 33.

[0110]

[0111]

[0112] Figure 4 shows an example of a movement pattern in which an eavesdropper (E) moves differently from legitimate users according to the GMM model.

[0114] The present invention proposes a method to improve security performance by adjusting the angles of the reflective panels of an intelligent reflective panel (RIS).

[0115] To compare the performance of the methods proposed in this invention, a performance evaluation was performed using the RPS (Random Phase Shift) algorithm. The RPS algorithm randomly adjusts the angle of the reflective panel, which can be mathematically expressed as Equation 34 below.

[0116]

[0117] When the angle is adjusted using the above RPS method, the SNR of the main channel and the eavesdropping channel is as shown in the following mathematical formulas 35 and 36.

[0118]

[0119]

[0121] In the first embodiment of the present invention, the angle of the reflective panel to improve security performance is adjusted to an angle that maximizes the real part of the SNR of the main channel. The present invention aims to maximize the security channel capacity (SC) by considering the quantized phase shift due to high phase shift resolution and hardware limitations, and the phase shift configuration in the UAV (U) is modeled as linear quantization of the Q level. This can be mathematically expressed as the following Equation 37.

[0122]

[0123] Here, represents the real part of the main channel's SNR. represents the angle range of the reflection panel. A search algorithm to find this is shown in Fig. 5.

[0125] In a second embodiment of the present invention, the angle of the reflective panel is adjusted to an angle that maximizes the security channel capacity. In the present invention, this is referred to as the OPS (Optimal Secrecy-Oriented Phase Shift) method. Mathematically, this is expressed as the following Equation 38.

[0126]

[0127] The SNR of the main channel and the listening channel when the angle is adjusted using the above OSP method is given by Equations 39 and 40.

[0128]

[0129]

[0130] The search algorithm for finding this is shown in Fig. 6.

[0132] In this invention, there are limitations to analyzing security performance using mathematical models because legitimate users and eavesdroppers (E) are involved. Therefore, this invention proposes a method to verify security performance using Transfer Learning, an artificial intelligence method. Transfer Learning involves adding additional layers to an existing learned model for training. Since it utilizes knowledge from the existing model, training is possible even with limited data, demonstrating superior performance compared to retraining the existing model. The model structure for Transfer Learning consists of a source model and a target model, which is similar to existing machine learning. In the Transfer Learning method, the source model corresponds to an already trained model, while the target model is a newly trained model.

[0133] Figure 7 shows a conceptual diagram of transfer learning according to an embodiment of the present invention.

[0134] As shown in Fig. 7, in the transfer learning method There are a Source Model and a Target Model for predicting (security channel capacity). The Source Model is an already trained model, typically a Deep Neural Network (DNN) with multiple hidden layers. On the other hand, the Target Model, where transfer learning takes place, consists of a portion of the Source Model and new hidden layers. In this case, the Source Model portion is not trained, while the Target Model portion is. That is, the input layer and the frozen layer within the Source Model are set with fixed parameters and do not participate in backpropagation during Target Model training. Additionally, the new hidden layer is positioned before the output layer. Along with the new layer, the fine-tuned layer is fine-tuned using new environment data. Fine-tuning aims to train only a portion of the Source Model and the new hidden layer using a small number of labeled samples from the target environment. Since fine-tuning learns parameters only from a limited number of layers, it can significantly reduce training costs. However, since it uses an already trained source model, the learning effect is superior to existing learning methods. The transfer learning approach proposed in this invention considers the use of a fine-tuned model. In this case, optimal parameters from the trained source model are transferred to the target model, thereby enhancing new training even if the target model was trained with only an insufficient amount of data.

[0135] As such, the transfer learning proposed in this invention increases the efficiency of learning by utilizing existing DNN models while adding new layers. Through this, significant performance improvement can be expected even with small amounts of training data.

[0136] The input data for transfer learning is as shown in the following mathematical equation 41.

[0137]

[0138] Here, is the transmission power to noise ratio, is the 3D coordinates of the source node, is the 3D coordinate of the UAV(U) equipped with RIS, is the 3D coordinate of the destination node (D), is the 3D coordinates of the eavesdropper (E), is the distance between the source node and the UAV(U), is the distance between the UAV (U) and the destination node (D), is the distance between the UAV (U) and the eavesdropper (E), is the angle between the source node and the R-th reflection panel of the RIS, is the angle between the R-th reflection panel of the RIS and the destination node (D), is the angle between the R-th reflective panel of the RIS and the eavesdropper (E), is the real part of the channel efficiency between the source node and the R-th reflection panel of the RIS, is the imaginary part of the channel efficiency between the source node and the R-th reflection panel of the RIS, is the real part of the channel efficiency between the R-th reflection panel of the RIS and the destination node (D), is the imaginary part of the channel efficiency between the R-th reflection panel of the RIS and the destination node (D), is the real part of the channel efficiency between the R-th reflective panel of the RIS and the eavesdropper (E), represents the imaginary part of the channel efficiency between the R-th reflection panel of the RIS and the eavesdropper (E).

[0139] Because the source model for transfer learning has all units in each layer connected to each other, the output at the l-th layer ( ) can be expressed as shown in the following mathematical formula 42.

[0140]

[0141] Here, represents the activation function of the L-th layer, and represents the output of the (L-1)th layer, i.e., the input of the Lth layer, and is the weight of the L-th layer, represents the bias of the L-th layer. In this invention, the activation function used is the ReLU (Rectified Linear Unit) function. This can be expressed mathematically as the following Equation 43.

[0142]

[0143] Through this, the source model infers the results Let's assume that. Through transfer learning, the source model updates the weights and biases between each layer, and the inference result of the source model ( ) and actual value( Reduces the error with ). At this time, RMSE (Root Mean Square Error) is used as a method to measure the error. Mathematically, the error in the k-th training is expressed as the following mathematical equation 44.

[0144]

[0145] This series of processes for transfer learning proceeds as shown in the algorithm of Fig. 8.

[0147] FIG. 9 is an example of simulation parameters for verifying the learning results according to an embodiment of the present invention, and FIG. 10 shows an example of the change in accuracy according to the number of training iterations by performing training according to the simulation parameters of FIG. 9. As shown in FIG. 10, it can be seen that the difference between the actual value and the inference result gradually decreases as the number of training iterations (Epoch) of the transfer learning model increases. In addition, it can be seen that the error of the transfer learning model is smaller than that of the source model training.

[0148] FIG. 11 is an example of the result of comparing the average security channel capacity when a UAV (U) has four reflective panels and an eavesdropper (E) moves with an RPG model according to an embodiment of the present invention. As shown in FIG. 11, it can be seen that the average security channel increases as the number of angles for adjusting the reflective panels increases. In addition, it can be seen that the simulation results and the average security channel capacity using a transfer learning model are almost identical.

[0149] FIG. 12 shows the secure channel capacity when an eavesdropper (E) moves according to an RPG movement model and the OSP method is applied to the RIS according to an embodiment of the present invention, FIG. 13 shows the secure channel capacity when an eavesdropper (E) moves according to an RWP model and the OSP method is applied to the RIS, and FIG. 14 shows the secure channel capacity when an eavesdropper (E) moves according to a GMM model and the OSP method is applied to the RIS. As shown in FIG. 12 to FIG. 14, the source model and the target model can closely predict the immediate secure channel capacity at 200 different locations indicated using various movement models, because the source model and the target model can learn the characteristics of node locations and channel coefficients during the training process. The average of the predictions by the source model and the target model has a similar pattern to the simulation average, which means that the source and target models successfully capture the characteristic features of the input data during training.

[0150] Figure 15 illustrates the effect of the number of learning layers on the inference result when performing transfer learning according to an embodiment of the present invention. As shown in Figure 15, it can be seen that the inference result of the average security channel capacity decreases as existing layers are newly learned. Through this, it can be seen that accuracy decreases as new learning is performed without utilizing the knowledge of the existing model.

[0152] In this way, the present invention can improve physical layer security performance by adjusting the angle of the intelligent reflective panel of the UAV (U) to maximize the real part of the SNR of the main channel or to maximize the security channel capacity. In addition, the physical layer security performance of a mobile network can be measured by predicting the security channel capacity in various node movement models using a trained artificial intelligence model.

[0153] The present invention is not limited to the embodiments described above, and it is obvious that various modifications and variations may be made by those skilled in the art within the scope of the technical concept of the present invention and the equivalent scope of the claims described below.

Claims

Claim 1 A method for adjusting the angle of an intelligent reflective panel, characterized in that, in a mobile network comprising a source node (S), a destination node (D), an eavesdropper (E), and a UAV (Unmanned Aerial Vehicle) (U) equipped with an intelligent reflective panel (RIS) having R reflective panels, a signal transmitted from the source node (S) is reflected by the intelligent reflective panel (RIS) of the UAV (U) and transmitted to the destination node (D), wherein the UAV (U) adjusts the angle of the reflective panel equipped in the intelligent reflective panel (RIS) to improve physical layer security performance, thereby preventing the signal reflected from the UAV (U) from being intercepted by the eavesdropper (E). Claim 2 A method for adjusting the angle of an intelligent reflective panel according to claim 1, wherein the angle of the reflective panel provided in the intelligent reflective panel (RIS) of the UAV (U) is adjusted to an angle that maximizes the real part of the SNR (Signal-to-noise Ratio) of the main channel, which is the path through which a signal transmitted from the source node (S) is reflected from the intelligent reflective panel (RIS) of the UAV (U) and transmitted to the destination node (D). Claim 3 In Clause 2, the SNR of the main channel is a mathematical formula (Here, average signal-to-noise ratio ( is the transmission power of the source node (S), represents the variance of the additional white Gaussian noise, which is the main channel noise, and Large-scale fading between source node (S) and UAV (U), Large-scale Fading between UAV(U) and destination node(D), is the amplitude reflection coefficient of the reflective panel, is the channel efficiency size between the source node (S) and the r-th reflector panel mounted on the UAV (U), is the channel efficiency size between the r-th reflector panel mounted on the UAV (U) and the destination node (D), is the angle of the r-th reflection panel, is the angle between the source node (S) and the r-th reflector panel mounted on the UAV (U), represents the angle between the r-th reflector panel mounted on the UAV (U) and the destination node (D), and the angle that maximizes the real part of the SNR of the main channel is the mathematical formula (Here, represents the real part of the main channel's SNR, and An intelligent reflective panel angle adjustment method characterized by (indicating the angle range of the reflective panel). Claim 4 In claim 1, the angle of the reflective panel equipped in the intelligent reflective panel (RIS) of the UAV (U) is the security channel capacity ( ), (Here, is the main channel capacity, An intelligent reflective panel angle adjustment method characterized by being adjusted to an angle that maximizes (which represents the eavesdropping channel capacity). Claim 5 In Clause 4, the above security channel capacity ( The angle that maximizes ) ) is a mathematical expression (Here, main channel capacity And, is the main channel's SNR (Signal-to-noise Ratio). average signal-to-noise ratio ( is the transmission power of the source node (S), represents the variance of the additional white Gaussian noise, which is the main channel noise, and Large-scale fading between source node (S) and UAV (U), Large-scale Fading between UAV(U) and destination node(D), is the amplitude reflection coefficient of the reflective panel, is the channel efficiency size between the source node (S) and the r-th reflector panel mounted on the UAV (U), is the channel efficiency size between the r-th reflector panel mounted on the UAV (U) and the destination node (D), is the angle of the r-th reflection panel, is the angle between the source node (S) and the r-th reflector panel mounted on the UAV (U), represents the angle between the r-th reflective panel mounted on the UAV (U) and the destination node (D), and the eavesdropping channel capacity as, Is average signal-to-noise ratio ( is the transmission power of the source node (S), represents the variance of additional white Gaussian noise, which is eavesdropping channel noise, and Large-scale fading between source node (S) and UAV (U), Large-Scale Fading between UAV(U) and eavesdropper(E), is the channel efficiency size between the source node (S) and the r-th reflector panel mounted on the UAV (U), is the channel efficiency size between the r-th reflective panel of the UAV (U) and the eavesdropper (E), is the angle of the r-th reflection panel, is the angle between the source node (S) and the r-th reflector panel mounted on the UAV (U), represents the angle between the r-th reflective panel mounted on the UAV (U) and the eavesdropper (E), An intelligent reflective panel angle adjustment method characterized by being obtained through (which indicates the angle range of the reflective panel). Claim 6 In claim 5, the angle of the reflective panel equipped in the intelligent reflective panel (RIS) of the above UAV(U) is the security channel capacity ( When adjusting to the angle that maximizes ), the main channel's SNR( ) is a mathematical expression It is calculated as, and the SNR of the eavesdropping channel( ) is a mathematical expression An intelligent reflective panel angle adjustment method characterized by being calculated as follows. Claim 7 In Clause 4, the above security channel capacity ( ) is predicted through transfer learning equipped with a source model that performs learning by having an input layer, multiple hidden layers, and an output layer; and a target model that receives parameters learned from the source model, performs learning with new environmental data, and outputs a result, wherein the parameters of the frozen layer and the fine-tune layer of the target model are copied from the source model, the input layer and the frozen layer are set with fixed parameters and do not participate in backpropagation during target model training, and the new layer and the fine-tune layer are trained with new environmental data to predict the secure channel capacity ( An intelligent reflective panel angle adjustment method characterized by outputting a predicted value. Claim 8 In claim 7, the input data (x) input to the source model for the transfer learning is (Here, is the transmission power to noise ratio, is the 3D coordinates of the source node, is the 3D coordinate of the UAV(U) equipped with RIS, is the 3D coordinate of the destination node (D), is the 3D coordinates of the eavesdropper (E), is the distance between the source node and the UAV(U), is the distance between the UAV (U) and the destination node (D), is the distance between the UAV (U) and the eavesdropper (E), is the angle between the source node and the R-th reflection panel of the RIS, is the angle between the R-th reflection panel of the RIS and the destination node (D), is the angle between the R-th reflective panel of the RIS and the eavesdropper (E), is the real part of the channel efficiency between the source node and the R-th reflection panel of the RIS, is the imaginary part of the channel efficiency between the source node and the R-th reflection panel of the RIS, is the real part of the channel efficiency between the R-th reflection panel of the RIS and the destination node (D), is the imaginary part of the channel efficiency between the R-th reflection panel of the RIS and the destination node (D), is the real part of the channel efficiency between the R-th reflective panel of the RIS and the eavesdropper (E), represents the imaginary part of the channel efficiency between the R-th reflection panel of the RIS and the eavesdropper (E), and the output of the l-th layer among the multiple hidden layers of the source model for transfer learning ( )silver (Here, is the activation function of the L-th layer, is the output of the (L-1)th layer (input of the Lth layer), is the weight of the L-th layer, represents the bias of the L-th layer), and the above activation function is the ReLU(Rectified Linear Unit) function( ) and the output layer of the above source model is the inference result An intelligent reflective panel angle adjustment method characterized by outputting Claim 9 In Clause 8, through the above transfer learning, the source model updates the weights and biases between each layer, and the inference result of the source model ( ) and actual value( Reduce the error with ), using RMSE (Root Mean Square Error) for the above error measurement, and the error in the k-th training is (Here, An intelligent reflective panel angle adjustment method characterized by being measured as (indicating the number of data points in a test data set).

Images