Intelligent beamforming method of STAR-RIS auxiliary security communication and sensing integrated system

Abstract

The invention discloses an intelligent beam forming method of an STAR-RIS assisted security communication integrated system, and relates to the technical field of communication, and the method comprises the following steps: S1, building a wireless communication scene comprising a base station, STAR-RIS, a legal user and an eavesdropper; s2, constructing a communication model and a sensing model of the STAR-RIS auxiliary security communication and sensing integrated system; s3, establishing a utility maximization problem of a corresponding system; s4, modeling the beam forming into a Markov decision process; s5, performing joint optimization on the transmission beam forming of the base station and the transmission and reflection coefficient matrix of the STAR-RIS by adopting a noise self-adaptive deep reinforcement learning algorithm; the STAR-RIS used in the invention adopts an energy segmentation protocol, and compared with a traditional passive RIS, the STAR-RIS can perform reflection transmission and transmission transmission on an incident signal, thereby realizing full-space coverage.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to an intelligent beamforming method of a STAR-RIS auxiliary safety interawareness integrated system. Background Art

[0002] With the development of 6G technology, many emerging applications have emerged, such as the Internet of Things, V2X communication, unmanned driving, smart homes, and drone networks. These functions all rely on accurate perception of the environment. Therefore, the sensing function is expected to become one of the basic services of the 6G network and become the next stage of development beyond the existing pure communication solution. At the same time, the combined use of millimeter wave frequencies and massive multiple input and multiple output (MIMO) technology makes the communication system and the radio sensing system have amazing similarities in hardware structure, channel characteristics, and information processing pipelines. Due to the shared use of hardware and wireless resources, radar and communication systems are also moving towards integration. For this reason, Integrated Sensing And Communication (ISAC) is considered an important driving factor and has attracted widespread attention.

[0003] Due to the broadcast and open nature of wireless signals, information in the synaesthesia integrated system is vulnerable to eavesdropping. As a key technology, Reconfigurable Intelligent Surface (RIS) can intelligently control the propagation environment, and is therefore an important way to achieve information security from the perspective of physical layer security by reconstructing the eavesdropping channel. In practical applications, RIS can provide signal enhancement for users on the same side of the BS by reflecting the received signal. However, for users on the other side, the signal will become weaker due to signal attenuation and scattering. To meet this challenge, some scholars have proposed a simultaneously transmissive and reflective RIS (STAR-RIS) to achieve 360° full coverage around the RIS panel.

[0004] For example, a Chinese patent with publication number CN118573236A discloses a sum rate optimization method for a STAR-RIS assisted synaesthesia integrated transmission system. The invention designs a sum rate optimization method for a STAR RIS assisted synaesthesia integrated transmission system, which maximizes the sum rate of the system by jointly adjusting the base station precoding matrix and the STAR-RIS phase shift matrix. However, the invention does not consider the important performance indicator of security. In addition, the invention adopts an alternating optimization algorithm for solving, and a series of sub-problems need to be solved in each optimization iteration, which may cause the algorithm complexity to be too high. In summary, the present invention designs an intelligent beamforming method for a STAR-RIS assisted security synaesthesia integrated system, uses a deep reinforcement learning algorithm, and studies a beamforming strategy based on overall system performance optimization in view of the presence of eavesdropping and optimized energy consumption in the system. Summary of the invention

[0005] The present invention aims to solve the above problems of the prior art. A smart beamforming method for a STAR-RIS assisted safety synaesthesia integrated system is proposed. The technical solution of the present invention is as follows:

[0006] An intelligent beamforming method for a STAR-RIS assisted safety synaesthesia integrated system comprises the following steps:

[0007] S1: Building a wireless communication scenario, which includes a base station, STAR-RIS, a legitimate user, and an eavesdropper;

[0008] S2: Construct the communication model and perception model of the STAR-RIS assisted safety synaesthesia integrated system;

[0009] S3: Based on the communication model and perception model, construct the utility maximization problem of the system;

[0010] S4: Modeling beamforming as a Markov decision process;

[0011] S5: A noise-adaptive deep reinforcement learning algorithm is used to jointly optimize the base station’s transmit beamforming and the STAR-RIS transmission-reflection coefficient matrix.

[0012] Furthermore, the wireless communication scenario in step S1 consists of a telepathy integrated base station, a STAR-RIS, T sensing targets, K users and an eavesdropper; the STAR-RIS adopts an energy splitting mode, that is, each element divides the incident signal into a reflected signal of a transmitted signal and transmits them simultaneously; the telepathy integrated base station is configured with M transmitting / receiving antennas; the STAR-RIS is configured with N elements, and each of the communication user and the eavesdropper is configured with a single antenna.

[0013] Furthermore, the communication model and perception model of the STAR-RIS assisted safety synaesthesia integration system in step 2 include:

[0014] Transmission and reflection coefficient matrices for STAR-RIS:

[0015]

[0016] β t,n ,θ t,n , β r,n ,θ r,n They represent the amplitude and phase of the transmission and reflection coefficient of the nth element of STAR-RIS respectively.

[0017] The echo signal vector received by the base station is expressed as follows:

[0018]

[0019] in represents the transmission beamforming matrix of the base station to the communication user, represents a dedicated sensing signal beamforming matrix of a base station; Indicates the information transmitted by the base station to the communication user. Indicates the signal sent by the base station to the sensing target; represents the noise of the echo signal, σ r 2 is the noise power; They represent the channel matrices from BS to target, STAR-RIS to target, and BS to STAR-RIS, respectively. T represents transpose; Φ r represents the reflection phase shift matrix of STAR-RIS;

[0020] The specific formula for the received signal of users in the transmission area and the reflection area is:

[0021]

[0022] in is the noise power, Φ i Represents the STAR-RIS transmission or reflection phase shift matrix. represents the channel from BS to the kth user and from STAR-RIS to the kth user;

[0023] The eavesdropper's received signal:

[0024]

[0025] in is the noise power, They represent the channels from BS to eavesdropper and from STAR-RIS to eavesdropper respectively;

[0026] The perceived signal-to-noise ratio of the base station is:

[0027]

[0028] in, (·) H represents conjugate transpose;

[0029] The signal-to-interference-to-noise ratio of users in the transmission area and the reflection area is as follows:

[0030]

[0031] Among them, w k represents the beamforming vector transmitted by the base station to user k, w j represents the beamforming vector transmitted by the base station to other users;

[0032] Calculate the eavesdropper's signal-to-interference-to-noise ratio. The specific formula is:

[0033]

[0034] The communication rate of user k is as follows: R k =log(1+γ k ), the eavesdropper’s communication rate, the specific formula is: R e,k =log(1+γ e,k ), the secure communication rate of user k, the specific formula is: in[·] + Indicates that only values ​​greater than 0 are meaningful; the total secure communication rate of the system, the specific formula is:

[0035]

[0036] System utility, the specific formula is:

[0037]

[0038] Where P is the base station transmit power, a represents the weight of the perceived performance, b represents the weight of the security performance, and c represents the weight of the power.

[0039] Furthermore, the utility maximization problem in step S3 is established as follows: taking the perceived signal-to-noise ratio of the base station and the secure communication rate of the legitimate user as constraints, the system utility of the synaesthesia integrated system is maximized by jointly optimizing the STAR-RIS phase shift matrix and the base station beamforming matrix, as follows:

[0040]

[0041]

[0042] C3:P≤P max

[0043]

[0044] Where W represents the precoding matrix of the base station, Φ t and Φ r represents the transmission and reflection surface coefficient matrix of STAR-RIS, R th They represent the secure communication rate and secure communication rate threshold of the kth user, γ t , γ th They represent the echo signal-to-noise ratio and echo signal-to-noise ratio threshold of the t-th perceived target, θ i,n Represents the phase of the transmission and reflection surface coefficients. Constraint C1 can ensure that the secure communication rate of all users is not lower than a threshold, ensuring user privacy. Constraint C2 can ensure that the signal-to-noise ratio of the echo signal of the perceived target is not lower than a threshold. Constraint C3 ensures that the transmission power of the base station does not exceed the upper limit. Constraint C4 indicates that the sum of the signal powers of the STAR-RIS transmission surface and the reflection surface will not exceed the power of the incident signal. Constraint C5 represents the phase limit of the STAR-RIS unit.

[0045] Furthermore, the process of jointly optimizing the transmit beamforming of the base station and the transmission reflection coefficient matrix of STAR-RIS using the noise adaptive deep reinforcement learning algorithm in step S4 includes the following steps:

[0046] S41: Define the system state space, action space and reward function, including:

[0047] The state space is specifically defined as:

[0048]

[0049] in, represents the square of the L2 norm of each column of the precoding matrix W at the y-th time step, is the secure communication rate between each user and the eavesdropper at the yth time step, is the echo signal-to-noise ratio of each perceived target at the yth time step, and the dimension of the state space is M+2K+T;

[0050] The action space is specifically defined as:

[0051]

[0052] in is the unprocessed base station precoding matrix output by the neural network at the yth time step, are the emission and reflection coefficients of STAR-RIS at the y-th time step, Re{·} means taking the real part, Im{·} means taking the imaginary part, δ is the power factor at the y-th time step, and the dimension of the action space is 4N+2M×(M+K)+1;

[0053] The reward function is specifically defined as:

[0054]

[0055] Among them, ξ is the penalty factor;

[0056] S42: Construct adaptive action noise and define action control strategies, including:

[0057] The maximum and minimum normalized cumulative rewards of x rounds in the current training {R e-x ,...,R e-1}, the specific formula is: Noise scale factor, the specific formula is: Among them, var(·) means to find the variance;

[0058] Define the action control strategy, the specific formula is:

[0059]

[0060] Among them, a ε is an action with noise, a is a deterministic action, ε is the noise sampled from a standard normal distribution, and β is the scaling factor of the noise.

[0061] Furthermore, in step S5, a noise-adaptive deep reinforcement learning algorithm is used to jointly optimize the transmit beamforming of the base station and the transmission reflection coefficient matrix of STAR-RIS, including:

[0062] Use the policy network to select actions, get rewards, transfer to the state of the next moment, and store the experience tuple in the experience replay pool; sample from the experience replay pool and update the value network of the agent; use the policy gradient to update the policy network based on the Q value generated by the value network; soft update the parameters of the policy target network and the value target network; update the noise scale factor at the end of each round.

[0063] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the intelligent beamforming method of the STAR-RIS auxiliary safety synaesthesia integration system as described in any one of the items is implemented.

[0064] A non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the intelligent beamforming method of the STAR-RIS assisted safety synaesthesia integrated system as described in any one of the items.

[0065] The advantages and beneficial effects of the present invention are as follows:

[0066] The present invention proposes a STAR-RIS-assisted security interawareness integrated system and an intelligent beamforming method. The steps shown in claim 1 meet the security of the ISAC system assisted by STAR-RIS; in claim 3, the existence of eavesdroppers is taken into account, and the utility function in microeconomics is used to measure the comprehensive performance of the proposed system in terms of security, perception, and energy consumption, that is, the system utility; in claim 5, a more flexible action space is designed to improve the correct execution of the intelligent agent's actions, and adaptive action noise is proposed, so that better strategies can be learned by better balancing the exploration and utilization of the intelligent agent. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a schematic diagram of the structure of the system model of the preferred embodiment provided by the present invention;

[0068] Figure 2 : Learning curves of the proposed algorithm under different learning rates;

[0069] Figure 3 : Learning curves of the proposed algorithm under different adaptive sampling windows;

[0070] Figure 4 : Relationship between different algorithm power and system utility;

[0071] Figure 5 : Relationship diagram of different algorithm power and system and safety rate;

[0072] Figure 6 : Relationship between the number of STAR-RIS units and system utility for different algorithms. DETAILED DESCRIPTION

[0073] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0074] The technical solution of the present invention to solve the above technical problems is:

[0075] The present invention will be described in detail below in conjunction with the accompanying drawings and simulation examples of the present invention.

[0076] S1: Build a wireless communication scenario including base stations, STAR-RIS, legitimate users and eavesdroppers;

[0077] S2: Construct the communication model and perception model of the STAR-RIS assisted safety synaesthesia integrated system;

[0078] S3: Establish the utility maximization problem of the corresponding system;

[0079] S4: Modeling beamforming as a Markov decision process;

[0080] S5: A noise-adaptive deep reinforcement learning algorithm is used to jointly optimize the base station’s transmit beamforming and the STAR-RIS transmission-reflection coefficient matrix;

[0081] Among them, the STAR-RIS assisted security synaesthesia integrated system in step S1 is composed of a synaesthesia integrated base station, a STAR-RIS, T sensing targets, K users and an eavesdropper; the STAR-RIS adopts an energy splitting mode, that is, each element divides the incident signal into a reflected signal of a transmitted signal and transmits them simultaneously; the synaesthesia integrated base station is configured with M transmitting / receiving frost function antennas; the STAR-RIS is configured with N elements, and the communication user and the eavesdropper are each configured with a single antenna.

[0082] The communication model and perception model of the STAR-RIS assisted safety synaesthesia integration system described in step 2 include:

[0083] Transmission and reflection coefficient matrices for STAR-RIS:

[0084]

[0085] The echo signal vector received by the base station is expressed as follows:

[0086]

[0087] in represents the transmission beamforming matrix of the base station to the communication user, represents a dedicated sensing signal beamforming matrix of a base station; Indicates the information transmitted by the base station to the communication user, Indicates the signal sent by the base station to the sensing target; represents the noise of the echo signal, is the noise power; They represent the channel matrices from BS to target, STAR-RIS to target, and BS to STAR-RIS, respectively. T represents transpose; Φr represents the reflection phase shift matrix of STAR-RIS;

[0088] The specific formula for the received signal of users in the transmission area and the reflection area is:

[0089]

[0090] in is the noise power. Φ i Represents the STAR-RIS transmission or reflection phase shift matrix. Represents the channel from BS to the k-th user and from STAR-RIS to the k-th user.

[0091] The eavesdropper's received signal:

[0092]

[0093] in is the noise power. Represents the channels from BS to eavesdropper and from STAR-RIS to eavesdropper.

[0094] The perceived signal-to-noise ratio of the base station is:

[0095]

[0096] in, (·) H represents the conjugate transpose.

[0097] The signal-to-interference-to-noise ratio of users in the transmission area and the reflection area is as follows:

[0098]

[0099] Among them, w k represents the beamforming vector transmitted by the base station to user k, w j represents the beamforming vector transmitted by the base station to other users;

[0100] Calculate the eavesdropper's signal-to-interference-to-noise ratio. The specific formula is:

[0101]

[0102] The communication rate of user k is as follows:

[0103] R k =log(1+γ k )

[0104] The communication rate of the eavesdropper is as follows:

[0105] Re,k =log(1+γ e,k )

[0106] The secure communication rate of user k is as follows:

[0107]

[0108] in[·] + Indicates that only values ​​greater than 0 are meaningful;

[0109] The total secure communication rate of the system is as follows:

[0110]

[0111] Furthermore, the utility maximization problem in step S3 is established as maximizing the system utility of the synaesthesia integration system by jointly optimizing the STAR-RIS phase shift matrix and the base station beamforming matrix, with the base station's perceived signal-to-noise ratio and the legitimate user's secure communication rate as constraints, as follows:

[0112]

[0113] C3:P≤P max

[0114]

[0115] Where W represents the precoding matrix of the base station, Φ t and Φ r represents the coefficient matrix of the transmission and reflection surfaces of STAR-RIS, P is the base station transmission power, a represents the weight of the perception performance, b represents the weight of the security performance, and c represents the weight of the power. Constraint C1 can ensure that the secure communication rate of all users is not lower than a threshold, ensuring the privacy of users. Constraint C2 can ensure that the signal-to-noise ratio of the echo signal of the perceived target is not lower than a threshold. Constraint C3 ensures that the transmission power of the base station does not exceed the upper limit. Constraint C4 indicates that the sum of the signal power of the transmission and reflection surfaces of STAR-RIS will not exceed the power of the incident signal. Constraint C5 represents the phase limit of the STAR-RIS unit.

[0116] Furthermore, the process of jointly optimizing the transmit beamforming of the base station and the transmission reflection coefficient matrix of STAR-RIS using the noise adaptive deep reinforcement learning algorithm in step S4 includes the following steps:

[0117] S41: Define the system state space, action space and reward function, including:

[0118] The state space is specifically defined as:

[0119]

[0120] in, represents the square of the L2 norm of each column of the precoding matrix W at the y-th time step, is the secure communication rate between each user and the eavesdropper at the yth time step, is the echo signal-to-noise ratio of each perceived target at the yth time step, and the dimension of the state space is M+2K+T;

[0121] The action space is specifically defined as:

[0122]

[0123] in is the unprocessed base station precoding matrix output by the neural network at the yth time step, are the emission and reflection coefficients of STAR-RIS at the y-th time step, Re{·} means taking the real part, Im{·} means taking the imaginary part, δ is the power factor at the y-th time step, and the dimension of the action space is 4N+2M×(M+K)+1;

[0124] The reward function is specifically defined as:

[0125]

[0126] Among them, ξ is the penalty factor.

[0127] S42: Construct adaptive action noise and define action control strategies, including:

[0128] The maximum and minimum normalized cumulative rewards of x rounds in the current training {R e-x ,...,R e-1}, the specific formula is:

[0129]

[0130] Noise scale factor, the specific formula is:

[0131]

[0132] Among them, var(·) means to find the variance;

[0133] Define the action control strategy, the specific formula is:

[0134]

[0135] Among them, a ε is the action with noise, a is the deterministic action, ε is the noise sampled from the standard normal distribution, and β is the scaling factor of the noise.

[0136] Furthermore, in step S5, a noise-adaptive deep reinforcement learning algorithm is used to jointly optimize the transmit beamforming of the base station and the transmission-reflection coefficient matrix of STAR-RIS, including:

[0137] Use the policy network to select actions, get rewards, transfer to the next state, and store the experience tuple in the experience replay pool; sample from the experience replay pool and update the value network of the agent; update the policy network using policy gradients based on the Q value generated by the value network; soft-update the parameters of the policy target network and the value target network. Update the noise scale factor at the end of each round.

[0138] In order to verify the superiority of the present invention, three reference schemes are set for comparison, namely: 1. DDPG algorithm based on improved action space, its action space has a power allocation factor, but it cannot adaptively adjust the action noise. 2. Action-constrained noise adaptive DDPG algorithm, which can adaptively adjust the action noise, but the action space has no power allocation factor. 3. Action-constrained DDPG algorithm, it cannot adaptively adjust the action noise, and the action space has no power allocation factor.

[0139] The simulation parameters of the system are set as follows: In terms of noise processing, the receiving noise power of the user, base station receiving signal and eavesdropper are all set to -90dBm, and the perception performance noise is set to -100dBm.

[0140] The learning rate of the Actor and Critic networks is 0.00001, the soft update coefficient is 0.1, the discount factor is 0.5, the sampling batch size is 64, the experience replay pool size is set to 10000, the total number of rounds is 3000, the step size per round is 1000, and the adaptive sampling window length is set to 25.

[0141] The position of the base station is (-35, -35), the position of STAR-RIS is (0, 0), the sensing target is randomly generated on the reflection surface of STAR-RIS, the user is randomly generated on both sides of STAR-RIS, and the eavesdropper is randomly generated on the reflection surface of STAR-RIS.

[0142] Large-scale fading is modeled as ρ0d -α, where ρ0 = 30 dB, d represents the distance between two points, and α represents the path loss index. The path loss index from base station to STAR-RIS, base station to eavesdropper, base station to user, STAR-RIS to user, and STAR-RIS to eavesdropper is 2.3, and the path loss index from base station to sensing target and STAR-RIS to sensing target is 2.7. The small-scale fading from base station to STAR-RIS, STAR-RIS to user, and STAR-RIS to eavesdropper follows the Rice distribution with a Rice factor of 3 dB. In addition, the small-scale fading from base station to target and STAR-RIS to target is modeled as a line-of-sight link.

[0143] Figure 2 The figure shows the learning curve of the proposed algorithm under different learning rates. When the learning rate of the Actor and Critic network is 0.0001, the algorithm shows the best performance under convergence. This is because when the learning rate is too large, the convergence is fast and the later performance is poor; when the learning rate is too small, the convergence is too slow, the training cost is increased, and the later performance improvement is limited. This proves that the proposed algorithm is very sensitive to the setting of the learning rate in the system under study, and it is very important to choose a suitable learning rate.

[0144] Figure 3 The figure shows the learning curve of the proposed algorithm under different adaptive sampling windows. The performance is better when the cumulative reward function window length is 25. This is because the window length affects the variance of the calculation, which in turn affects the noise scale. When the window length is small, the variance is too large, which makes the noise scale too small, causing the agent to fall into the local optimum; when the window length is large, the variance is too small, which makes the noise scale too large, making it difficult for the agent to converge to the current optimum. This shows that the proposed algorithm is very sensitive to the setting of the cumulative reward function window length in the system under study, and it is very important to choose an appropriate cumulative reward function window length.

[0145] Figure 4 The relationship between the power and system utility of different algorithms is shown in the figure. The system utility of all schemes increases with the increase of power. However, when the transmission power reaches a certain value, the growth rate of all schemes gradually slows down. This is because as the power increases, the growth of communication and perception performance gradually saturates, and an obvious marginal effect appears. This shows that in practical applications, power cannot be blindly increased to pursue performance. The proposed scheme has better performance than scheme 2, and the system utility increase is up to 14.6%. This is because the proposed action space and action mapping method can allocate power more flexibly. The proposed scheme has better performance than scheme 1, and the system utility increase is up to 23.5%. This is because the proposed adaptive action noise can better balance exploration and utilization, thereby obtaining a better strategy. Compared with scheme 3, it shows the worst performance at all power levels, and the system utility is lower than other schemes. This shows that without adaptive noise and the proposed action space design, the system utility is significantly limited.

[0146] Figure 5 The relationship between the power of different algorithms and the system and safety rate is shown in the figure. The total system safety rate of all schemes increases with the increase of power, but when the transmission power reaches a certain value, the growth rate of all schemes gradually slows down, which is similar to the trend of system utility. The total system safety rate of the proposed scheme increases by up to 29.3% compared with scheme 2, by up to 54.0% compared with scheme 1, and by up to 221.6% compared with scheme 3.

[0147] Figure 6 The relationship between the number of STAR-RIS units and system utility of different algorithms. As the number of units increases, the system utility of all schemes increases. This is because more units can provide more array gain. The proposed scheme achieves the performance of other schemes with more STAR-RIS units under the condition of fewer STAR-RIS units, thus saving the cost of STAR-RIS.

[0148] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions.

[0149] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0150] The above embodiments should be understood to be only used to illustrate the present invention and not to limit the protection scope of the present invention. After reading the contents of the present invention, technicians can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A smart beamforming method for a STAR-RIS assisted safety synaesthesia integrated system, characterized in that: The following steps are involved: S1: Building a wireless communication scenario, which includes a base station, STAR-RIS, a legitimate user, and an eavesdropper; S2: Construct the communication model and perception model of the STAR-RIS assisted safety synaesthesia integrated system; S3: Based on the communication model and perception model, construct the utility maximization problem of the system; S4: Modeling beamforming as a Markov decision process; S5: A noise-adaptive deep reinforcement learning algorithm is used to jointly optimize the base station’s transmit beamforming and the STAR-RIS transmission-reflection coefficient matrix.

2. The intelligent beamforming method of the STAR-RIS assisted safety synaesthesia integrated system according to claim 1 is characterized in that: The wireless communication scenario in step S1 consists of a telepathy base station, a STAR-RIS, T sensing targets, K users and an eavesdropper; the STAR-RIS adopts an energy splitting mode, that is, each element divides the incident signal into a reflected signal of a transmitted signal and transmits them simultaneously; the telepathy base station is configured with M transmitting / receiving antennas; the STAR-RIS is configured with N elements, and each of the communication user and the eavesdropper is configured with a single antenna.

3. The intelligent beamforming method of the STAR-RIS assisted safety synaesthesia integrated system according to claim 2 is characterized in that: The communication model and perception model of the STAR-RIS assisted safety synaesthesia integration system in step 2 include: Transmission and reflection coefficient matrices for STAR-RIS: β t,n ,θ t,n , β r,n ,θ r,n They represent the magnitude and phase of the transmission and reflection coefficient of the nth element of STAR-RIS respectively; The echo signal vector received by the base station is expressed as follows: in represents the transmission beamforming matrix of the base station to the communication user, represents a dedicated sensing signal beamforming matrix of a base station; Indicates the information transmitted by the base station to the communication user. Indicates the signal sent by the base station to the sensing target; represents the noise of the echo signal, is the noise power; They represent the channel matrices from BS to target, STAR-RIS to target, and BS to STAR-RIS, respectively. T represents transpose; Φ r represents the reflection phase shift matrix of STAR-RIS; The specific formula for the received signal of users in the transmission area and the reflection area is: in is the noise power, Φ i Represents the STAR-RIS transmission or reflection phase shift matrix. represents the channel from BS to the kth user and from STAR-RIS to the kth user; The eavesdropper's received signal: in is the noise power, They represent the channels from BS to eavesdropper and from STAR-RIS to eavesdropper respectively; The perceived signal-to-noise ratio of the base station is: in, (·) H represents conjugate transpose; The signal-to-interference-to-noise ratio of users in the transmission area and the reflection area is as follows: Among them, w k represents the beamforming vector transmitted by the base station to user k, w j represents the beamforming vector transmitted by the base station to other users; Calculate the eavesdropper's signal-to-interference-to-noise ratio. The specific formula is: The communication rate of user k is as follows: R k =log(1+γ k ), the eavesdropper’s communication rate, the specific formula is: R e,k =log(1+γ e,k ), the secure communication rate of user k, the specific formula is: in[·] + Indicates that only values ​​greater than 0 are meaningful; the total secure communication rate of the system, the specific formula is: System utility, the specific formula is: Where P is the base station transmit power, a represents the weight of the perceived performance, b represents the weight of the security performance, and c represents the weight of the power.

4. The intelligent beamforming method of the STAR-RIS assisted safety synaesthesia integrated system according to claim 3 is characterized in that: The utility maximization problem in step S3 is established as follows: taking the perceived signal-to-noise ratio of the base station and the secure communication rate of the legitimate user as constraints, the system utility of the synaesthesia integrated system is maximized by jointly optimizing the STAR-RIS phase shift matrix and the base station beamforming matrix, as follows: C3:P≤P max Where W represents the precoding matrix of the base station, Φ t and Φ r represents the transmission and reflection surface coefficient matrix of STAR-RIS, R th They represent the secure communication rate and secure communication rate threshold of the kth user, γ t , γ th They represent the echo signal-to-noise ratio and echo signal-to-noise ratio threshold of the t-th perceived target, θ i,n Represents the phase of the transmission and reflection surface coefficients. Constraint C1 can ensure that the secure communication rate of all users is not lower than a threshold, ensuring user privacy. Constraint C2 can ensure that the signal-to-noise ratio of the echo signal of the perceived target is not lower than a threshold. Constraint C3 ensures that the transmission power of the base station does not exceed the upper limit. Constraint C4 indicates that the sum of the signal powers of the STAR-RIS transmission surface and the reflection surface will not exceed the power of the incident signal. Constraint C5 represents the phase limit of the STAR-RIS unit.

5. The intelligent beamforming method of the STAR-RIS assisted safety synaesthesia integrated system according to claim 4 is characterized in that: The process of jointly optimizing the transmit beamforming of the base station and the transmission reflection coefficient matrix of STAR-RIS using the noise adaptive deep reinforcement learning algorithm in step S4 includes the following steps: S41: Define the system state space, action space and reward function, including: The state space is specifically defined as: in, represents the square of the L2 norm of each column of the precoding matrix W at the y-th time step, is the secure communication rate between each user and the eavesdropper at the yth time step, is the echo signal-to-noise ratio of each perceived target at the yth time step, and the dimension of the state space is M+2K+T; The action space is specifically defined as: in is the unprocessed base station precoding matrix output by the neural network at the yth time step, are the emission and reflection coefficients of STAR-RIS at the y-th time step, Re{·} means taking the real part, Im{·} means taking the imaginary part, δ is the power factor at the y-th time step, and the dimension of the action space is 4N+2M×(M+K)+1; The reward function is specifically defined as: Among them, ξ is the penalty factor; S42: Construct adaptive action noise and define action control strategies, including: The maximum and minimum normalized cumulative rewards of x rounds in the current training {R e-x ,...,R e-1 }, the specific formula is: Noise scale factor, the specific formula is: Among them, var(·) means to find the variance; Define the action control strategy, the specific formula is: Among them, a ε is an action with noise, a is a deterministic action, ε is the noise sampled from a standard normal distribution, and β is the scaling factor of the noise.

6. The intelligent beamforming method of the STAR-RIS assisted safety synaesthesia integrated system according to claim 5 is characterized in that: In step S5, a noise-adaptive deep reinforcement learning algorithm is used to jointly optimize the transmit beamforming of the base station and the transmission reflection coefficient matrix of STAR-RIS, including: Use the policy network to select actions, get rewards, transfer to the state of the next moment, and store the experience tuple in the experience replay pool; sample from the experience replay pool and update the value network of the agent; use the policy gradient to update the policy network based on the Q value generated by the value network; soft update the parameters of the policy target network and the value target network; update the noise scale factor at the end of each round.

7. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the intelligent beamforming method of the STAR-RIS auxiliary safety synaesthesia integration system as claimed in any one of claims 1 to 6 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the intelligent beamforming method of the STAR-RIS assisted safety synaesthesia integration system according to any one of claims 1 to 6 is implemented.

Images