Intelligent reflecting surface assisted VLC communication system secure transmission design method

Abstract

The invention discloses a secure transmission design method for a VLC communication system assisted by an intelligent reflecting surface. The method comprises the following steps: establishing an IRS-assisted RSMA indoor visible light communication system; obtaining secrecy energy efficiency according to the ratio of the total secrecy rate to the total power consumption; establishing an optimization problem of secrecy energy efficiency; the optimization problem is relaxed; modeling the optimization problem as a Markov decision process; solving a mixed integer nonlinear programming problem of secrecy energy efficiency maximization based on a DS-PPO algorithm of deep reinforcement learning; when the reward value gradually tends to be stable in the training process, the current result is a local or global optimal solution; the QoS requirement of a user is ensured to be met by optimizing resource allocation, and the strength of a signal received by an eavesdropper is weakened through beam forming and the intelligent reflection capability of an IRS, so that the security of the system is improved; the secrecy energy efficiency of the system is maximized by jointly optimizing beam forming, direct current bias, RSMA common rate and IRS incidence matrix.

Description

Technical Field

[0001] The present invention relates to the technical field of visible light communication, and in particular to a design method for secure transmission of a VLC communication system assisted by an intelligent reflective surface. Background Art

[0002] Visible Light Communication (VLC) technology based on light-emitting diodes (LEDs) is regarded as an important complementary technology to traditional RF communications. VLC has both lighting and communication functions, which can not only significantly expand the available spectrum resources, but also provide a green and low-cost wireless access method for future networks. A large number of empirical studies have shown that VLC is superior to RF communications in key indicators such as bandwidth and data transmission rate.

[0003] However, VLC also faces technical challenges that cannot be ignored. Due to its relatively short signal wavelength, the propagation channel is easily affected by shadows and obstructions, which often leads to discontinuous communication coverage and reduced quality of service (QoS). At the same time, the intelligent reflecting surface (IRS) technology is becoming a cutting-edge research direction of common concern in academia and industry due to its ability to actively reconstruct wireless propagation channels. Specifically, the optical IRS is composed of many passive reflection units, which can accurately control the reflection direction of the incident signal by deflecting their own angles, thereby changing the characteristics of the wireless channel and increasing the spatial degrees of freedom (DoF) of communication. With its highly controllable and intelligent signal reflection mechanism, IRS can effectively alleviate the problem of discontinuous communication coverage in VLC and significantly improve the overall system performance. On the other hand, due to the inherent broadcast characteristics of the VLC system and its reliance on line-of-sight links, potential eavesdroppers (Eve) can easily intercept confidential information transmitted to legitimate users (LUs), resulting in a large information security risk in the system. To address this issue, physical layer security (PLS) technology has become an effective way to ensure communication security. In the research of PLS, IRS is becoming a key technology to enhance communication security with its ability to accurately reconstruct the wireless channel environment. Specifically, IRS can actively direct signals to legitimate users while effectively suppressing signal reception by potential eavesdroppers, thereby significantly improving communication security.

[0004] In addition, the deployment of IRS inevitably introduces more communication links, which will further aggravate inter-user interference (IUI) in multi-user scenarios. Rate-splitting Multiple Access (RSMA) technology, as a cutting-edge multiple access technology, provides an effective way to alleviate IUI. In fact, RSMA cleverly combines the advantages of space division multiple access and non-orthogonal multiple access technology. It splits user information into a public part and a private part: the public part is jointly encoded into a public stream, and the private part is encoded into a private stream separately. Through the successive interference cancellation (SIC) technology, users can first decode and remove the public stream, and then decode their own private stream.

[0005] So far, although there have been many studies on IRS-assisted VLC systems and RSMA VLC systems, the collaborative application of IRS and RSMA in VLC systems has not been systematically explored. In addition, since VLC systems face serious information security risks, it is particularly important to design an effective wireless resource allocation scheme to improve information security. Summary of the invention

[0006] In response to the needs in the prior art, the present invention provides a design method for secure transmission of a VLC communication system assisted by an intelligent reflective surface, aiming to improve the information security and energy efficiency of the network.

[0007] A method for designing secure transmission of a VLC communication system assisted by an intelligent reflective surface comprises the following steps: Step 1: Establish an IRS-assisted RSMA indoor visible light communication system; Step 2: Calculate the ratio of the total confidentiality rate of the system to the total power consumption to obtain the confidentiality energy efficiency; Step 3: Establish the optimization problem of confidentiality energy efficiency; Step 4: Solve the optimization problem, specifically: Step 4.1: Relax the optimization problem; Step 4.2: Model the optimization problem as a Markov decision process. Step 4.3: Use the DS-PPO algorithm based on deep reinforcement learning to solve the mixed integer nonlinear programming problem of maximizing confidentiality energy efficiency; Step 4.4: When the reward value gradually stabilizes during training, the fluctuation range narrows to a certain stable range and there is no significant upward or downward trend, the current result is the local or global optimal solution.

[0008] Further, the optimization problem of confidentiality energy efficiency is specifically:

[0009] The optimization goal is to jointly optimize the beamforming , DC bias , Public Information Rate and IRS correlation matrix , To maximize the system's confidentiality energy efficiency SEE, the following optimization problem is constructed:

[0010] In this optimization problem, constraint (C1) is the user's service quality constraint, requiring that the receiving rate of each LU is not less than the service quality threshold ; (C2) and (C8) are RSMA public information rate constraints, ensuring that all LUs can successfully decode the public stream; (C2) Indicates (C3) limits the total power consumption of the system Stay within your power budget ; (C4) and (C9) require that the LED emission signal remain in the linear operating domain ; Constraints (C5)-(C7) are derived from the IRS incidence matrix and The definition of ensures that each IRS unit can be associated with at most one LED and one LU at the same time. For the The information rate obtained by decoding the private stream and The matrices are No. The nth item of the column and the matrix No. The nth item in the column.

[0011] Further: Step 4.1 is specifically:

[0012] Will Equivalently converted to ,Right now:

[0013] in, is the index set of the merge matrix, the matrix Defined as , among which, when , indicating that the IRS unit in The first LED and the A reflection channel is established between the LUs;

[0014] By transforming the binary matrix Convert to a continuous matrix , Relaxed to :

[0015] in, is a matrix The relaxed form of ; through the minimum distance projection criterion, The relaxation results of The optimal projection index of IRS units ,pass Calculate; where, yes No. OK, is an auxiliary vector whose elements are 1, and the rest are 0; in addition, is defined as an all-zero vector, so the corresponding index is given by:

[0016] Further: Step 4.2 is specifically:

[0017] The Markov decision process consists of an agent and a quaternion Composition, of which , , and Represent the action space, state space, reward function and discount factor respectively; according to , the action space, state space, reward function and discount factor are defined as follows:

[0018] The action space represents the set of actions that the agent can perform, which corresponds to the range of values ​​of the optimization variable, and the optimization variable is the beamforming vector , DC bias , Public Information Rate , and the relaxed IRS incidence matrix ;

[0019] The state space refers to all possible state representations of the environment. These possible state representations are usually composed of various characteristics of the environment and reflect the overall state of the system. Contains the current action ,award , and the signal-to-interference-to-noise ratio set ;

[0020] The reward function is calculated based on the given state and action, aiming to guide the agent to learn a near-optimal strategy; specifically, the reward function combines the penalty term of SEE and constraint, namely

[0021] in, , , , They are penalty items for different constraints. When the value is 1, it means the constraint is satisfied, otherwise it is not satisfied. The specific expressions are as follows:

[0022] Further: Step 4.3 is specifically:

[0023] The DS-PPO algorithm consists of two neural networks: the policy network and the value network. The policy network is responsible for and strategies Perform actions The choice of its network parameters is express; The probability distribution of the output action is defined, namely ; The value network is responsible for passing the value function To evaluate the action The value of is its network parameter. Specifically, the agent First, according to the current state and strategies Select Action ,Right now ; Then, through the interaction between the agent and the environment, the corresponding reward is obtained and the next state ; At the same time, the state transfer data obtained through interaction is stored in the experience pool; next, off-policy sampling and on-policy sampling are used to extract it from the experience pool and state transition samples for updating the neural network;

[0024] Policy network update: First, use GAE to estimate the advantage, that is,

[0025] in, , and define for ; Then calculate the loss function of the policy network, that is:

[0026] in, Represents the probability ratio of the current action to the previous action;

[0027] Finally, according to the loss function Policy network parameters Perform a mini-batch gradient descent update, i.e.

[0028] in, , represents the learning rate of the policy network;

[0029] Value network update: The policy network takes the return of GAE as the target value, that is, , and by minimizing the predicted value With target value The MSE update value network is:

[0030] Then, similar to the update of the policy network, according to the loss function Value network parameters Perform a mini-batch gradient descent update, namely:

[0031] in, Represents the learning rate of the value network.

[0032] Further: IRS assisted RSMA indoor visible light communication system is:

[0033] LEDs are evenly deployed on the ceiling, with An IRS with a mirror reflection unit is installed on the wall; there is a potential Eve on the ground trying to intercept the information sent by the LEDs to the user;

[0034] In the RSMA scheme, Information about LUs Will be divided into public parts and private part ; Subsequently, the common parts of all users are merged into , and encoded as a public stream ;No. The private part of each user is encoded independently as a private stream ; Each signal flow satisfies , ; The system's transmission signal Expressed as ,in, and Public Stream and private The beamforming vector of is the DC bias vector, Represents the DC bias current; the LED's emission signal remains in the linear operating domain, that is, , ,in , and Respectively represent the minimum and maximum driving current of the LED.

[0035] Further: Confidential Energy Efficiency SEE is:

[0036] in, and They represent the association matrices between IRS units and LUs and between IRS units and LEDs, respectively; For beamforming, is the DC bias, is the public information rate; is the total confidentiality rate, is the total power consumption of the system, is the public wiretapping rate, is a vector of all 1s, is the forward voltage of the LED, Represents the constant power consumption of the circuit module and IRS.

[0037] The beneficial effects of the present invention are as follows: by optimizing resource allocation, it ensures that the QoS requirements of users are met, and by using beamforming and the intelligent reflection capability of IRS, the signal strength received by the eavesdropper is weakened, thereby improving the security of the system; at the same time, multiple constraints such as power budget, RSMA public information decoding and LED linear working domain are also met during the optimization process, thereby maximizing the system confidentiality energy efficiency by jointly optimizing beamforming, DC bias, RSMA public rate and IRS association matrix. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flow chart of the present invention; Figure 2 It is a model diagram of the IRS-assisted RSMA VLC network system; Figure 3 This is a schematic diagram of the DS-PPO algorithm based on deep reinforcement learning; Figure 4 is the training curve of the DS-PPO algorithm; Figure 5 It is a comparison chart of SEE training curves of different methods; Figure 6 It is a comparison chart of SEE between RSMA and SDMA under different numbers of IRS units; Figure 7 It is the impact of the number of LUs on the SEE and total power of the system. DETAILED DESCRIPTION

[0039] The present invention is described in detail below in conjunction with the accompanying drawings. The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention. The directional terms such as left, middle, right, top, and bottom in the embodiments of the present invention are only relative concepts or are based on the normal use state of the product, and should not be considered as restrictive.

[0040] While LEDs provide communication services to LUs, a potential Eve attempts to intercept the transmitted information. By deploying IRS to accurately reconstruct the wireless channel environment, actively direct the signal to LUs, and suppress Eve's signal reception, the security of communication can be improved. The specific scheme is shown in the figure. Figure 2 Based on this model, the present invention provides a design method for secure transmission of a VLC communication system assisted by an intelligent reflective surface, which can optimize resource allocation according to model parameters to maximize the confidentiality efficiency of the system.

[0041] like Figure 1 As shown, the technical solution adopted by the present invention to solve the technical problem is as follows:

[0042] The first step is to build the IRS-assisted RSMA indoor VLC system model:

[0043] In an indoor visible light communication network, LEDs are evenly deployed on the ceiling, with An IRS with a mirror reflection unit is installed on the wall. With the help of IRS, LEDs transmit signals to the receiver through both line-of-sight and non-line-of-sight channels in RSMA mode. LUs provide high-quality communication services. At the same time, there is a potential Eve on the ground trying to intercept the information sent by LEDs to LUs.

[0044] In the RSMA scheme, Information about LUs Will be divided into public parts and private part . Subsequently, the common parts of all LUs are combined into , and encoded as a public stream ;No. The private part of each LU is encoded independently as a private stream Each signal flow satisfies , The system's transmission signal Expressed as ,in, and Public Stream and private The beamforming vector of is the DC bias vector, In order to avoid signal clipping, the LED's transmission signal must be kept in the linear operating domain, that is, , ,in , and Respectively represent the minimum and maximum driving current of the LED.

[0045] 1) Line-of-sight channel: LED transmitters usually conform to the Lambertian model, so the channel gain can be determined by the geometric position of the transmitter and receiver. The first LED and the The channel gain between LUs can be expressed as:

[0046] in, is the Lambert index, is the physical area of ​​PD. , and denote the half-intensity field angle, radiation angle and incident angle, respectively. Indicates LED to The line-of-sight transmission distance of each LU, is the optical filter gain, represents the optical concentrator gain. Similarly, The channel gain between the LED and the eavesdropper is:

[0047] 2) Non-line-of-sight channel gain: Optical reflection is divided into diffuse reflection and specular reflection, of which diffuse reflection is usually ignored because the channel gain is much smaller than specular reflection. The LEDs pass through the IRS units to The non-line-of-sight channel gain of a LU can be expressed as:

[0048] in, , is the IRS reflection coefficient, and Respectively LED to IRS units and IRS units to Since optical signals are highly directional, non-line-of-sight channels are only effective when the IRS unit is associated with both the LED and the LU. Therefore, the line-of-sight channel gain depends on the channel gain and two binary incidence matrices, namely:

[0049] in, and denote the association matrices between IRS units and LUs and between IRS units and LEDs, respectively. or indicates association, and vice versa. LED to The non-line-of-sight channel gain of each LU is:

[0050] in, and They are No. column sum No. List, and The matrices are No. The nth item of the column and the matrix No. And because the reflection path is almost a straight line, each IRS unit can only be associated with one LED and one LU at most, that is, , Considering both line-of-sight and non-line-of-sight channels, The total channel gain of LU is:

[0051] in, and Respectively represent The line-of-sight channel gain and non-line-of-sight channel gain of each LU.

[0052] The second step is to calculate the ratio of the total confidentiality rate of the system to the total power consumption to obtain the confidentiality energy efficiency:

[0053] No. The information rate obtained by a LU by decoding the public stream and the private stream can be expressed as:

[0054] At the same time, in order to ensure that all LUs can successfully decode the common stream, all LU common information rates should meet ,in Indicates The public information rate of LUs. The total information transmission rate of LU is On the other hand, when Eve tries to eavesdrop on the data stream sent to LU, the public eavesdropping rate and the private eavesdropping rate are expressed as:

[0055] Therefore, the total confidentiality rate is expressed as:

[0056] in, and They represent the confidentiality rates of public and private flows respectively.

[0057] In addition, the total power consumption of the system It can be expressed as:

[0058] in, is a vector of all 1s, is the forward voltage of the LED, Represents the constant power consumption of the circuit module and IRS.

[0059] The variables to be optimized are the beamforming vector , public information rate , DC bias and IRS correlation matrix , , then the confidentiality energy efficiency SEE of the system under consideration can be expressed as:

[0060] The third step is to establish the optimization problem of confidentiality energy efficiency:

[0061] The optimization goal is to jointly optimize the beamforming , DC bias , Public Information Rate and IRS correlation matrix , To maximize the system's confidentiality energy efficiency SEE, the following optimization problem can be constructed based on the model:

[0062] In this optimization problem, constraint (C1) is the LU (user) quality of service (QoS) constraint, requiring that the receiving rate of each LU is not less than the QoS threshold ; (C2) and (C8) are RSMA common information rate constraints, ensuring that all LUs can successfully decode the common stream. Indicates (C3) limits the total power consumption of the system Stay within your power budget ; In order to avoid signal clipping, (C4) and (C9) require the LED emission signal to remain in the linear operating domain ; Constraints (C5)-(C7) are derived from the IRS incidence matrix and The definition of ensures that each IRS unit can be associated with at most one LED and one LU at the same time.

[0063] The fourth step is to design an algorithm to solve the optimization problem:

[0064] because It is a mixed integer nonlinear programming problem, and traditional optimization methods are challenging to solve. At the same time, deep reinforcement learning has attracted the attention of the academic community because it can effectively handle highly computational optimization problems. Therefore, this paper proposes a dual sampling-proximal policy optimization (DS-PPO) algorithm based on deep reinforcement learning to solve .

[0065] 1) Relax the optimization problem

[0066] Due to the constraints (C5)−(C7), the matrix and has column orthogonality. Therefore, can be equivalently converted to ,Right now:

[0067] in, is the index set of the merge matrix, the matrix Defined as , among which, when , indicating that the IRS unit in The first LED and the A reflection channel is established between the LUs.

[0068] Since there are integer variables, solving the problem To solve this problem, a feasible approach is to first optimize its relaxed form , and then the result is restored to binary by the minimum distance projection criterion. Therefore, by converting the binary matrix Convert to a continuous matrix , can be relaxed to:

[0069] in, is a matrix The relaxed form of . By using the minimum distance projection criterion, The relaxation result of is converted into a binary result. Specifically, The optimal projection index of IRS units Available through Calculate. Among them, yes No. OK, is an auxiliary vector whose elements are 1, and the rest are 0. In addition, is defined as a vector of all zeros. The corresponding index is then given by:

[0070] 2) Model the optimization problem as a Markov decision process (MDP)

[0071] Markov decision process (MDP) is the modeling basis of reinforcement learning. An MDP consists of an agent and a quaternion Composition, of which , , and Represent the action space, state space, reward function and discount factor respectively. , the action space, state space, reward function and discount factor are defined as follows:

[0072] The action space represents the set of actions that the agent can perform. In the present invention, these action sets correspond to the optimization variables (i.e., the beamforming vectors , DC bias , Public Information Rate , and the relaxed IRS incidence matrix ) value range;

[0073] The state space refers to all possible states of the environment. These states are usually composed of various characteristics of the environment and reflect the overall state of the system. In the present invention, the state space Contains the current action ,award , and the signal-to-interference-to-noise ratio set ;

[0074] The reward function is calculated based on the given state and action, aiming to guide the agent to learn a near-optimal strategy. Specifically, the reward function combines the penalty term of SEE and constraints, namely:

[0075] in, , , , They are penalty items for different constraints. When the value is 1, it means the constraint is satisfied, otherwise it is not satisfied. The specific expressions are as follows:

[0076] 3) DS-PPO algorithm design

[0077] This paper proposes a DS-PPO algorithm based on deep reinforcement learning to solve the mixed integer nonlinear programming problem of maximizing confidentiality energy efficiency. The algorithm diagram is shown in Figure 3 . The DS-PPO algorithm has two improvements on the traditional PPO algorithm: ① Combining the two sampling mechanisms of off-policy sampling and on-policy sampling to enhance sample diversity; ② Using generalized advantage estimation (GAE) instead of traditional RTG for advantage estimation to better control the bias and variance of policy gradient estimation. The DS-PPO algorithm contains two neural networks: the policy network (actor network) and the value network (critic network). The policy network is responsible for and strategies Perform actions The choice of its network parameters is express. The probability distribution of the output action is defined (subject to Gaussian distribution), that is, ; The value network is responsible for passing the value function To evaluate the action The value of is its network parameter. Specifically, the agent First, according to the current state and strategies Select Action ,Right now . Then, after the interaction between the agent and the environment, the corresponding reward is obtained and the next state At the same time, the state transition data obtained through interaction is stored in the experience pool. Next, off-policy sampling and on-policy sampling are used to extract and state transition samples for updating the neural network.

[0078] Policy network update: First use GAE to estimate the advantage, that is:

[0079] in, , and define for Then the loss function of the policy network is calculated, i.e.

[0080] in, represents the probability ratio of the current action to the previous action. Finally, according to the loss function Policy network parameters Perform a mini-batch gradient descent update, namely:

[0081] in, , Represents the learning rate of the policy network.

[0082] Value network update: The policy network takes the return of GAE as the target value, that is, , and by minimizing the predicted value With target value The MSE update value network is:

[0083] Then, similar to the update of the policy network, according to the loss function Value network parameters Perform a mini-batch gradient descent update, namely:

[0084] in, Represents the learning rate of the value network.

[0085] When the reward value gradually stabilizes during training, the fluctuation range narrows to a certain stable range and there is no significant upward or downward trend, it indicates that the strategy iteration of the neural network is close to saturation and the current result is close to the local or global optimal solution, and the training can be ended.

[0086] The specific process of the DS-PPO algorithm is summarized as follows:

[0087]

[0088] Figure 4 The training curve of the DS-PPO method is shown, showing good convergence. Figure 4 In (a), we can see that the gap between the SEE curve and the reward curve gradually decreases with the increase of training steps, which indicates that the strategy of the reinforcement learning agent is becoming more and more accurate; Figure 4 (b) shows that the achievable rate of each LU can meet the QoS requirements.

[0089] Figure 5 The SEE achieved by different methods is depicted. It can be seen that the DS-PPO method proposed in this patent can achieve higher SEE than other baseline methods. Specifically, the SEE achieved by DS-PPO is about 11.5% higher than that of PPO, because DS-PPO uses a more advanced GAE method for advantage estimation. In addition, DS-PPO converges faster than PPO, thanks to the combination of on-policy sampling and off-policy sampling.

[0090] Figure 6 The SEE achieved by RSMA and SDMA with different numbers of IRS units are compared. First, it can be seen that with the increase of IRS units, the SEE values ​​achieved by RSMA and SDMA will increase. This is because more IRS units can provide more spatial degrees of freedom. In addition, the SEE achieved by RSMA is always higher than that of SDMA, thanks to RSMA's flexible interference management technology.

[0091] Figure 7 (a) and Figure 7 (b) The SEE and total power of the network under different numbers of legal users are plotted. Figure 7 As can be seen in (a), as the number of users increases, the SEE value first increases and then decreases. This is because when there are fewer users, the increase in the number of users can more effectively utilize the spatial freedom provided by the LED; but when there are more users, more users will cause serious inter-user interference, thereby reducing SEE. In addition, the increase in inter-user interference will lead to stricter QoS requirements, thus consuming more power, such as Figure 7 (b) as shown.

[0092] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A design method for secure transmission of a VLC communication system assisted by an intelligent reflective surface, characterized in that: The following steps are involved: Step 1: Establish an IRS-assisted RSMA indoor visible light communication system; Step 2: Calculate the ratio of the total confidentiality rate of the system to the total power consumption to obtain the confidentiality energy efficiency; Step 3: Establish the optimization problem of confidentiality energy efficiency; Step 4: Solve the optimization problem, specifically: Step 4.1: Relax the optimization problem; Step 4.2: Model the optimization problem as a Markov decision process. Step 4.3: Use the DS-PPO algorithm based on deep reinforcement learning to solve the mixed integer nonlinear programming problem of maximizing confidentiality energy efficiency; Step 4.4: When the reward value gradually stabilizes during training, the fluctuation range narrows to a certain stable range and there is no significant upward or downward trend, the current result is the local or global optimal solution.

2. The method for designing secure transmission of a VLC communication system assisted by an intelligent reflective surface according to claim 1, characterized in that: The optimization problem of confidentiality energy efficiency is specifically: The optimization goal is to jointly optimize the beamforming , DC bias, public information rate and IRS correlation matrix , To maximize the system's confidentiality energy efficiency SEE, the following optimization problem is constructed: ； In this optimization problem, constraint (C1) is the user's service quality constraint, requiring that the receiving rate of each LU is not less than the service quality threshold ; (C2) and (C8) are RSMA public information rate constraints, ensuring that all LUs can successfully decode the public stream; (C2) Indicates (C3) limits the total power consumption of the system Stay within your power budget ; (C4) and (C9) require that the LED emission signal remain in the linear operating domain ; Constraints (C5)-(C7) are derived from the IRS incidence matrix and The definition of ensures that each IRS unit can be associated with at most one LED and one LU at the same time; For the The information rate obtained by each LU by decoding the private stream, and The matrices are No. The nth item of the column and the matrix No. The nth item in the column.

3. The method for designing secure transmission of a VLC communication system assisted by an intelligent reflective surface according to claim 2, characterized in that: Step 4.1 is as follows: Will Equivalently converts to ,Right now: ； in, is the index set of the merge matrix, the matrix Defined as , among which, when , indicating that the IRS unit in The first LED and the A reflection channel is established between the LUs; By transforming the binary matrix Convert to a continuous matrix , Relaxed to : ； in, is a matrix The relaxed form of ; through the minimum distance projection criterion, The relaxation results of The optimal projection index of IRS units ,pass Calculate; where, yes No. OK, is an auxiliary vector whose elements are 1, and the rest are 0; in addition, is defined as an all-zero vector, so the corresponding index is given by: 。 4. The method for designing secure transmission of a VLC communication system assisted by an intelligent reflective surface according to claim 3, characterized in that: Step 4.2 is as follows: The Markov decision process consists of an agent and a quaternion Composition, of which , , and Represent the action space, state space, reward function and discount factor respectively; according to , the action space, state space, reward function and discount factor are defined as follows: The action space represents the set of actions that the agent can perform, which corresponds to the range of values ​​of the optimization variable, and the optimization variable is the beamforming vector , DC bias , Public Information Rate , and the relaxed IRS incidence matrix ; The state space refers to all possible state representations of the environment. These possible state representations are usually composed of various characteristics of the environment and reflect the overall state of the system. Contains the current action ,award , and the signal-to-interference-to-noise ratio set ; The reward function is calculated based on the given state and action, aiming to guide the agent to learn a near-optimal strategy; specifically, the reward function combines the penalty term of SEE and constraint, namely: ； in, , , , They are penalty items for different constraints. When the value is 1, it means the constraint is satisfied, otherwise it is not satisfied. The specific expressions are as follows: 。 5. The method for designing secure transmission of a VLC communication system assisted by an intelligent reflective surface according to claim 4, characterized in that: Step 4.3 is as follows: The DS-PPO algorithm consists of two neural networks: the policy network and the value network. The policy network is responsible for and strategies Perform actions The choice of its network parameters is express; The probability distribution of the output action is defined, namely ; The value network is responsible for passing the value function To evaluate the action The value of is its network parameter. Specifically, the agent First, according to the current state and strategies Select Action ,Right now ; Then, through the interaction between the agent and the environment, the corresponding reward is obtained and the next state ; At the same time, the state transfer data obtained through interaction is stored in the experience pool; next, off-policy sampling and on-policy sampling are used to extract it from the experience pool and state transition samples for updating the neural network; Policy network update: First, use GAE to estimate the advantage, that is, ； in, , and define for ; Then calculate the loss function of the policy network, that is: ； in, Represents the probability ratio of the current action to the previous action; Finally, according to the loss function Policy network parameters Perform a mini-batch gradient descent update, i.e. ； in, , represents the learning rate of the policy network; Value network update: The policy network takes the return of GAE as the target value, that is, , and by minimizing the predicted value With target value The MSE update value network is: ； Then, similar to the update of the policy network, according to the loss function Value network parameters Perform a mini-batch gradient descent update, namely: ； in, Represents the learning rate of the value network.

6. The method for designing secure transmission of a VLC communication system assisted by an intelligent reflective surface according to claim 1, characterized in that: The IRS-assisted RSMA indoor visible light communication system is: LEDs are evenly deployed on the ceiling, with An IRS with a mirror reflection unit is installed on the wall; there is a potential Eve on the ground trying to intercept the information sent by the LEDs to the user; In the RSMA scheme, Information about LUs Will be divided into public parts and private part ; Subsequently, the common parts of all users are merged into , and encoded as a public stream ;No. The private part of each user is encoded independently as a private stream ; Each signal flow satisfies , ; The system's transmission signal Expressed as ,in, and Public Stream and private The beamforming vector of is the DC bias vector, Represents the DC bias current; the LED's emission signal remains in the linear operating domain, that is, , ,in , and Respectively represent the minimum and maximum driving current of the LED.

7. The method for designing secure transmission of a VLC communication system assisted by an intelligent reflective surface according to claim 6, characterized in that: The confidentiality energy efficiency SEE is: ； in, and They represent the association matrices between IRS units and LUs and between IRS units and LEDs, respectively; For beamforming, is the DC bias, is the public information rate; is the total confidentiality rate, is the total power consumption of the system, is the public wiretapping rate, is a vector of all 1s, is the forward voltage of the LED, Represents the constant power consumption of the circuit module and IRS.

Images