Safe and timely transmission method of short packet ISAC system assisted by air computing

By using the meta reinforcement learning algorithm of equalization field A3C in a short-packet ISAC system assisted by air computing, the air computing transceiver and block length are optimized, and the system's security and information freshness are solved, and the security and timeliness of information transmission are achieved. .

CN119095052APending Publication Date: 2024-12-06NORTH CHINA ELECTRIC POWER UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411180757.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The short-pack ISAC system assisted by air computing has challenges in ensuring communication security and information freshness. The prior art cannot meet the system's security and information freshness requirements at the same time.

Method used

A meta-reinforcement learning algorithm based on balanced field A3C is adopted to optimize the air computing transceiver settings and block length, and multi-objective optimization problems are designed, combining the safety information age interrupt probability, air computing mean square error and target detection error to achieve safe and timely transmission.

Benefits of technology

It ensures the security and timeliness of information transmission, solves the multi-objective optimization problem, realizes the security and timeliness of the system, and meets the requirements of information freshness and communication security.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119095052A_ABST
    Figure CN119095052A_ABST
Patent Text Reader

Abstract

The invention discloses a safe and timely transmission method of a short packet ISAC system assisted by air computing. According to the method, an element reinforcement learning method of an equilibrium field A3C is adopted, the setting of an air computing transceiver and the block length of the short packet ISAC system assisted by air computing are controlled, and meanwhile, the information freshness, the communication safety, the air computing precision and the sensing precision of user information are ensured. By solving the problem of multi-objective optimization, joint optimization of air computing performance, communication performance and sensing performance is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of wireless communications, and in particular relates to a secure and timely transmission method for an air computing-assisted short packet ISAC system. Background Art

[0002] With the convergence of communication, perception and computing technologies, the next generation of wireless networks is considered to be an important force in promoting emerging industrial IoT applications, including self-driving cars, extended reality, artificial intelligence and digital twins, which require large amounts of data and low latency. To support these applications, IoT devices need to collect large amounts of data from the environment and pass it to the server for further fusion and computing. The Integrated Sensing and Communication (ISAC) technology can collect and transmit data at the physical layer simultaneously by using the same spectrum and signal in radar sensing and data communication, thereby improving the spectrum efficiency of the system. In addition, with the emergence of air computing technology, ISAC systems have made significant progress in reducing energy consumption and shortening processing time. Air computing processes signals by eliminating interference generated when data is transmitted simultaneously, thereby reducing the latency required for transmission.

[0003] Due to the spectrum sharing and broadcast characteristics of wireless transmission during the signal superposition process of air computing, the short packet ISAC system assisted by air computing faces unique security challenges. At the same time, due to the demand for information freshness in industrial Internet of Things applications, ensuring the freshness of information is also crucial. Existing research cannot guarantee the security and freshness of wireless communication systems at the same time. Therefore, it is necessary to study the problem of simultaneously ensuring security and information freshness of the short packet ISAC system assisted by air computing, design an effective transmission method, and meet the accuracy of air computing and perception. Summary of the invention

[0004] In order to study the problems of physical layer security, perception accuracy and information freshness in an air computing-assisted short packet ISAC system, the present invention discloses a secure and timely transmission method for an air computing-assisted short packet ISAC system.

[0005] The embodiments of the present invention provide the following technical methods:

[0006] A secure and timely transmission method for an air computing-assisted short packet ISAC system, the method comprising:

[0007] Step A, establish a short packet-based RIS-assisted ISAC system with eavesdroppers, perform air computing transceiver settings, and build communication, perception, and computing models;

[0008] Step B, calculating the aerial calculation mean square error and target detection estimation error representing the aerial calculation accuracy and perception accuracy;

[0009] Step C, calculating the security information age and security information age interruption probability which characterize communication security and information freshness;

[0010] Step D, establishing a multi-objective optimization problem based on air computing transceiver and block length design to simultaneously minimize the safety information age interruption probability, air computing mean square error and target detection estimation error;

[0011] In step E, a meta-reinforcement learning algorithm based on the equilibrium field A3C is designed to transform the multi-objective optimization problem into multiple single-objective optimization problems for optimization, and a weight update strategy between optimization objectives is designed to reduce the error of manual parameter adjustment and achieve safe and timely transmission.

[0012] Wherein, step A specifically includes:

[0013] A1, established an airborne computing-assisted short-packet ISAC system, which includes a public target, an eavesdropper, and an N a Access point with M antennas and N configurations s =N tx +N rx Radar sensors with 10 antennas each can simultaneously send detection signals and data symbols to the access point for air calculations. tx For transmitting signal, N rx The signal sent by the sensor has the dual functions of sensing the target and sending data to the access point. At the same time, the eavesdropper is equipped with N e root antennas, attempting to eavesdrop on data sent to the access point;

[0014] A2, formulates a transmission protocol, divides the signal transmission process into different time slots, and synchronizes the data transmission operations of different sensors according to the clock signal broadcast by the server. At the same time, the channel state information between the access point and each sensor is estimated in the pilot signal broadcast by each sensor and then transmitted to the access point;

[0015] A3, construct the signal model and define the data symbol transmitted by the mth sensor in the tth time slot as The data symbol transmitted by the i-th sensor in the t-th time slot is K represents the number of functions calculated over the air, assuming that the data symbols are independent and identically distributed, with mean 0 and unit variance, that is, A and W m They represent the receive and transmit beamformers of the air computing, referred to as the air computing transceiver, and are transmitted by applying the transmit beamformer The signal emitted by the mth sensor can be expressed as x m [t] = W m s m [t];

[0016] A4, building a communication and air computing model. In order to prevent eavesdroppers from interfering, the sensor introduces artificial noise when transmitting signals to the access point. in Considering the imperfect CSI caused by the uncertainty in the channel, the communication channel and the artificial noise channel are modeled as and in and are the statistics of the communication channel and the artificial noise channel, ΔH m and ΔG m are the error terms of the communication channel and the artificial noise channel respectively. Since the target is far away from the access point, it is assumed that the target reflection signal disappears at the access point, so the signal at the receiving end can be written as At the same time, the signal received by the eavesdropper can be written as H em , W em and ΔG em For the eavesdropper's receiving receiver, the communication channel matrix, the transmitter in the air and the artificial noise channel error terms are calculated, represents Gaussian white noise, which obeys The distribution of is the noise power of the communication channel;

[0017] A5, construct the perception model. In order to ensure the quality of the perception signal, suppose the target reflection signal received by the mth sensor Represented as z m [t] = G mm W m s m [t]+Ω m [t]+n r [t],Ω m [t] = ∑ i∈M / {m} G im W i s i [t]+∑ i∈M / {m} Q im W i s i [t] is the interference signal, G mm is the statistical target response matrix, W i is the airborne computing transmitter of the ith sensor, and is the target response matrix and direct radar matrix between the ith and mth sensors, represents Gaussian white noise, which obeys The distribution of To sense the noise power of the channel, according to the matched filtering principle, the statistical matrix of the sensor signal can be expressed as is the statistical channel matrix.

[0018] Wherein, step B specifically includes:

[0019] B1, the calculation does not take into account the air computing mean square error set by the air computing transceiver. It is assumed that the air computing pre-processing and post-processing functions of the data sent by the sensor are linear. Therefore, in order to calculate the received signal y(t) and the ideal signal y ideal (t), and the time slot sequence is hidden for the convenience of calculation, and the error caused by the artificial noise channel CSI is quantified as And the mean square error calculated in the air is Similarly, the mean square error between the eavesdropper signal and the ideal signal is given here to evaluate the physical layer security performance of the system, and the error caused by CSI in the eavesdropper's artificial noise channel is quantified as So we can get the mean square error of the communication signal at the eavesdropper:

[0020] B2, calculate the air-to-air mean square error considering the optimal air-to-air transceiver setting, and define the optimal channel estimator under short packet communication as In each iteration, the access point broadcasts the current receiver A to all sensor nodes, and the sensor calculates the current optimal transmitter W m , after which the access point will obtain the effective CSI vector g when all nodes send pilot signals simultaneously. g can be expressed as At the same time, the effective CSI matrix F can be obtained by sending pilot signals from all sensors, and F can be expressed as Combined with the effective CSI vector matrix, the optimal transceiver setting under the influence of imperfect CSI and limited block length is defined as and The optimal W m By setting it into the MSE at the receiver, we can get the mean square error of the system calculated over the air based on the optimal receiver A.

[0021]

[0022] B3, define the target detection error. Each sensor estimates the position of the target based on its own position and the angle and distance information extracted from the reflected radar signal. Then, each sensor transmits the estimated position of the target to the server through air calculation, so that the server obtains the average estimated target position. The design of beamforming will also affect the perception accuracy. Let the maximum number of time slots in the transmission process be T m , N mThe distribution of N m The probability density of can be written as in G can be obtained by minimizing the log-likelihood function mm The maximum likelihood estimate of Maximum Likelihood Estimation The expression can be written as The log-likelihood function expression is: Taking the derivative of this function, under the condition that the derivative is zero, we can get G mm The maximum likelihood estimate of The estimated mean square error of can be calculated as MSE G It is defined as the target detection error.

[0023] Wherein, step C specifically includes:

[0024] C1, define information age. First, information age is defined as the time since the latest information update package was generated. The information age curve reaches its peak before the target node receives the information update, and the peak value is ΔT(l) = T D (l)-T A (l-1), T D (l) and T A (l) respectively represent the departure time and the departure time of the lth state update. To maintain consistency, let T A (0) = 0. Under the first-come, first-served strategy, the departure time of the lth state update can be expressed as T SC (j, l) represents the cumulative transmission time from the lth update to the jth update (l<j), T SC (j,l) is defined as T S (n) represents the transmission time of the nth update. The total residence time T(l) of each update should not be less than T S (n), similarly, let the interval between the arrival times of adjacent state update packets be X(l-1,l)=T A (l)-T A (l-1), we can get the arrival time interval between status update packets and status update packets as Suppose the residence time of a single data packet is

[0025] T(l)=T D (l)-T A (l) = max{T SC (j,l)-X(j,l)}, so the information age can be updated as

[0026] ΔT(l)=T D (l)-T A (l-1) = X(l-1,l) + T(l);

[0027] C2, define the security information age interruption probability. Based on the definition of information age, the security information age is defined as the information age of the eavesdropper minus the information age of the receiver. Specifically, the instantaneous peak information age of the eavesdropper and the receiver is defined as ΔE(l) and ΔM(l). Then the security information age of the system can be expressed as: ΔS(l) = [ΔE(l) - ΔM(l)] +, [·] + is a positive function, and the security information age interruption probability is defined as μ th The threshold value of the information lag between the eavesdropping user and the receiving user in terms of the amount of channel usage;

[0028] C3, since it is impossible to directly use mathematical analysis methods to derive the closed expression of the safety information age interruption probability, the present invention uses the moment generating function method to derive the closed expression of the safety information age, and defines M according to the moment generating function X (θ)=E[e θX ], where θ>0 is a real variable used to generate and describe the moment of the random variable, X is the variable for which the moment generating function is required, and the moment generating function of the safety information age interruption probability is defined as M ΔS(l) (θ)=E[e θΔS(l) ]=M ΔE(l) (θ)M ΔM(l) (-θ), and at the same time, the present invention assumes that the arrival time service interval of these data packets obeys Poisson distribution, that is, the arrival process of the data packets is a Poisson process with parameter λ, and the parameters of the receiving end and the eavesdropper are the same, and the service time of each data packet at the receiving end and the eavesdropper obeys the parameters μ m and μ e Then, let the length of the communication data block transmitted by the short packet ISAC system assisted by air computing be N. According to the above assumptions, the expression of the probability of safety information age interruption can be obtained as Where T represents the unit time of each channel symbol, ε m and ε e are the decoding error probabilities of data packets transmitted for user m and the eavesdropper, respectively.

[0029] Wherein, step D specifically comprises:

[0030] D1, in order to comprehensively measure the security, information freshness and perception accuracy of the short packet ISAC system assisted by air computing, a multi-objective optimization problem is proposed. Specifically, in order to better meet the requirements of the receiving end users for the freshness of air computing information and the physical layer security, and at the same time study the relationship between communication and perception performance and computing in the short packet ISAC system assisted by air computing, the present invention jointly optimizes the air computing transceiver A, W m and block length N, taking into account the channel perception and target detection accuracy as well as the signal quality at the eavesdropper, further minimizing the probability of security information age interruption;

[0031] D2, the optimization problem is expressed as

[0032]

[0033] sC D ≥R S +R E

[0034]

[0035] MSE e >η e

[0036]

[0037] Among them, C D ≥R S +R E is the reliability constraint, C D is the information transmission rate, R S is the security information transmission rate, R E is the eavesdropper's information transmission rate; For security constraints, is the probability of information leakage, δ∈[0,1] is the confidentiality constraint; MSE e >η e is the eavesdropper signal quality constraint, η e is the eavesdropper signal quality threshold; is the transmission power limit at the sensor, and P is the maximum transmission power of the user.

[0038] Wherein, step E specifically includes:

[0039] E1, to solve the multi-objective optimization problem, introduce a decomposition strategy and introduce J weight vectors satisfy j∈J, each weight vector corresponds to a sub-problem, and solving each sub-optimization problem obtains a set of potential Pareto optimal solutions. When all J optimization problems are solved, the expected Pareto frontier can be obtained. The multi-objective optimization problem is decomposed into a set of sub-problems, and each sub-problem is regarded as a single-objective optimization problem and expressed as a Markov decision process;

[0040] E2, defines the state space, the channel error term when the agent is in the state at time t in and They represent the error terms of the communication channel and the artificial noise channel at time t, and the error term in the eavesdropper channel at time t. And the transmission power P at time t t Composition, expressed as s t ={H t ,H et ,P t}, the state of the agent is determined by the state at the previous moment and the action taken at the previous moment;

[0041] E3, define the action space, the action of the agent is composed of the air calculation transceiver and block length at the receiving end, defined as a t ={A,W m ,N}, in each state, the agent takes different actions according to the established strategy to obtain the maximum reward;

[0042] E4, define the strategy space and function, the strategy of the agent is defined as given state s t Select an action t The probability of t |s t ), and defines a parameter θ a A deep neural network, this policy function takes the state of the agent as input and outputs a policy;

[0043] E5, define the reward function. In order to minimize the probability of safety information interruption, the mean square error of air calculation and the target detection estimation error, the reward function is defined as Different from the single-objective Markov decision process, in the multi-objective Markov decision process, in the state s t Next select action a t Then return a set of rewards [R1,...,R J ] T , using R(a t |s t ) to represent the reward vector;

[0044] E6, define the value function, which measures the value of each state, that is, the future return that can be achieved in each state. For an initial state s0, define its value function as where γ t is the discount factor;

[0045] E7, perform task training, train on a set of training tasks, update a set of sub-model parameters, and train the sub-model parameters through asynchronous advantage action evaluation (A3C). The state value function of A3C can be expressed as where θ v are the parameters of the deep neural network used to estimate the value function. A3C uses the v-step reward to update its parameters. The action value of the v-th step is defined as Introducing the advantage function To determine the advantage of the action taken in the current state, during the task training phase, first randomly sample from the given weight vector distribution p(λ) Weight vector For each weight vector λ j , the loss function of the policy network (Actor) is in, is the entropy of the strategy, ξ is the step size, and the cumulative gradient of the Actor is calculated as

[0046] where θ a ′ is the thread parameter of Actor, and the loss function of the value network (Critic) is The cumulative gradient of Critic is calculated as where θ v ' is the thread parameter of Critic;

[0047] E8, in A3C, each agent updates its parameters after interacting with the environment and passes them to the global network. During the training process, the agent uses the accumulated gradients to independently update the global network model parameters and periodically synchronizes the latest parameters from the global network to the local network. This process can use the equilibrium field approximation principle. The algorithm that combines the equilibrium field approximation and A3C is called the equilibrium field A3C algorithm. The local strategy affects the overall optimization and introduces the mean function mean(·). The mean of the parameter gradient can be used to update the policy parameters and value parameters of the j-th sub-model. and

[0048]

[0049] E9, design the weight update strategy between optimization targets. In each round of iteration, according to the performance of the current weight strategy, evaluate the mutual influence between the weights and calculate the Nash equilibrium point. At this equilibrium point, no single weight can achieve better results through unilateral adjustment, so as to determine the optimal weight combination for the next step. The weight update rule is: α is the learning rate, It is about j The gradient of the utility function U, the Nash equilibrium condition is This formula indicates that any fine-tuning of a single weight at the Nash equilibrium point will not further improve the utility function U, thus achieving the optimal weight update;

[0050] E10, meta-training stage, aggregates the strategy trajectories obtained from task training and adjusts the meta-model parameters through differentiation. The parameter update rule is:

[0051]

[0052] Where ε represents the update step size. After the training is completed, the model will adapt to the new task according to the optimized parameters. It only needs to fine-tune the meta-model to obtain the corresponding Pareto frontier, thereby achieving safe and reliable transmission.

[0053] Compared with the prior art, the above technical method has the following advantages:

[0054] The present invention ensures the security and timeliness of information transmission through a short packet ISAC system assisted by air computing. In order to simultaneously measure the freshness of information and communication security, the probability of interruption of the age of security information is introduced, and the perception accuracy and air computing mean square error are considered at the same time. A meta-reinforcement learning algorithm based on the balanced field A3C set by the air computing transceiver is designed to simultaneously minimize SAOP, target detection error and air computing mean square error, thereby solving the multi-objective optimization problem and ensuring the security and timeliness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical methods in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A flowchart of a secure and timely transmission method for an air computing-assisted short packet ISAC system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] As described in the background technology section, how to improve the accuracy of aerial computing and perception while meeting the strict requirements of time-sensitive services in the emerging industrial Internet of Things for communication security and information freshness is an urgent problem to be solved by people in this field.

[0058] The core idea of ​​the present invention is to use the probability of security information age interruption as a measure of communication security and information freshness in an air computing-assisted short packet ISAC system, while considering the accuracy of air computing and perception, and using the meta-reinforcement learning algorithm of balanced field A3C to solve the multi-objective optimization problem. Compared with traditional transmission methods, this method can efficiently meet the strict requirements of time-sensitive services on security and timeliness.

[0059] See also Figure 1 The embodiment of the present invention provides a method for secure and timely transmission of a short packet ISAC system assisted by air computing, the method comprising:

[0060] Step A, establish a short packet-based RIS-assisted ISAC system with eavesdroppers, perform air computing transceiver settings, and build communication, perception, and computing models;

[0061] Step B, calculating the aerial calculation mean square error and target detection estimation error representing the aerial calculation accuracy and perception accuracy;

[0062] Step C, calculating the security information age and security information age interruption probability which characterize communication security and information freshness;

[0063] Step D, establishing a multi-objective optimization problem based on air computing transceiver and block length design to simultaneously minimize the safety information age interruption probability, air computing mean square error and target detection estimation error;

[0064] In step E, a meta-reinforcement learning algorithm based on the equilibrium field A3C is designed to transform the multi-objective optimization problem into multiple single-objective optimization problems for optimization, and a weight update strategy between optimization objectives is designed to reduce the error of manual parameter adjustment and achieve safe and timely transmission.

[0065] Wherein, step A specifically includes:

[0066] A1, established an airborne computing-assisted short-packet ISAC system, which includes a public target, an eavesdropper, and an N a Access point with M antennas and N configurations s =N tx +N rx Radar sensors with 10 antennas each can simultaneously send detection signals and data symbols to the access point for air calculations. tx For transmitting signal, N rxThe signal sent by the sensor has the dual functions of sensing the target and sending data to the access point. At the same time, the eavesdropper is equipped with N e root antennas, attempting to eavesdrop on data sent to the access point;

[0067] A2, formulates a transmission protocol, divides the signal transmission process into different time slots, and synchronizes the data transmission operations of different sensors according to the clock signal broadcast by the server. At the same time, the channel state information between the access point and each sensor is estimated in the pilot signal broadcast by each sensor and then transmitted to the access point;

[0068] A3, construct the signal model and define the data symbol transmitted by the mth sensor in the tth time slot as The data symbol transmitted by the i-th sensor in the t-th time slot is K represents the number of functions calculated over the air, assuming that the data symbols are independent and identically distributed, with mean 0 and unit variance, that is, A and W m They represent the receive and transmit beamformers of the air computing, referred to as the air computing transceiver, and are transmitted by applying the transmit beamformer The signal emitted by the mth sensor can be expressed as x m [t] = W m s m [t];

[0069] A4, building a communication and air computing model. In order to prevent eavesdroppers from interfering, the sensor introduces artificial noise when transmitting signals to the access point. in Considering the imperfect CSI caused by the uncertainty in the channel, the communication channel and the artificial noise channel are modeled as and in and are the statistics of the communication channel and the artificial noise channel, ΔH m and ΔG m are the error terms of the communication channel and the artificial noise channel respectively. Since the target is far away from the access point, it is assumed that the target reflection signal disappears at the access point, so the signal at the receiving end can be written as At the same time, the signal received by the eavesdropper can be written as H em , W em and ΔG em For the eavesdropper's receiving receiver, the communication channel matrix, the transmitter in the air and the artificial noise channel error terms are calculated, represents Gaussian white noise, which obeys The distribution of is the noise power of the communication channel;

[0070] A5, construct the perception model. In order to ensure the quality of the perception signal, suppose the target reflection signal received by the mth sensor Represented as z m [t] = G mm W m s m [t]+Ω m [t]+n r [t],Ω m [t] = ∑ i∈M / {m} G im W i s i [t]+∑ i∈M / {m} Q im W i s i [t] is the interference signal, G mm is the statistical target response matrix, W i is the airborne computing transmitter of the ith sensor, and is the target response matrix and direct radar matrix between the ith and mth sensors, represents Gaussian white noise, which obeys The distribution of To sense the noise power of the channel, according to the matched filtering principle, the statistical matrix of the sensor signal can be expressed as is the statistical channel matrix.

[0071] Wherein, step B specifically includes:

[0072] B1, the calculation does not take into account the air computing mean square error set by the air computing transceiver. It is assumed that the air computing pre-processing and post-processing functions of the data sent by the sensor are linear. Therefore, in order to calculate the received signal y(t) and the ideal signal y ideal (t), and for the convenience of calculation, the present invention hides the time slot sequence and quantifies the error caused by the artificial noise channel CSI as And the mean square error calculated in the air is Similarly, the mean square error between the eavesdropper signal and the ideal signal is given here to evaluate the physical layer security performance of the system, and the error caused by CSI in the eavesdropper's artificial noise channel is quantified as So we can get the mean square error of the communication signal at the eavesdropper:

[0073] B2, calculate the air-to-air mean square error considering the optimal air-to-air transceiver setting, and define the optimal channel estimator under short packet communication as In each iteration, the access point broadcasts the current receiver A to all sensor nodes, and the sensor calculates the current optimal transmitter W m , after which the access point will obtain the effective CSI vector g when all nodes send pilot signals simultaneously. g can be expressed as At the same time, the effective CSI matrix F can be obtained by sending pilot signals from all sensors, and F can be expressed as Combined with the effective CSI vector matrix, the optimal transceiver setting under the influence of imperfect CSI and limited block length is defined as and The optimal W m By setting it into the MSE at the receiver, we can get the mean square error of the system calculated over the air based on the optimal receiver A.

[0074]

[0075] B3, define the target detection error. Each sensor estimates the position of the target based on its own position and the angle and distance information extracted from the reflected radar signal. Then, each sensor transmits the estimated position of the target to the server through air calculation, so that the server obtains the average estimated target position. The design of beamforming will also affect the perception accuracy. Let the maximum number of time slots in the transmission process be T m , N m The distribution of N m The probability density of can be written as in G can be obtained by minimizing the log-likelihood function mm The maximum likelihood estimate of Maximum Likelihood Estimation The expression can be written as The log-likelihood function expression is: Taking the derivative of this function, under the condition that the derivative is zero, we can get G mm The maximum likelihood estimate of The estimated mean square error of can be calculated as MSE G It is defined as the target detection error.

[0076] Wherein, step C specifically includes:

[0077] C1, define information age. First, information age is defined as the time since the latest information update package was generated. The information age curve reaches its peak before the target node receives the information update, and the peak value is ΔT(l) = T D (l)-T A (l-1), T D (l) and TA (l) respectively represent the departure time and the departure time of the lth state update. To maintain consistency, let T A (0) = 0. Under the first-come, first-served strategy, the departure time of the lth state update can be expressed as T SC (j, l) represents the cumulative transmission time from the lth update to the jth update (l<j), T SC (j,l) is defined as T S (n) represents the transmission time of the nth update. The total residence time T(l) of each update should not be less than T S (n), similarly, let the interval between the arrival times of adjacent state update packets be X(l-1,l)=T A (l)-T A (l-1), we can get the arrival time interval between status update packets and status update packets as Suppose the residence time of a single data packet is

[0078] T(l)=T D (l)-T A (l) = max{T SC (j,l)-X(j,l)}, so the information age can be updated as

[0079] ΔT(l)=T D (l)-T A (l-1) = X(l-1,l) + T(l);

[0080] C2, define the security information age interruption probability. Based on the definition of information age, the security information age is defined as the information age of the eavesdropper minus the information age of the receiver. Specifically, the instantaneous peak information age of the eavesdropper and the receiver is defined as ΔE(l) and ΔM(l). Then the security information age of the system can be expressed as: ΔS(l) = [ΔE(l) - ΔM(l)] + , [·] + To take a positive function, define the probability of security information age interruption as μ th The threshold value of the information lag between the eavesdropping user and the receiving user in terms of the amount of channel usage;

[0081] C3, since it is impossible to directly use mathematical analysis methods to derive the closed expression of the safety information age interruption probability, the present invention uses the moment generating function method to derive the closed expression of the safety information age, and defines M according to the moment generating function X (θ)=E[e θX], where θ>0 is a real variable used to generate and describe the moment of the random variable, X is the variable for which the moment generating function is required, and the moment generating function of the safety information age interruption probability is defined as M ΔS(l) (θ)=E[e θΔS(l) ]=M ΔE(l) (θ)M ΔM(l) (-θ), and at the same time, the present invention assumes that the arrival time service interval of these data packets obeys Poisson distribution, that is, the arrival process of the data packets is a Poisson process with parameter λ, and the parameters of the receiving end and the eavesdropper are the same, and the service time of each data packet at the receiving end and the eavesdropper obeys the parameters μ m and μ e Then, let the length of the communication data block transmitted by the short packet ISAC system assisted by air computing be N. According to the above assumptions, the expression of the probability of safety information age interruption can be obtained as Where T represents the unit time of each channel symbol, ε m and ε e are the decoding error probabilities of data packets transmitted for user m and the eavesdropper, respectively.

[0082] Wherein, step D specifically comprises:

[0083] D1, in order to comprehensively measure the security, information freshness and perception accuracy of the short packet ISAC system assisted by air computing, a multi-objective optimization problem is proposed. Specifically, in order to better meet the requirements of the receiving end users for the freshness of air computing information and the physical layer security, and at the same time study the relationship between communication and perception performance and computing in the short packet ISAC system assisted by air computing, the present invention jointly optimizes the air computing transceiver A, W m and block length N, taking into account the channel perception and target detection accuracy as well as the signal quality at the eavesdropper, further minimizing the probability of security information age interruption;

[0084] D2, the optimization problem is expressed as

[0085]

[0086] sC D ≥R S +R E

[0087]

[0088] MSE e >η e

[0089]

[0090] Among them, C D ≥RS +R E is the reliability constraint, C D is the information transmission rate, R S is the security information transmission rate, R E is the eavesdropper's information transmission rate; For security constraints, is the probability of information leakage, δ∈[0,1] is the confidentiality constraint; MSE e >η e is the eavesdropper signal quality constraint, η e is the eavesdropper signal quality threshold; is the transmission power limit at the sensor, and P is the maximum transmission power of the user.

[0091] Wherein, step E specifically includes:

[0092] E1, to solve the multi-objective optimization problem, introduce a decomposition strategy and introduce J weight vectors satisfy j∈J, each weight vector corresponds to a sub-problem, and solving each sub-optimization problem obtains a set of potential Pareto optimal solutions. When all J optimization problems are solved, the expected Pareto frontier can be obtained. The multi-objective optimization problem is decomposed into a set of sub-problems, and each sub-problem is regarded as a single-objective optimization problem and expressed as a Markov decision process;

[0093] E2, defines the state space, the channel error term when the agent is in the state at time t in and They represent the error terms of the communication channel and the artificial noise channel at time t, and the error term in the eavesdropper channel at time t. And the transmission power P at time t t Composition, expressed as s t ={H t ,H et ,P t}, the state of the agent is determined by the state at the previous moment and the action taken at the previous moment;

[0094] E3, define the action space, the action of the agent is composed of the air calculation transceiver and block length at the receiving end, defined as a t ={A,W m ,N}, in each state, the agent takes different actions according to the established strategy to obtain the maximum reward;

[0095] E4, define the strategy space and function, the strategy of the agent is defined as given state s t Select an action t The probability oft |s t ), and defines a parameter θ a A deep neural network, this policy function takes the state of the agent as input and outputs a policy;

[0096] E5, define the reward function. In order to minimize the probability of safety information interruption, the mean square error of air calculation and the target detection estimation error, the reward function is defined as Different from the single-objective Markov decision process, in the multi-objective Markov decision process, in the state s t Next select action a t Then return a set of rewards [R1,...,R J ] T , using R(a t |s t ) to represent the reward vector;

[0097] E6, define the value function, which measures the value of each state, that is, the future return that can be achieved in each state. For an initial state s0, define its value function as where γ t is the discount factor;

[0098] E7, perform task training, train on a set of training tasks, update a set of sub-model parameters, and train the sub-model parameters through asynchronous advantage action evaluation (A3C). The state value function of A3C can be expressed as where θ v are the parameters of the deep neural network used to estimate the value function. A3C uses the v-step reward to update its parameters. The action value of the v-th step is defined as Introducing the advantage function To determine the advantage of the action taken in the current state, during the task training phase, first randomly sample from the given weight vector distribution p(λ) Weight vector For each weight vector λ j , the loss function of the policy network (Actor) is in, is the entropy of the strategy, ξ is the step size, and the cumulative gradient of the Actor is calculated as

[0099] where θ a ′ is the thread parameter of Actor, and the loss function of the value network (Critic) is The cumulative gradient of Critic is calculated as where θ v ' is the thread parameter of Critic;

[0100] E8, in A3C, each agent updates its parameters after interacting with the environment and passes them to the global network. During the training process, the agent uses the accumulated gradients to independently update the global network model parameters and periodically synchronizes the latest parameters from the global network to the local network. This process can use the equilibrium field approximation principle. The algorithm that combines the equilibrium field approximation and A3C is called the equilibrium field A3C algorithm. The local strategy affects the overall optimization and introduces the mean function mean(·). The mean of the parameter gradient can be used to update the policy parameters and value parameters of the j-th sub-model. and

[0101]

[0102] E9, design the weight update strategy between optimization targets. In each round of iteration, according to the performance of the current weight strategy, evaluate the mutual influence between the weights and calculate the Nash equilibrium point. At this equilibrium point, no single weight can achieve better results through unilateral adjustment, so as to determine the optimal weight combination for the next step. The weight update rule is: α is the learning rate, It is about j The gradient of the utility function U, the Nash equilibrium condition is This formula indicates that any fine-tuning of a single weight at the Nash equilibrium point will not further improve the utility function U, thus achieving the optimal weight update;

[0103] E10, meta-training stage, aggregates the strategy trajectories obtained from task training and adjusts the meta-model parameters through differentiation. The parameter update rule is:

[0104]

[0105] Where ε represents the update step size. After the training is completed, the model will adapt to the new task according to the optimized parameters. It only needs to fine-tune the meta-model to obtain the corresponding Pareto frontier, thereby achieving safe and reliable transmission.

[0106] Compared with the prior art, the above technical method has the following advantages:

[0107] The present invention ensures the security and timeliness of information transmission through a short packet ISAC system assisted by air computing. In order to simultaneously measure the freshness of information and communication security, the probability of interruption of the age of security information is introduced, and the perception accuracy and air computing mean square error are considered at the same time. A meta-reinforcement learning algorithm based on the balanced field A3C set by the air computing transceiver is designed to simultaneously minimize SAOP, target detection error and air computing mean square error, thereby solving the multi-objective optimization problem and ensuring the security and timeliness of the system.

[0108] The present invention discloses a secure and timely transmission method for an air computing-assisted short packet ISAC system. The method adopts a meta-reinforcement learning method of a balanced field A3C to control the air computing transceiver settings and block length of the air computing-assisted short packet ISAC system, while ensuring the information freshness, communication security, air computing accuracy and perception accuracy of user information. By solving multi-objective optimization problems, the joint optimization of air computing performance, communication performance and perception performance is achieved.

[0109] The various parts in this manual are described in a progressive manner, and each part focuses on the differences from other parts. The same or similar parts between the various parts can be referenced to each other.

[0110] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A secure and timely transmission method for a short packet ISAC system assisted by air computing, characterized in that: The steps include: Step A, establish a short packet-based RIS-assisted ISAC system with eavesdroppers, set up air computing transceivers, and build communication, perception, and computing models; Step B, calculating the aerial calculation mean square error and target detection estimation error representing the aerial calculation accuracy and perception accuracy; Step C, calculating the security information age and security information age interruption probability which characterize communication security and information freshness; Step D, establishing a multi-objective optimization problem based on air computing transceiver and block length design to simultaneously minimize the safety information age interruption probability, air computing mean square error and target detection estimation error; Step E: Design a meta-reinforcement learning algorithm based on the balanced field A3C, transform the multi-objective optimization problem into multiple single-objective optimization problems for optimization, and design a weight update strategy between optimization objectives to reduce the error of manual parameter adjustment and achieve safe and timely transmission; Wherein, step A specifically includes: A1, established an airborne computing-assisted short-packet ISAC system, which includes a public target, an eavesdropper, and an N a Access point with M antennas and N configurations s =N tx +N rx Radar sensors with 10 antennas each can simultaneously send detection signals and data symbols to the access point for air calculations. tx For transmitting signal, N rx The signal sent by the sensor has the dual functions of sensing the target and sending data to the access point. At the same time, the eavesdropper is equipped with N e root antennas, attempting to eavesdrop on data sent to the access point; A2, formulates a transmission protocol, divides the signal transmission process into different time slots, and synchronizes the data transmission operations of different sensors according to the clock signal broadcast by the server. At the same time, the channel state information between the access point and each sensor is estimated in the pilot signal broadcast by each sensor and then transmitted to the access point; A3, construct the signal model and define the data symbol transmitted by the mth sensor in the tth time slot as The data symbol transmitted by the i-th sensor in the t-th time slot is K represents the number of functions calculated over the air, assuming that the data symbols are independent and identically distributed, with mean 0 and unit variance, that is, A and W m They represent the receive and transmit beamformers of the air computing, referred to as the air computing transceiver, and are transmitted by applying the transmit beamformer The signal emitted by the mth sensor can be expressed as x m [t] = W m s m [t]; A4, building a communication and air computing model. In order to prevent eavesdroppers from interfering, the sensor introduces artificial noise when transmitting signals to the access point. in Considering the imperfect CSI caused by the uncertainty in the channel, the communication channel and the artificial noise channel are modeled as and in and are the statistics of the communication channel and the artificial noise channel, ΔH m and ΔG m are the error terms of the communication channel and the artificial noise channel respectively. Since the target is far away from the access point, it is assumed that the target reflection signal disappears at the access point, so the signal at the receiving end can be written as At the same time, the signal received by the eavesdropper can be written as H em , W em and ΔG em For the eavesdropper's receiving receiver, the communication channel matrix, the transmitter in the air and the artificial noise channel error terms are calculated, represents Gaussian white noise, which obeys The distribution of is the noise power of the communication channel; A5, construct the perception model. In order to ensure the quality of the perception signal, suppose the target reflection signal received by the mth sensor Represented as z m [t] = G mm W m s m [t]+Ω m [t]+n r [t],Ω m [t] = Σ i∈M / {m} G im W i s i [t]+Σ i∈M / {m} Q im W i s i [t] is the interference signal, G mm is the statistical target response matrix, W i is the airborne computing transmitter of the ith sensor, and is the target response matrix and direct radar matrix between the ith and mth sensors, represents Gaussian white noise, which obeys The distribution of To sense the noise power of the channel, according to the matched filtering principle, the statistical matrix of the sensor signal can be expressed as is the statistical channel matrix; Wherein, step B specifically includes: B1, the calculation does not take into account the air computing mean square error set by the air computing transceiver. It is assumed that the air computing pre-processing and post-processing functions of the data sent by the sensor are linear. Therefore, in order to calculate the received signal y(t) and the ideal signal y ideal (t), and the time slot sequence is hidden for the convenience of calculation, and the error caused by the artificial noise channel CSI is quantified as And the mean square error calculated in the air is Similarly, the mean square error between the eavesdropper signal and the ideal signal is given here to evaluate the physical layer security performance of the system, and the error caused by CSI in the eavesdropper's artificial noise channel is quantified as So we can get the mean square error of the communication signal at the eavesdropper: B2, calculate the air-to-air mean square error considering the optimal air-to-air transceiver setting, and define the optimal channel estimator under short packet communication as In each iteration, the access point broadcasts the current receiver A to all sensor nodes, and the sensor calculates the current optimal transmitter W m , after which the access point will obtain the effective CSI vector g when all nodes send pilot signals simultaneously. g can be expressed as At the same time, the effective CSI matrix F can be obtained by sending pilot signals from all sensors, and F can be expressed as Combined with the effective CSI vector matrix, the optimal transceiver setting under the influence of imperfect CSI and limited block length is defined as and The optimal W m By setting it into the MSE at the receiver, we can get the mean square error of the system calculated over the air based on the optimal receiver A. B3, define the target detection error. Each sensor estimates the position of the target based on its own position and the angle and distance information extracted from the reflected radar signal. Then, each sensor transmits the estimated position of the target to the server through air calculation, so that the server obtains the average estimated target position. The design of beamforming will also affect the perception accuracy. Let the maximum number of time slots in the transmission process be T m , N m The distribution of N m The probability density of can be written as in G can be obtained by minimizing the log-likelihood function mm The maximum likelihood estimate of Maximum Likelihood Estimation The expression can be written as The log-likelihood function expression is: Taking the derivative of this function, under the condition that the derivative is zero, we can get G mm The maximum likelihood estimate of The estimated mean square error of can be calculated as MSE G Defined as target detection error; Wherein, step C specifically includes: C1, define information age. First, information age is defined as the time since the latest information update package was generated. The information age curve reaches its peak before the target node receives the information update, and the peak value is ΔT(l) = T D (l)-T A (l-1), T D (l) and T A (l) respectively represent the departure time and the departure time of the lth state update. To maintain consistency, let T A (0) = 0. Under the first-come, first-served strategy, the departure time of the lth state update can be expressed as T SC (j, l) represents the cumulative transmission time from the lth update to the jth update (l<j), T SC (j,l) is defined as T S (n) represents the transmission time of the nth update. The total residence time T(l) of each update should not be less than T S (n), similarly, let the interval between the arrival times of adjacent state update packets be X(l-1,l)=T A (l)-T A (l-1), we can get the arrival time interval between status update packets and status update packets as Assume that the residence time of a single data packet is T(l)=T D (l)-T A (l) = max{T SC (j,l)-X(j,l)}, so the information age can be updated as ΔT(l)=T D (l)-T A (l-1)=X(l-1,l)+T(l); C2, define the security information age interruption probability. Based on the definition of information age, the security information age is defined as the information age of the eavesdropper minus the information age of the receiver. Specifically, the instantaneous peak information age of the eavesdropper and the receiver is defined as ΔE(l) and ΔM(l). Then the security information age of the system can be expressed as: ΔS(l) = [ΔE(l) - ΔM(l)] + , [·] + To take a positive function, define the probability of security information age interruption as μ th The threshold value of the information lag between the eavesdropping user and the receiving user in terms of the amount of channel usage; C3, since it is impossible to directly use mathematical analytical methods to derive the closed expression of the safety information age interruption probability, the closed expression of the safety information age is derived using the moment generating function method. According to the moment generating function definition M X (θ)=E[e θX ], where θ>0 is a real variable used to generate and describe the moment of the random variable, X is the variable for which the moment generating function is required, and the moment generating function of the safety information age interruption probability is defined as M ΔS(l) (θ)=E[e θΔS(l) ]=M ΔE(l) (θ)M ΔM(l) (-θ), and at the same time, it is assumed that the arrival time service interval of these data packets follows a Poisson distribution, that is, the arrival process of the data packets is a Poisson process with parameter λ, and the parameters of the receiving end and the eavesdropper are the same, and the service time of each data packet at the receiving end and the eavesdropper follows the parameters μ m and μ e Then, let the length of the communication data block transmitted by the short packet ISAC system assisted by air computing be N. According to the above assumptions, the expression of the probability of safety information age interruption can be obtained as Where T represents the unit time of each channel symbol, ε m and ε e are the decoding error probabilities of data packets transmitted by user m and the eavesdropper respectively; Wherein, step D specifically comprises: D1,In order to comprehensively measure the security, information freshness and perception accuracy of the short packet ISAC system assisted by air computing, a multi-objective optimization problem is proposed. Specifically, in order to better meet the requirements of users at the receiving end for the freshness of air computing information and physical layer security, the relationship between communication and perception performance and computing in the short packet ISAC system assisted by air computing is studied. By jointly optimizing the air computing transceiver A, W m and block length N, taking into account the channel perception and target detection accuracy as well as the signal quality at the eavesdropper, further minimizing the probability of security information age interruption; D2, the optimization problem is expressed as s.t.C D ≥R S +R E MSE e >the e Among them, C D ≥R S +R E is the reliability constraint, C D is the information transmission rate, R S is the security information transmission rate, R E is the eavesdropper's information transmission rate; For security constraints, is the probability of information leakage, δ∈[0,1] is the confidentiality constraint; MSE e >η e is the eavesdropper signal quality constraint, η e is the eavesdropper signal quality threshold; is the transmission power limit at the sensor, and P is the maximum transmission power of the user; Wherein, step E specifically includes: E1, to solve the multi-objective optimization problem, introduce a decomposition strategy and introduce J weight vectors satisfy Each weight vector corresponds to a sub-problem. Solving each sub-optimization problem yields a set of potential Pareto optimal solutions. When all J optimization problems are solved, the desired Pareto frontier can be obtained. The multi-objective optimization problem is decomposed into a set of sub-problems. Each sub-problem is regarded as a single-objective optimization problem and expressed as a Markov decision process. E2, defines the state space, the channel error term when the agent is in the state at time t in and They represent the error terms of the communication channel and the artificial noise channel at time t, and the error term in the eavesdropper channel at time t. And the transmission power P at time t t Composition, expressed as s t ={H t ,H et ,P t }, the state of the agent is determined by the state at the previous moment and the action taken at the previous moment; E3, define the action space, the action of the agent is composed of the air calculation transceiver and block length at the receiving end, defined as a t ={A,W m ,N}, in each state, the agent takes different actions according to the established strategy to obtain the maximum reward; E4, define the strategy space and function, the strategy of the agent is defined as given state s t Select an action t The probability of t |s t ), and defines a parameter θ a A deep neural network, this policy function takes the state of the agent as input and outputs a policy; E5, define the reward function. In order to minimize the probability of safety information interruption, the mean square error of air calculation and the target detection estimation error, the reward function is defined as Different from the single-objective Markov decision process, in the multi-objective Markov decision process, in the state s t Next select action a t Then return a set of rewards [R1,...,R J ] T , using R(a t |s t ) to represent the reward vector; E6, define the value function, which measures the value of each state, that is, the future return that can be achieved in each state. For an initial state s0, define its value function as where γ t is the discount factor; E7, perform task training, train on a set of training tasks, update a set of sub-model parameters, and train the sub-model parameters through asynchronous advantage action evaluation (A3C). The state value function of A3C can be expressed as where θ v are the parameters of the deep neural network used to estimate the value function. A3C uses the v-step reward to update its parameters. The action value of the v-th step is defined as Introducing the advantage function To determine the advantage of the action taken in the current state, during the task training phase, first randomly sample from the given weight vector distribution p(λ) Weight vector For each weight vector λ j , the loss function of the policy network (Actor) is in, is the entropy of the strategy, ξ is the step size, and the cumulative gradient of the Actor is calculated as where θ a ′ is the thread parameter of Actor, and the loss function of the value network (Critic) is The cumulative gradient of Critic is calculated as where θ v ' is the thread parameter of Critic; E8, in A3C, each agent updates its parameters after interacting with the environment and passes them to the global network. During the training process, the agent uses the accumulated gradients to independently update the global network model parameters and periodically synchronizes the latest parameters from the global network to the local network. This process can use the equilibrium field approximation principle. The algorithm that combines the equilibrium field approximation and A3C is called the equilibrium field A3C algorithm. The local strategy affects the overall optimization and introduces the mean function mean(·). The mean of the parameter gradient can be used to update the policy parameters and value parameters of the j-th sub-model. and E9, design the weight update strategy between optimization targets. In each round of iteration, according to the performance of the current weight strategy, evaluate the mutual influence between the weights and calculate the Nash equilibrium point. At this equilibrium point, no single weight can achieve better results through unilateral adjustment, so as to determine the optimal weight combination for the next step. The weight update rule is: α is the learning rate, It is about j The gradient of the utility function U, the Nash equilibrium condition is This formula indicates that any fine-tuning of a single weight at the Nash equilibrium point will not further improve the utility function U, thus achieving the optimal weight update; E10, meta-training stage, aggregates the strategy trajectories obtained from task training and adjusts the meta-model parameters through differentiation. The parameter update rule is: Where ε represents the update step size. After training, the model will adapt to the new task based on the optimized parameters. Only by fine-tuning the meta-model can the corresponding Pareto front be obtained, thus achieving safe and reliable transmission.

Citation Information

Cited By

  • Limited block length optimization method and system of cellular-free communication perception integrated system

    CN120342546A