Sensing quality based integrated sensing and communication transmission optimization method for internet of vehicles

Abstract

The application relates to the technical field of integrated communication and sensing calculation of Internet of Vehicles, and discloses a sensing quality-based integrated communication and sensing calculation transmission optimization method of Internet of Vehicles, which comprises the following specific steps: establishing a vehicle-infrastructure cooperation system in a target environment; dividing time slots and updating vehicle positions; constructing a sensing model based on RSU side sensing and vehicle side sensing; respectively calculating maximum RSU side and vehicle side sensing condition mutual information based on the sensing model; in each time slot, judging whether the maximum RSU side condition is greater than the maximum vehicle side condition mutual information; if yes, the RSU sensing mode is adopted in the time slot; if not, the vehicle-RSU cooperative processing sensing mode is adopted in the time slot; and the minimum time delay and energy consumption results under each time slot are calculated. The application solves the problems of large calculation amount and easy falling into a local optimal solution in the prior art, and has the characteristics of being capable of improving the delay of the vehicle-mounted network and reducing the energy consumption of the vehicle-mounted network.

Description

Technical Field

[0001] This invention relates to the field of integrated communication, sensing, and computing technology in the Internet of Vehicles (IoV), and more specifically, to an IoV-based integrated communication, sensing, and computing transmission optimization method based on sensor quality. Background Technology

[0002] In recent years, autonomous driving technology has attracted widespread attention from industry and academia. In the intelligent transportation networks of future cities, autonomous driving technology and its various applications require higher precision and lower latency. With the integration of vehicle infrastructure, vehicle technology, and IoT technology, a large amount of spectrum resources are being used in the vehicle-to-everything (V2X) network, leading to spectrum scarcity. Sensor-Integrated Communication (ISAC) technology, derived from 5G millimeter-wave communication, integrates communication, radar sensing, and camera technologies into a single system. Utilizing a unified transceiver frequency band for communication and sensing, it can fully utilize the spectrum and reduce information transmission latency between different types of communication and sensing, thus improving spectrum utilization efficiency. Simultaneously, edge computing offloading technology plays a crucial role in the converged sensor-integrated communication (ISCC) system. The onboard network can interact with roadside infrastructure via communication links, offloading data tasks to roadside units with powerful computing capabilities as a service platform, effectively reducing transmission latency and overall energy consumption.

[0003] In existing research on spectrum subcarrier allocation and power allocation in communication and sensing, traditional constrained convex optimization methods are commonly used to solve the problem. However, in large vehicle networks, multi-objective optimization and a series of complex nonlinear constraints are involved. Moreover, in edge offloading, considering the duality of local and edge computing offloading tasks, traditional optimization methods may generate huge computational loads and get trapped in local optima. Some existing studies use deep reinforcement learning algorithms, such as DQN or DDPG, which can handle large-scale state and action spaces and learn optimal decisions in dynamic and uncertain environments, becoming powerful tools for solving edge computing task offloading problems. However, the research mentioned above that considers the integration of communication, sensing, and computing rarely considers the issue of vehicle sensing quality assessment. In autonomous driving, perception quality is an important indicator affecting vehicle safety. In the integrated communication, sensing, and computing system that combines vehicles and roadside infrastructure, many studies consider equipping the vehicle side with a joint sensing system (JCR), ignoring the powerful computing capabilities of roadside units. These units can also be equipped with a joint sensing system and directly compute vehicle perception information, thus reducing V2I transmission latency and improving the efficiency and safety of autonomous driving.

[0004] In summary, existing technologies suffer from high computational costs and are prone to getting trapped in local optima. Therefore, how to invent an integrated optimization method for vehicle networking that can save computational costs and achieve global optimal solutions is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] To address the problems of high computational load and susceptibility to local optima in existing technologies, this invention provides an integrated transmission optimization method for vehicle networking based on sensor quality, which can improve the latency of the vehicle network and reduce its energy consumption.

[0006] To achieve the above-mentioned objectives of this invention, the technical solution adopted is as follows:

[0007] A vehicle-to-everything (V2X) communication, sensing, and computing integrated transmission optimization method based on sensor quality includes the following specific steps:

[0008] In the target environment, a vehicle-infrastructure cooperative system consisting of several roadside units (RSUs) and several vehicles is established, in which each vehicle and RSU is equipped with an intelligent integrated communication, sensing, and computing device consisting of a communication processor, a computing processor, and sensors.

[0009] Divide time slots and update vehicle positions;

[0010] Construct a perception model based on RSU-side perception and vehicle-side perception;

[0011] The perception conditional mutual information of the RSU side and the vehicle side is calculated based on the perception model; the perception conditional mutual information of the RSU side and the vehicle side is maximized by optimizing the perception conditional mutual information.

[0012] In each time slot, determine whether the maximized RSU side condition is greater than the maximized vehicle side condition mutual information;

[0013] If so, then in this time slot, the RSU perception mode, in which the vehicle's sensing information relies entirely on the RSU side perception, is adopted, and the total latency and total energy consumption of the vehicle in the RSU perception mode are calculated; if not, then in this time slot, the vehicle-RSU collaborative processing perception mode, which integrates vehicle-side perception and RSU side perception, is adopted, and the total latency and total energy consumption of the vehicle in the vehicle-RSU collaborative processing perception mode are calculated.

[0014] Calculate the total latency and total energy consumption of the vehicle in each time slot; optimize the total latency and total energy consumption to obtain the result that minimizes latency and energy consumption.

[0015] Determine if the vehicle is traveling in the last time slot. If so, the optimization ends; otherwise, the time slots are re-divided and the vehicle's position is updated.

[0016] Preferably, a perception model based on RSU-side perception and vehicle-side perception is constructed, specifically as follows:

[0017] Let there be n time slots; in each time slot, the RSU collects vehicle environmental information through sensors for RSU-side sensing; let N be the number of frequency band subcarriers allocated to each vehicle i by the RSU. i (n), the carriers allocated to communication by the RSU are The carriers assigned to sensing are Under the premise of spectrum sharing, a carrier allocation factor α is established. N , Let P be the total power supplied by RSU to vehicle i. i (n), the carriers allocated to communication by the RSU are The power allocated to sensing is Then there is

[0018] Simultaneously, vehicle-side perception is performed in each time slot. Specifically, the vehicle collects information through its own sensors, performs calculations using its vehicle computing processor, and sends the sensor data to the RSU side via its vehicle communication processor. Let N be the number of subcarriers allocated to vehicle j. j (n), the carrier wave allocated to communication for vehicle j is The carriers assigned to sensing are Then there is That is, under the premise of spectrum sharing, a carrier allocation factor β is established. N ,Right now Assume that the total power of vehicle j in time slot n is P. j (n), the carrier assigned to the vehicle for communication is P. j c (n), the power allocated to sensing is P j r (n), then P j c (n)+P j r (n)=P j (n), setting the power allocation factor β P ,β P *P j (n)=P j c (n), (1-β) p )*P k (n)=P j r (n); Further refine the power required for each subcarrier, assuming the power of the k-th sensing carrier, then allocate it to the sensing carrier. The power is: Then there is Similarly, assuming the power of the q-th communication carrier, it is allocated to the communication carrier. The power is: Then there is

[0019] For RSU-side sensing, data is transmitted via radar sensing in time slot n. A continuous OFDM signal is represented as:

[0020]

[0021] Where f c It is the carrier frequency, N s It is the number of symbols per OFDM subcarrier. This refers to the number of subcarriers allocated to sensing, where a subcarrier is defined as... It is the waveform amplitude, c k,l (n) represents the phase code of the RSU subcarrier, l is the OFDM symbol length, and T s The pulse duration is represented by t, and the timing parameter is t.

[0022] For vehicle-side perception, data is transmitted via radar sensing in time slot n. A continuous OFDM signal is represented as:

[0023]

[0024] Where f c It is the carrier frequency, N s It is the number of symbols per OFDM subcarrier. This refers to the number of subcarriers allocated to sensing, where a subcarrier is defined as... It's the amplitude, c k,l (n) represents the subcarrier phase code, Δf represents the frequency increment, l is the OFDM symbol length, and T s This represents the pulse duration.

[0025] Furthermore, based on the perception model, the perception conditional mutual information on the RSU side and the vehicle side is calculated separately. The specific steps are as follows:

[0026] Let the channel gain of the sensing model sensed on the RSU side be: The RSU radar cross section is represented by λ, where λ represents the length of the radar wave, and d represents the radar cross section. n This represents the distance detected by the radar; then there will be an echo signal. in Indicates Gaussian noise;

[0027] The calculated conditional mutual information of the RSU side is as follows:

[0028]

[0029] in The interference of RSU communication on sensing is represented by the formula: T P f represents the total duration of OFDM. k Let Δf represent the frequency of the k-th subcarrier, and Δf be the subcarrier frequency interval. Represents the Fourier transform of Gaussian white noise. It is the RSU transmit antenna gain. It is the RSU receiver antenna gain. It is the channel gain for communication;

[0030] Let the channel gain of the vehicle-side perception model be: This represents the radar cross-section of the vehicle, where λ represents the length of the vehicle's radar wave; the echo signal is then represented as: in Indicates Gaussian noise;

[0031] The calculated conditional mutual information of perception on the vehicle side is as follows:

[0032]

[0033] in The interference of vehicle communication on sensing signals is represented by the formula: In addition to interference from the vehicle's own transmitted waveform communication and sensing, the vehicle's sensing mutual information is also affected by interference from the RSU sensing waveform, which can be expressed by the formula: T P f represents the total duration of OFDM. k This represents the frequency of the k-th subcarrier; Represents the Fourier transform of Gaussian white noise. It is the gain of the vehicle's transmitting antenna. It refers to the vehicle's receiving antenna gain. It is the channel gain for communication.

[0034] Furthermore, optimizing the perception conditional mutual information to maximize the perception conditional mutual information between the RSU side and the vehicle side involves the following steps:

[0035] Constructing the first optimization function based on conditional mutual information sensed on the RSU side:

[0036]

[0037] Wherein, constraint C1 represents the range of values ​​for the subcarrier allocation for communication and sensing on the RSU side; C2 represents the power value constraint for communication and sensing subcarriers; C3 represents the maximum power constraint that the power allocation of the RSU in each time slot shall not exceed; C4 represents the minimum requirement for mutual information of sensing conditions on the RSU side in each time slot to ensure sensing quality; C5 ensures the minimum downlink communication quality requirement in each time slot, and the transmission rate shall be lower than the set threshold.

[0038] Constructing a second optimization function based on vehicle-side perception conditional mutual information:

[0039]

[0040] Wherein, constraint C1 represents the range constraint of subcarrier allocation for vehicle-side communication and sensing; C2 represents the power constraint of vehicle-side communication and sensing subcarriers; C3 represents the constraint that the power allocation of the vehicle in each time slot does not exceed the maximum power; C4 represents the minimum requirement of mutual information of vehicle-side sensing conditions in each time slot to ensure sensing quality; C5 ensures the minimum requirement of uplink communication quality in each time slot, and the transmission rate can be lower than the set threshold.

[0041] Based on the first and second optimization functions, the sensing conditional mutual information between the RSU side and the vehicle side is optimized by allocating subcarriers and power of the communication and sensing beams of the vehicle and RSU, thereby obtaining the maximized sensing conditional mutual information between the RSU side and the vehicle side.

[0042] Furthermore, in the RSU perception mode, the RSU collects vehicle environmental information through sensors, and the results calculated by the computing processor are directly returned to the vehicle side by the communication processor, thus realizing vehicle perception. In the vehicle-RSU collaborative processing perception mode, the vehicle collects vehicle environmental information through its own sensors, and sends the sensing data to the RSU side through the vehicle computing processor and measurement communication processor. The RSU integrates the vehicle environmental information collected by the vehicle with the vehicle environmental information it perceives, and then returns the data results to the vehicle side. In both perception modes, the RSU acts as the central controller for reinforcement learning, ensuring maximum perception conditional mutual information between the RSU and vehicle sides by allocating subcarriers and power for the communication and perception beams of the vehicle and RSU. After determining the perception mode based on the perception conditional mutual information, the RSU determines the allocation of computing resources. The communication and perception beams share a frequency band, and different subcarriers are allocated through OFDM technology to realize communication and perception functions.

[0043] Furthermore, the total latency and total energy consumption are calculated in RSU perception mode or vehicle-RSU cooperative processing perception mode. The specific steps are as follows:

[0044] Each vehicle's uplink assumption is modeled as a 3GPP standard path loss model:

[0045]

[0046] Where f(n) is the uplink communication frequency, denoted by d RSU-car (n) represents the distance between the vehicle and the RSU. Considering the varying distances between the vehicle and the RSU model, the uplink channel gain is:

[0047] The calculation yields:

[0048]

[0049] in The interference of the sensed signal on the communication signal in the signal transmitted by the vehicle is represented as:

[0050]

[0051] Therefore, B j =β N *N j (n)*Δf is the bandwidth occupied by RSU communication, and σ(n) is Gaussian noise;

[0052] Within the same carrier frequency band used by the RSU, the spectrum resources for communication and sensing are divided by splitting sub-carriers. The RSU uses a designated downlink to transmit sensing data back to the vehicle side. The downlink is modeled as flat fading, and the path loss of the vehicle for each n time slot is modeled as follows:

[0053]

[0054] The channel gain between the RSU and the vehicle is used To uniformly represent the communication channel gain of different subcarriers; using the formula: Indicates the channel gain for different time slots;

[0055] Calculate the rate based on the channel gain:

[0056]

[0057] in The interference of sensing signals on communication signals in V2I is represented as:

[0058]

[0059] B i =α N *N i (n)*Δf is the bandwidth occupied by RSU communication, and σ(n) is Gaussian noise;

[0060] The computing resources on the RSU side and the onboard computing resources are integrated and allocated; let f(n) be the CPU resources for the RSU's processing tasks in each time slot n, and f(n) be the CPU resources used to process the RSU's own sensing information. RSU (n), where f is the computational resource used to process the vehicle-RSU perception mode. car Adding the resource allocation factor μ to (n), we have μf(n)=f RSU (n), (1-μ)f(n)=f car (n);

[0061] If RSU sensing mode is used, then:

[0062] In RSU perception mode, assuming vehicle m uses RSU perception mode in time slot n, the amount of sensor data that the RSU needs to sense and process for vehicle m in each time slot is... The delay that needs to be handled is In each time slot, after processing the data, the RSU needs to send it to each vehicle. Let the processed data for each vehicle be... The downlink transmission delay is:

[0063]

[0064] The delay of vehicle m in RSU full processing mode within time slot n is calculated as follows:

[0065]

[0066] Regarding energy consumption, the energy consumption of vehicle m in each time slot n using the RSU perception model There are two parts. One part is the computational cost of RSU processing the perception task, where κ(n) is the computational resource factor.

[0067]

[0068] The other part is the energy consumption of transmitting communication resource blocks to the vehicle.

[0069]

[0070] The total latency of vehicle m in RSU perception processing mode within time slot n is calculated. for:

[0071]

[0072] If the vehicle-RSU cooperative perception mode is adopted, then:

[0073] In the vehicle-RSU cooperative sensing mode, let the size of the sensing data received by vehicle m in time slot n be... In each time slot, the RSU needs to sense the size of the sensor data that vehicle m needs to process. The vehicle transmits sensor data to the RSU's communication computing resources for processing, with an uplink transmission latency of [missing information]. The results of data processing uploaded to RSU's computing resources are as follows: In vehicle-RSU mode, the processing delay for each vehicle m in time slot n is...

[0074] In terms of energy consumption, the energy consumption of vehicle m in each time slot n when using vehicle-RSU cooperative processing consists of two parts: one part is the communication consumption of vehicle m transmitting perception information to the RSU. Part of it is the computational cost of the RSU processing the raw sensing data sent by the vehicle.

[0075]

[0076] There is energy consumption in the vehicle-RSU collaborative processing.

[0077]

[0078] Furthermore, the total latency and total energy consumption of the vehicle in each time slot are calculated. The specific steps are as follows: Suppose that vehicle m can choose two perception processing modes in each time slot n, and let η represent the perception mode selection method. When η=1, it means that the RSU perception mode is used, and when η=0, it means that the vehicle-RSU cooperative perception mode is used. The resulting processing latency τ m (n) is represented as:

[0079]

[0080] Then vehicle m consumes energy e in each time slot n. m (n):

[0081]

[0082] The total vehicle delay T is obtained for each time slot:

[0083]

[0084] The total energy consumption result E is obtained:

[0085]

[0086] Furthermore, optimizing the total latency and total energy consumption results to minimize the latency and energy consumption results involves the following steps:

[0087] Suppose that the vehicle-infrastructure cooperative system has a total of M ∈ {1,2,…,m…,M} vehicles; introduce T max and E max As the maximum delay and energy consumption, the optimized delay and energy consumption are normalized, and σ is introduced. As a weighting factor balancing latency and energy consumption, a third optimization function is constructed based on the total latency and total energy consumption results:

[0088]

[0089]

[0090] Wherein, constraint C1 represents the value limit of the computing resource allocation factor; C2 represents that the energy consumption of each time slot cannot exceed the maximum value; C3 represents the processing delay constraint under each time slot, ensuring that the processing delay can meet the service requirements; C4 represents the binary nature of the perception mode selection, either choosing the RSU perception mode or choosing the vehicle-RSU cooperative perception mode; C5 represents the delay and energy consumption weight factor allocation constraint of the optimization problem.

[0091] Based on the third optimization function, the total latency and total energy consumption results are optimized by calculating resource allocation, resulting in minimized latency and energy consumption.

[0092] Furthermore, the mutual information of perception conditions between the RSU and vehicle sides is optimized by allocating subcarriers and power of the communication and sensing beams of the vehicle and RSU. The specific optimization steps are as follows:

[0093] Initialize a quadruple (S, A, T, R) encompassing the system state set S, action set A, state transition probability set T, and real-valued reward function R. State S describes the environment, the specific situation or state the agent is in when performing an action; action A is the decision or behavior taken by the agent in a given state; the state transition probability set T typically describes the probability distribution of the agent transitioning from one state to another after performing a specific action; reward R is the core feedback signal in reinforcement learning, used to evaluate the effectiveness of the agent's specific action in a specific state; the goal of the MDP process is to find a policy π that maximizes the long-term cumulative reward, expressed by the formula: Where γ∈(0,1) is a discount factor used to balance the importance of immediate rewards and future rewards;

[0094] The optimization process of the perception conditional mutual information between the RSU side and the vehicle side is set as a Markov decision process using a low-level deep learning network DDPG. It is assumed that the perception conditional mutual information on the vehicle side is affected by the power allocation of communication and perception on the vehicle side and the subcarrier allocation of communication and perception on the vehicle side; the perception conditional mutual information on the RSU side is affected by the power allocation of communication and perception on the RSU side and the subcarrier allocation of communication and perception on the vehicle side.

[0095] The DDPG algorithm is used to instantiate the perception conditional mutual information on the vehicle side and the RSU side respectively, and the DDPG framework on the vehicle side and the DDPG framework on the RSU side are updated simultaneously.

[0096] In RSU-perceptual conditional mutual information DDPG reinforcement learning, the task is defined as maximizing the perceptual conditional mutual information on the RSU side; the state S in time slot n is defined as: These include the power allocated to each vehicle by the RSU, the communication allocated power, sensing allocated power, communication allocated subcarriers, sensing allocated subcarriers, and the downlink communication rate transmitted by the RSU to the vehicles; the action space is defined as follows: This includes the communication-sensing power allocation factor and its subcarrier allocation factor for each RSU (Roadside Unit). The power allocation factor determines the power allocation ratio between communication and sensing on the RSU side, while the subcarrier allocation factor determines the subcarrier allocation ratio between the communication beam and sensing beam on the RSU side. Both the power allocation factor and the subcarrier allocation factor are continuous values, ranging from (0,1). In the RSU-side optimization problem, the reward is defined as... The conditional mutual information on the RSU side of each time slot is used as the reward value of the reward function, and the maximum conditional mutual information is obtained according to different power and subcarrier allocation conditions.

[0097] In vehicle-side perception conditional mutual information (DDPG) reinforcement learning, the task is defined as maximizing the conditional mutual information on the vehicle side; the state S in time slot n is defined as: These include the vehicle's communication allocation power, sensing allocation power, communication allocation subcarriers, sensing allocation subcarriers, and the vehicle's uplink communication rate; the definition of the action space is... This includes the vehicle's power allocation factor and subcarrier allocation factor. The power allocation factor determines the power allocation ratio between communication and sensing on the vehicle side, while the subcarrier allocation factor determines the subcarrier allocation ratio between the communication beam and sensing beam on the vehicle side. Both the power allocation factor and the subcarrier allocation factor are continuous values, ranging from (0,1). The reward is defined as... The vehicle-side conditional mutual information of each time slot is used as the reward value of the reward function, and the maximum conditional mutual information is obtained according to different power and subcarrier allocation conditions.

[0098] Furthermore, the total latency and total energy consumption are optimized by calculating resource allocation. The specific steps are as follows:

[0099] A high-level deep learning network is used to set the optimization process of total latency and total energy consumption as a Markov decision process; in the n time slot, the state S is defined as: s n =(S m (n),η,μ(n),E(n),B(n)), the state includes the size of the sensing data, the selection strategy of the sensing mode, the computing resources allocated by RSU to the two modes respectively, the remaining energy of RSU and the remaining bandwidth of processing; action space a n =(μ(n)) includes a computational resource allocation factor, which determines the computational resources allocated to RSU perception and vehicle-RSU cooperative perception. This factor is a continuous number with values ​​between (0,1). In the reward design of high-level reinforcement learning networks, the optimization problem is to minimize latency and energy consumption. The reward design is the opposite of the optimization problem. The reward value is optimized based on the allocation of computing resources in different modes.

[0100] The beneficial effects of this invention are as follows:

[0101] This invention establishes an integrated collaborative processing environment based on a vehicle-infrastructure (V2I) cooperative system. It constructs a perception model by combining the sensor receivers (JCR) of the onboard intelligent sensing and computing (ISC) device and the roadside unit (RSU) intelligent sensing and computing (LCU) device. This ensures optimized perception quality on both the onboard and RSU sides. Based on this optimized perception quality, it selects either an RSU perception mode or a vehicle-RSU joint perception mode to minimize latency and energy consumption of the entire V2I-Roadside Infrastructure (Roadside Infrastructure) joint system. Therefore, this invention solves the problems of high computational complexity and susceptibility to local optima in existing technologies. Attached Figure Description

[0102] Figure 1 This is a flowchart illustrating the integrated transmission optimization method for vehicle networking based on sensor quality according to the present invention.

[0103] Figure 2 This is a schematic diagram of the vehicle-infrastructure cooperation system constructed by the present invention.

[0104] Figure 3 This is a block diagram illustrating the optimization algorithm used in this invention. Detailed Implementation

[0105] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0106] Example 1

[0107] like Figure 1 As shown, a vehicle-to-everything (V2X) communication, sensing, and computing integrated transmission optimization method based on sensor quality includes the following specific steps:

[0108] like Figure 2 As shown, a vehicle-infrastructure cooperative system consisting of several roadside units (RSUs) and several vehicles is established in the target environment. Each vehicle and RSU is equipped with an intelligent sensing and computing integrated device (ISCC) consisting of a communication processor, a computing processor, and sensors.

[0109] In this embodiment, the sensors include millimeter-wave radar, cameras, etc. Both the vehicle and the RSU can collect environmental information about the vehicle's operation and calculate the optimal conditions for autonomous driving, enhancing road visibility and providing early warnings. The communication and sensing beams share a frequency band, and different subcarriers are allocated through OFDM technology to achieve communication and sensing functions.

[0110] Divide time slots and update vehicle positions;

[0111] Construct a perception model based on RSU-side perception and vehicle-side perception;

[0112] The perception conditional mutual information of the RSU side and the vehicle side is calculated based on the perception model; the perception conditional mutual information of the RSU side and the vehicle side is maximized by optimizing the perception conditional mutual information.

[0113] In each time slot, determine whether the maximized RSU side condition is greater than the maximized vehicle side condition mutual information;

[0114] If so, then in this time slot, the RSU perception mode, in which the vehicle's sensing information relies entirely on the RSU side perception, is adopted, and the total latency and total energy consumption of the vehicle in the RSU perception mode are calculated; if not, then in this time slot, the vehicle-RSU collaborative processing perception mode, which integrates vehicle-side perception and RSU side perception, is adopted, and the total latency and total energy consumption of the vehicle in the vehicle-RSU collaborative processing perception mode are calculated.

[0115] Calculate the total latency and total energy consumption of the vehicle in each time slot; optimize the total latency and total energy consumption to obtain the result that minimizes latency and energy consumption.

[0116] Determine if the vehicle is traveling in the last time slot. If so, the optimization ends; otherwise, the time slots are re-divided and the vehicle's position is updated.

[0117] Example 2

[0118] More specifically, in one embodiment, a perception model based on RSU-side perception and vehicle-side perception is constructed as follows:

[0119] Let there be n time slots; in each time slot, the RSU collects vehicle environmental information through sensors for RSU-side sensing; let N be the number of frequency band subcarriers allocated to each vehicle i by the RSU. i (n), the carriers allocated to communication by the RSU are The carriers assigned to sensing are Under the premise of spectrum sharing, a carrier allocation factor α is established. N , Let P be the total power supplied by RSU to vehicle i. i (n), the carriers allocated to communication by the RSU are The power allocated to sensing is Then there is

[0120] Simultaneously, vehicle-side perception is performed in each time slot. Specifically, the vehicle collects information through its own sensors, performs calculations using its vehicle computing processor, and sends the sensor data to the RSU side via its vehicle communication processor. Let N be the number of subcarriers allocated to vehicle j. j (n), the carrier wave allocated to communication for vehicle j is The carriers assigned to sensing are Then there is That is, under the premise of spectrum sharing, a carrier allocation factor β is established. N ,Right now Assume that the total power of vehicle j in time slot n is P. j (n), the carrier assigned to the vehicle for communication is P. j c (n), the power allocated to sensing is P j r (n), then P j c (n)+P j r (n)=P j (n), setting the power allocation factor β P ,β P *P j (n)=P j c (n), (1-β) P )*P j (n)=P j r (n); Further refine the power required for each subcarrier, assuming the power of the k-th sensing carrier, then allocate it to the sensing carrier. The power is: Then there is Similarly, assuming the power of the q-th communication carrier, it is allocated to the communication carrier. The power is: Then there is

[0121] For RSU-side sensing, data is transmitted via radar sensing in time slot n. A continuous OFDM signal is represented as:

[0122]

[0123] Where f c It is the carrier frequency, N s It is the number of symbols per OFDM subcarrier. This refers to the number of subcarriers allocated to sensing, where a subcarrier is defined as... It is the waveform amplitude, c k,l (n) represents the phase code of the RSU subcarrier, l is the OFDM symbol length, and T s The pulse duration is represented by t, and the timing parameter is t.

[0124] For vehicle-side perception, data is transmitted via radar sensing in time slot n. A continuous OFDM signal is represented as:

[0125]

[0126] Where f c It is the carrier frequency, N s It is the number of symbols per OFDM subcarrier. This refers to the number of subcarriers allocated to sensing, where a subcarrier is defined as... It's the amplitude, c k,l (n) represents the subcarrier phase code, Δf represents the frequency increment, l is the OFDM symbol length, and T s This represents the pulse duration.

[0127] In one specific embodiment, the perception conditional mutual information of the RSU side and the vehicle side is calculated based on the perception model, and the specific steps are as follows:

[0128] Let the channel gain of the sensing model sensed on the RSU side be: The RSU radar cross section is represented by λ, where λ represents the length of the radar wave, and d represents the radar cross section. n This represents the distance detected by the radar; then there will be an echo signal. in Indicates Gaussian noise;

[0129] The calculated conditional mutual information of the RSU side is as follows:

[0130]

[0131] in The interference of RSU communication on sensing is represented by the formula: T P f represents the total duration of OFDM. k Let Δf represent the frequency of the k-th subcarrier, and Δf be the subcarrier frequency interval. Represents the Fourier transform of Gaussian white noise. It is the RSU transmit antenna gain. It is the RSU receiver antenna gain. It is the channel gain for communication;

[0132] Let the channel gain of the vehicle-side perception model be: This represents the radar cross-section of the vehicle, where λ represents the length of the vehicle's radar wave; the echo signal is then represented as: in Indicates Gaussian noise;

[0133] The calculated conditional mutual information of perception on the vehicle side is as follows:

[0134]

[0135] in The interference of vehicle communication on sensing signals is represented by the formula: In addition to interference from the vehicle's own transmitted waveform communication and sensing, the vehicle's sensing mutual information is also affected by interference from the RSU sensing waveform, which can be expressed by the formula: T P f represents the total duration of OFDM. k This represents the frequency of the k-th subcarrier; Represents the Fourier transform of Gaussian white noise. It is the gain of the vehicle's transmitting antenna. It refers to the vehicle's receiving antenna gain. It is the channel gain for communication.

[0136] In one specific embodiment, optimizing the perception conditional mutual information to obtain the maximized perception conditional mutual information between the RSU side and the vehicle side involves the following steps:

[0137] Constructing the first optimization function based on conditional mutual information sensed on the RSU side:

[0138]

[0139] Wherein, constraint C1 represents the range of values ​​for the subcarrier allocation for communication and sensing on the RSU side; C2 represents the power value constraint for communication and sensing subcarriers; C3 represents the maximum power constraint that the power allocation of the RSU in each time slot shall not exceed; C4 represents the minimum requirement for mutual information of sensing conditions on the RSU side in each time slot to ensure sensing quality; C5 ensures the minimum downlink communication quality requirement in each time slot, and the transmission rate shall be lower than the set threshold.

[0140] Constructing a second optimization function based on vehicle-side perception conditional mutual information:

[0141]

[0142] Wherein, constraint C1 represents the range constraint of subcarrier allocation for vehicle-side communication and sensing; C2 represents the power constraint of vehicle-side communication and sensing subcarriers; C3 represents the constraint that the power allocation of the vehicle in each time slot does not exceed the maximum power; C4 represents the minimum requirement of mutual information of vehicle-side sensing conditions in each time slot to ensure sensing quality; C5 ensures the minimum requirement of uplink communication quality in each time slot, and the transmission rate can be lower than the set threshold.

[0143] Based on the first and second optimization functions, the sensing conditional mutual information between the RSU side and the vehicle side is optimized by allocating subcarriers and power of the communication and sensing beams of the vehicle and RSU, thereby obtaining the maximized sensing conditional mutual information between the RSU side and the vehicle side.

[0144] In one specific embodiment, in the RSU perception mode, the RSU collects vehicle environmental information through sensors, and the results calculated by the computing processor are directly returned to the vehicle side by the communication processor to achieve vehicle perception. In the vehicle-RSU collaborative processing perception mode, the vehicle collects vehicle environmental information through its own sensors, and sends the sensing data to the RSU side through the vehicle computing processor and measurement communication processor. The RSU integrates the vehicle environmental information collected by the vehicle and the vehicle environmental information it perceives, and then returns the data results to the vehicle side. In both perception modes, the RSU acts as the central controller for reinforcement learning. By allocating subcarriers and power of the communication and perception beams of the vehicle and the RSU, it ensures the maximum mutual information of perception conditions between the RSU side and the vehicle side. After determining the perception mode based on the mutual information of perception conditions, the RSU determines the allocation of computing resources. The communication and perception beams share a frequency band, and different subcarriers are allocated through OFDM technology to achieve communication and perception functions.

[0145] In one specific embodiment, the total latency and total energy consumption are calculated in RSU perception mode or vehicle-RSU cooperative processing perception mode, and the specific steps are as follows:

[0146] Each vehicle's uplink assumption is modeled as a 3GPP standard path loss model:

[0147]

[0148] Where f(n) is the uplink communication frequency, denoted by d RSU-car (n) represents the distance between the vehicle and the RSU. Considering the varying distances between the vehicle and the RSU model, the uplink channel gain is:

[0149] The calculation yields:

[0150]

[0151] in The interference of the sensed signal on the communication signal in the signal transmitted by the vehicle is represented as:

[0152]

[0153] Therefore, B j =β N *N j (n)*Δf is the bandwidth occupied by RSU communication, and σ(n) is Gaussian noise;

[0154] Within the same carrier frequency band used by the RSU, the spectrum resources for communication and sensing are divided by splitting sub-carriers. The RSU uses a designated downlink to transmit sensing data back to the vehicle side. The downlink is modeled as flat fading, and the path loss of the vehicle for each n time slot is modeled as follows:

[0155]

[0156] The channel gain between the RSU and the vehicle is used To uniformly represent the communication channel gain of different subcarriers; using the formula: Indicates the channel gain for different time slots;

[0157] Calculate the rate based on the channel gain:

[0158]

[0159] in The interference of sensing signals on communication signals in V2I is represented as:

[0160]

[0161] B i =α N *N i (n)*Δf is the bandwidth occupied by RSU communication, and σ(n) is Gaussian noise;

[0162] The computing resources on the RSU side and the onboard computing resources are integrated and allocated; let f(n) be the CPU resources for the RSU's processing tasks in each time slot n, and f(n) be the CPU resources used to process the RSU's own sensing information. RSU (n), where f is the computational resource used to process the vehicle-RSU perception mode. car Adding the resource allocation factor μ to (n), we have μf(n)=f RSU (n), (1-μ)f(n)=f car (n);

[0163] If RSU sensing mode is used, then:

[0164] In RSU perception mode, assuming vehicle m uses RSU perception mode in time slot n, the amount of sensor data that the RSU needs to sense and process for vehicle m in each time slot is... The delay that needs to be handled is In each time slot, after processing the data, the RSU needs to send it to each vehicle. Let the processed data for each vehicle be... The downlink transmission delay is:

[0165]

[0166] The delay of vehicle m in RSU full processing mode within time slot n is calculated as follows:

[0167]

[0168] Regarding energy consumption, the energy consumption of vehicle m in each time slot n using the RSU perception model There are two parts. One part is the computational cost of RSU processing the perception task, where K(n) is the computational resource factor.

[0169]

[0170] The other part is the energy consumption of transmitting communication resource blocks to the vehicle.

[0171]

[0172] The total latency of vehicle m in RSU perception processing mode within time slot n is calculated. for:

[0173]

[0174] If the vehicle-RSU cooperative perception mode is adopted, then:

[0175] In the vehicle-RSU cooperative sensing mode, let the size of the sensing data received by vehicle m in time slot n be... In each time slot, the RSU needs to sense the size of the sensor data that vehicle m needs to process. The vehicle transmits sensor data to the RSU's communication computing resources for processing, with an uplink transmission latency of [missing information]. The results of data processing uploaded to RSU's computing resources are as follows: In vehicle-RSU mode, the processing delay for each vehicle m in time slot n is...

[0176] In terms of energy consumption, the energy consumption of vehicle m in each time slot n when using vehicle-RSU cooperative processing consists of two parts: one part is the communication consumption of vehicle m transmitting perception information to the RSU. Part of it is the computational cost of the RSU processing the raw sensing data sent by the vehicle.

[0177]

[0178] There is energy consumption in the vehicle-RSU collaborative processing.

[0179]

[0180] In one specific embodiment, the total latency and total energy consumption of the vehicle in each time slot are calculated. The specific steps are as follows: Suppose that vehicle m can select two perception processing modes in each time slot n, and let η represent the perception mode selection method. When η=1, it means that the RSU perception mode is used, and when η=0, it means that the vehicle-RSU cooperative perception mode is used. The resulting processing latency τ m (n) is represented as:

[0181]

[0182] Then vehicle m consumes energy e in each time slot n. m (n):

[0183]

[0184] The total vehicle delay T is obtained for each time slot:

[0185]

[0186] The total energy consumption result E is obtained:

[0187]

[0188] In one specific embodiment, optimizing the total latency and total energy consumption results to obtain the minimized latency and energy consumption results involves the following steps:

[0189] Suppose that the vehicle-infrastructure cooperative system has a total of M ∈ {1,2,…,m…,M} vehicles; introduce T max and E max As the maximum delay and energy consumption, the optimized delay and energy consumption are normalized, and σ is introduced. As a weighting factor balancing latency and energy consumption, a third optimization function is constructed based on the total latency and total energy consumption results:

[0190]

[0191] Wherein, constraint c1 represents the value limit of the computational resource allocation factor; C2 represents that the energy consumption of each time slot cannot exceed the maximum value; C3 represents the processing delay constraint under each time slot, ensuring that the processing delay can meet the service requirements; C4 represents the binary nature of the perception mode selection, either choosing the RSU perception mode or choosing the vehicle-RSU cooperative perception mode; C5 represents the delay and energy consumption weight factor allocation constraints of the optimization problem.

[0192] Based on the third optimization function, the total latency and total energy consumption results are optimized by calculating resource allocation, resulting in minimized latency and energy consumption.

[0193] Example 3

[0194] In one specific embodiment, such as Figure 3 As shown, the sensing conditional mutual information between the RSU and vehicle sides is optimized by allocating subcarriers and power of the communication and sensing beams of the vehicle and RSU. The specific optimization steps are as follows:

[0195] Initialize a quadruple (S, A, T, R) encompassing the system state set S, action set A, state transition probability set T, and real-valued reward function R. State S describes the environment, the specific situation or state the agent is in when performing an action; action A is the decision or behavior taken by the agent in a given state; the state transition probability set T typically describes the probability distribution of the agent transitioning from one state to another after performing a specific action; reward R is the core feedback signal in reinforcement learning, used to evaluate the effectiveness of the agent's specific action in a specific state; the goal of the MDP process is to find a policy π that maximizes the long-term cumulative reward, expressed by the formula: Where γ∈(0,1) is a discount factor used to balance the importance of immediate rewards and future rewards;

[0196] The optimization process of the perception conditional mutual information between the RSU side and the vehicle side is set as a Markov decision process using a low-level deep learning network DDPG. It is assumed that the perception conditional mutual information on the vehicle side is affected by the power allocation of communication and perception on the vehicle side and the subcarrier allocation of communication and perception on the vehicle side; the perception conditional mutual information on the RSU side is affected by the power allocation of communication and perception on the RSU side and the subcarrier allocation of communication and perception on the vehicle side.

[0197] The DDPG algorithm is used to instantiate the perception conditional mutual information on the vehicle side and the RSU side respectively, and the DDPG framework on the vehicle side and the DDPG framework on the RSU side are updated simultaneously.

[0198] In RSU-perceptual conditional mutual information DDPG reinforcement learning, the task is defined as maximizing the perceptual conditional mutual information on the RSU side; the state S in time slot n is defined as: These include the power allocated to each vehicle by the RSU, the communication allocated power, sensing allocated power, communication allocated subcarriers, sensing allocated subcarriers, and the downlink communication rate transmitted by the RSU to the vehicles; the action space is defined as follows: This includes the communication-sensing power allocation factor and its subcarrier allocation factor for each RSU (Roadside Unit). The power allocation factor determines the power allocation ratio between communication and sensing on the RSU side, while the subcarrier allocation factor determines the subcarrier allocation ratio between the communication beam and sensing beam on the RSU side. Both the power allocation factor and the subcarrier allocation factor are continuous values, ranging from (0,1). In the RSU-side optimization problem, the reward is defined as... The conditional mutual information on the RSU side of each time slot is used as the reward value of the reward function, and the maximum conditional mutual information is obtained according to different power and subcarrier allocation conditions.

[0199] In vehicle-side perception conditional mutual information (DDPG) reinforcement learning, the task is defined as maximizing the conditional mutual information on the vehicle side; the state S in time slot n is defined as: These include the vehicle's communication allocation power, sensing allocation power, communication allocation subcarriers, sensing allocation subcarriers, and the vehicle's uplink communication rate; the definition of the action space is... This includes the vehicle's power allocation factor and subcarrier allocation factor. The power allocation factor determines the power allocation ratio between communication and sensing on the vehicle side, while the subcarrier allocation factor determines the subcarrier allocation ratio between the communication beam and sensing beam on the vehicle side. Both the power allocation factor and the subcarrier allocation factor are continuous values, ranging from (0,1). The reward is defined as... The vehicle-side conditional mutual information of each time slot is used as the reward value of the reward function, and the maximum conditional mutual information is obtained according to different power and subcarrier allocation conditions.

[0200] In one specific embodiment, the total latency and total energy consumption results are optimized by calculating resource allocation. The specific steps are as follows:

[0201] A high-level deep learning network is used to set the optimization process of total latency and total energy consumption as a Markov decision process; in the n time slot, the state S is defined as: s n =(S m (n),η,μ(n),E(n),B(n)), the state includes the size of the sensing data, the selection strategy of the sensing mode, the computing resources allocated by RSU to the two modes respectively, the remaining energy of RSU and the remaining bandwidth of processing; action space a n =(μ(n)) includes a computational resource allocation factor, which determines the computational resources allocated to RSU perception and vehicle-RSU cooperative perception. This factor is a continuous number with values ​​between (0,1). In the reward design of high-level reinforcement learning networks, the optimization problem is to minimize latency and energy consumption. The reward design is the opposite of the optimization problem. The reward value is optimized based on the allocation of computing resources in different modes.

[0202] Therefore, this invention establishes a vehicle-infrastructure (RSU) cooperative sensing system, which combines the communication and sensing integrated technology ISAC with the edge computing technology MEC, and proposes two sensing and computing modes: RSU sensing and RSU-vehicle cooperative sensing. The computing resource allocation process of the two modes is modeled as a Markov decision process and solved using a reinforcement learning algorithm, so as to minimize the total latency and total energy consumption of the system.

[0203] This invention also ensures the sensing quality of the vehicle-infrastructure (RSU) cooperative sensing system. By optimizing the sensing conditional mutual information of the vehicle side and the RSU side as sensing quality indicators, the communication sensing subcarriers and their power allocation process of the vehicle side and the RSU side are modeled as Markov decision processes to maximize the sensing conditional mutual information of the vehicle side and the RSU side, thereby meeting the requirements for sensing quality.

[0204] This invention employs a two-layer reinforcement learning network to optimize the proposed problem. The lower-layer reinforcement learning network aims to optimize the perception conditional mutual information between the vehicle and RSU sides, while the higher-layer reinforcement learning network aims to minimize the system's total latency and energy consumption. The decision and reward functions of the higher layer change according to the different decision factors passed from the lower layer, providing a more suitable perception mode strategy.

[0205] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A vehicle-to-everything (V2X) communication, sensing, and computational integration transmission optimization method based on sensor quality, characterized in that: The specific steps include the following: In the target environment, a vehicle-infrastructure cooperative system consisting of several roadside units (RSUs) and several vehicles is established, in which each vehicle and RSU is equipped with an intelligent integrated communication, sensing, and computing device consisting of a communication processor, a computing processor, and sensors. Divide time slots and update vehicle positions; Construct a perception model based on RSU-side perception and vehicle-side perception; The perception conditional mutual information of the RSU side and the vehicle side is calculated based on the perception model; the perception conditional mutual information of the RSU side and the vehicle side is maximized by optimizing the perception conditional mutual information. In each time slot, determine whether the maximized RSU-side conditional mutual information is greater than the maximized vehicle-side conditional mutual information; If so, then in this time slot, the RSU perception mode, in which the vehicle's sensing information relies entirely on the RSU side perception, is adopted, and the total latency and total energy consumption of the vehicle in the RSU perception mode are calculated; if not, then in this time slot, the vehicle-RSU collaborative processing perception mode, which integrates vehicle-side perception and RSU side perception, is adopted, and the total latency and total energy consumption of the vehicle in the vehicle-RSU collaborative processing perception mode are calculated. Calculate the total latency and total energy consumption of the vehicle in each time slot; optimize the total latency and total energy consumption to obtain the result that minimizes latency and energy consumption. Determine if the vehicle is traveling in the last time slot. If so, the optimization ends; otherwise, the time slots are re-divided and the vehicle's position is updated.

2. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 1, characterized in that: Construct a perception model based on RSU-side perception and vehicle-side perception, specifically as follows: Let the time slot be In each time slot, the RSU collects vehicle environmental information through sensors to perform RSU-side perception. The number of frequency band subcarriers allocated to each vehicle i by the RSU is denoted as follows: The carriers allocated to communication by the RSU are The carriers assigned to the sensing are , = Under the premise of spectrum sharing, a carrier allocation factor is established. , , ; Record RSU to vehicle The total power is The power allocated to communication by the RSU is The power allocated to sensing is Then there is ; Simultaneously, vehicle-side perception is performed in each time slot. Specifically, the vehicle collects information through its own sensors, performs calculations using its vehicle computing processor, and sends the sensor data to the RSU side via its vehicle communication processor. Let the number of subcarriers allocated to vehicle j be... The communication carrier assigned to vehicle j is The carriers assigned to the sensing are Then there is = That is, under the premise of spectrum sharing, a carrier allocation factor is established. ,Right now , Assume that the total power of vehicle j in time slot n is The power allocated to the vehicle for communication is The power allocated to sensing is Then there is Establish a power allocation factor , , ; Further refining the power required for each subcarrier, let's assume it's allocated to the first... The power of each sensing carrier is: Then there is Similarly, suppose the first The power of each communication carrier is allocated to the communication carrier. The power is: Then there is ; For RSU-side sensing, data is transmitted via radar sensing in time slot n. A continuous OFDM signal is represented as: in It is the carrier frequency. It is the number of symbols per OFDM subcarrier. This refers to the number of subcarriers allocated to sensing, where a subcarrier is defined as... , It is the waveform amplitude. Phase code representing RSU subcarriers, OFDM symbol length, The pulse duration is represented by t, and the timing parameter is t. For vehicle-side perception, data is transmitted via radar sensing in time slot n. A continuous OFDM signal is represented as: in It is the carrier frequency. It is the number of symbols per OFDM subcarrier. This refers to the number of subcarriers allocated to sensing, where a subcarrier is defined as... , It's the amplitude. Represents frequency increment. OFDM symbol length, This represents the pulse duration.

3. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 2, characterized in that: The perception conditional mutual information is calculated based on the perception model on both the RSU side and the vehicle side. The specific steps are as follows: Let the channel gain of the sensing model sensed on the RSU side be: , Indicates the RSU radar cross section. Represents the length of the radar wave. This represents the distance detected by the radar; then there will be an echo signal. ,in The calculated conditional mutual information of the RSU side is as follows: in The interference of RSU communication on sensing is represented by the formula: Indicates the total duration of OFDM. This represents the frequency of the k-th subcarrier. It is the subcarrier frequency spacing. The Fourier transform of Gaussian white noise, It is the RSU transmit antenna gain. It is the RSU receiver antenna gain. It is the channel gain for communication; Let the channel gain of the vehicle-side perception model be: , This represents the radar cross-section of the vehicle. The length of the vehicle radar wave represents the echo signal, which is then represented as: ,in ; The calculated conditional mutual information of perception on the vehicle side is as follows: in The interference of vehicle communication on sensing signals is represented by the formula: In addition to interference from the vehicle's own transmitted waveform communication and sensing, the vehicle's sensing mutual information is also affected by interference from the RSU sensing waveform, which can be expressed by the formula: , Indicates the total duration of OFDM. This represents the frequency of the k-th subcarrier; The Fourier transform of Gaussian white noise, It is the gain of the vehicle's transmitting antenna. It refers to the vehicle's receiving antenna gain. It is the channel gain for communication.

4. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 3, characterized in that: To optimize the perception conditional mutual information and obtain the maximum perception conditional mutual information between the RSU side and the vehicle side, the specific steps are as follows: Constructing the first optimization function based on conditional mutual information sensed on the RSU side: P1: Among the constraints This represents the range constraints for subcarrier allocation values ​​for communication and sensing on the RSU side. Represents the power constraints for communication and sensing subcarriers; This means that the power allocation of the RSU in each time slot does not exceed the maximum power constraint; This represents the minimum requirement for RSU-side sensing conditional mutual information in each time slot to ensure sensing quality; Ensure that the downlink communication quality meets the minimum requirements for each time slot, and that the transmission rate does not fall below the set threshold; Constructing a second optimization function based on vehicle-side perception conditional mutual information: P2: Among the constraints This represents the constraint on the range of subcarrier allocation values ​​for vehicle-side communication and sensing. This represents the power constraints for the vehicle-side communication and sensing subcarriers; This indicates that the power allocation of the vehicle in each time slot does not exceed the maximum power constraint; This represents the minimum requirement for vehicle-side sensing conditional mutual information for each time slot to ensure sensing quality; Ensure that the uplink communication quality meets the minimum requirements for each time slot, and that the transmission rate does not fall below the set threshold; Based on the first and second optimization functions, the sensing conditional mutual information between the RSU side and the vehicle side is optimized by allocating subcarriers and power of the communication and sensing beams of the vehicle and RSU, thereby obtaining the maximized sensing conditional mutual information between the RSU side and the vehicle side.

5. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 1, characterized in that: In the RSU perception mode, the RSU collects vehicle environmental information through sensors, and the results calculated by the computing processor are directly returned to the vehicle side by the communication processor, thus realizing vehicle perception. In the vehicle-RSU collaborative processing perception mode, the vehicle collects vehicle environmental information through its own sensors, and sends the sensing data to the RSU side through the vehicle computing processor and measurement communication processor. The RSU integrates the vehicle environmental information collected by the vehicle and the vehicle environmental information it perceives, and then returns the data results to the vehicle side. In both perception modes, the RSU acts as the central controller for reinforcement learning. By allocating subcarriers and power for the communication and perception beams of the vehicle and the RSU, it ensures the maximum mutual information of perception conditions between the RSU and the vehicle side. After determining the perception mode based on the mutual information of perception conditions, the RSU determines the allocation of computing resources. The communication and perception beams share a frequency band, and different subcarriers are allocated through OFDM technology to realize communication and perception functions.

6. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 4, characterized in that: The specific steps for calculating the total latency and total energy consumption in RSU perception mode or vehicle-RSU cooperative processing perception mode are as follows: Each vehicle's uplink assumption is modeled as a 3GPP standard path loss model: in It is the frequency for uplink communication, used Considering the distance variations between the vehicle and the RSU model, the uplink channel gain is: The calculation yields: in The interference of the sensed signal on the communication signal in the signal transmitted by the vehicle is represented as: Therefore, we can conclude that... Bandwidth used for RSU communication ; Within the same carrier band used by the RSU, the spectrum resources for communication and sensing are divided by splitting sub-carriers. The RSU uses a designated downlink to send the sensing data back to the vehicle side; the downlink is modeled as flat fading, assuming the vehicle uses each The path loss of a time slot is modeled as follows: The channel gain between the RSU and the vehicle is used To uniformly represent the communication channel gain of different subcarriers; using the formula: Indicates the channel gain for different time slots; Calculate the rate based on the channel gain: in The interference of sensing signals on communication signals in V2I is represented as: Bandwidth used for RSU communication ; Integrate and allocate computing resources on the RSU side and onboard computing resources; assuming that in each time slot n, the CPU resources for the RSU's processing tasks are... The CPU resources used to process the RSU's own sensor information are The computing resources used to process vehicle-RSU perception modes are Add computing resource allocation factor That is, , ; If RSU sensing mode is used, then: In RSU perception mode, assuming vehicle m uses RSU perception mode in time slot n, the amount of sensor data that the RSU needs to sense and process for vehicle m in each time slot is... The delay that needs to be handled is In each time slot, after processing the data, the RSU needs to send it to each vehicle. Let the processed data for each vehicle be... Then the downlink transmission delay is: The delay of vehicle m in RSU full processing mode within time slot n is calculated as follows: In terms of energy consumption, each time slot Energy consumption of vehicle m using the RSU perception model There are two parts. One part is the computational cost of RSU processing the perception task, in which... To calculate resource factors; The other part is the energy consumption of transmitting communication resource blocks to the vehicle. : The total latency of vehicle m in RSU perception processing mode within time slot n is calculated. for: ； If the vehicle-RSU cooperative perception mode is adopted, then: In the vehicle-RSU cooperative sensing mode, let the size of the sensing data received by vehicle m in time slot n be... In each time slot, the RSU needs to sense the size of the sensor data that vehicle m needs to process. The vehicle transmits sensor data to the RSU's communication computing resources for processing, with an uplink transmission latency of [missing information]. The data uploaded to RSU's computing resources was processed as follows: In vehicle-RSU mode, the processing delay for each vehicle m in time slot n is... ; In terms of energy consumption, the energy consumption of vehicle m in each time slot n when using vehicle-RSU cooperative processing consists of two parts: one part is the communication consumption of vehicle m transmitting perception information to the RSU. One part is the computational cost of the RSU processing the raw sensing data sent by the vehicle. : There is energy consumption in the vehicle-RSU collaborative processing. : 。 7. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 6, characterized in that: The specific steps for calculating the total latency and total energy consumption of the vehicle in each time slot are as follows: Assume that vehicle m can select two sensing processing modes in each time slot n, using... This indicates the mode selection method for perception. This indicates that RSU sensing mode is being used. This indicates the processing latency incurred when using the vehicle-RSU cooperative perception mode. Represented as: Then vehicle m consumes energy in each time slot n. : Obtain the total vehicle delay in each time slot. : The total energy consumption result is obtained. ： 。 8. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 7, characterized in that: The specific steps to optimize the total latency and total energy consumption to minimize the latency and energy consumption are as follows: Suppose the vehicle-infrastructure cooperation system has a total of M vehicles; introduce... and As the maximum latency and energy consumption, the optimized latency and energy consumption are normalized, and at the same time, the following is introduced: As a weighting factor balancing latency and energy consumption, a third optimization function is constructed based on the total latency and total energy consumption results: P3: Among the constraints This indicates the restrictions on the values ​​of the resource allocation factor. This means that the energy consumption of each time slot cannot exceed the maximum value; This represents the processing latency constraint for each time slot, ensuring that the processing latency meets service requirements. This indicates the binary nature of the perception mode selection: either select the RSU perception mode or select the vehicle-RSU cooperative perception mode. This represents the constraints on the assignment of delay and energy consumption weight factors in the optimization problem; Based on the third optimization function, the total latency and total energy consumption results are optimized by calculating resource allocation, resulting in minimized latency and energy consumption.

9. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 8, characterized in that: The sensing conditional mutual information between the RSU and vehicle sides is optimized by allocating subcarriers and power for the communication and sensing beams of the vehicle and RSU. The specific optimization steps are as follows: Initialize a quadruple Covering the system state set Action Collection State transition probability set and real-value reward function Sets, states It describes the environment, the specific context or state in which the agent program is performing a certain action; action It refers to the decisions or actions taken by the agent in a given state; the set of state transition probabilities. It is typically used to describe the probability distribution of an agent transitioning from one state to another after performing a specific action; reward It is the core feedback signal in reinforcement learning, used to evaluate the effectiveness of an agent taking a specific action in a specific state; the goal of the MDP process is to find a policy π that maximizes the long-term cumulative reward, expressed by the formula: ,in It is a discount factor used to balance the importance of immediate rewards and future rewards; The optimization process of the perception conditional mutual information between the RSU side and the vehicle side is set as a Markov decision process using a low-level deep learning network DDPG. It is assumed that the perception conditional mutual information on the vehicle side is affected by the power allocation of communication and perception on the vehicle side and the subcarrier allocation of communication and perception on the vehicle side; the perception conditional mutual information on the RSU side is affected by the power allocation of communication and perception on the RSU side and the subcarrier allocation of communication and perception on the vehicle side. The DDPG algorithm is used to instantiate the perception conditional mutual information on the vehicle side and the RSU side respectively, and the DDPG framework on the vehicle side and the DDPG framework on the RSU side are updated simultaneously. In RSU-perceptual conditional mutual information DDPG reinforcement learning, the task is defined as maximizing the perceptual conditional mutual information on the RSU side; the state S in time slot n is defined as: These include the power allocated to each vehicle by the RSU, the communication allocated power, sensing allocated power, communication allocated subcarriers, sensing allocated subcarriers, and the downlink communication rate sent by the RSU to the vehicle. The definition of action space This includes the communication-sensing power allocation factor and its subcarrier allocation factor for each RSU (Roadside Unit). The power allocation factor determines the power allocation ratio between communication and sensing on the RSU side, while the subcarrier allocation factor determines the subcarrier allocation ratio between the communication beam and sensing beam on the RSU side. Both the power allocation factor and the subcarrier allocation factor are continuous values, ranging from [value range missing]. Between; in the RSU-side optimization problem, the reward is defined as... The conditional mutual information of each time slot RSU side is used as the reward value of the reward function, and the maximum conditional mutual information is obtained according to different power and subcarrier allocation conditions. In vehicle-side perception conditional mutual information (DDPG) reinforcement learning, the task is defined as maximizing the conditional mutual information on the vehicle side; the state S in time slot n is defined as: These include the vehicle's communication allocation power, sensing allocation power, communication allocation subcarrier, sensing allocation subcarrier, and vehicle uplink communication rate, respectively. The definition of action space This includes the vehicle's power allocation factor and subcarrier allocation factor. The power allocation factor determines the power allocation ratio between communication and sensing on the vehicle side, while the subcarrier allocation factor determines the subcarrier allocation ratio between the communication beam and sensing beam on the vehicle side. Both the power allocation factor and the subcarrier allocation factor are continuous values, with a range of values ​​within... Between; defining the reward The vehicle-side conditional mutual information of each time slot is used as the reward value of the reward function, and the maximum conditional mutual information is obtained according to different power and subcarrier allocation conditions.

10. The integrated transmission optimization method for vehicle networking based on sensor quality according to claim 9, characterized in that: The total latency and total energy consumption results are optimized by calculating resource allocation. The specific steps are as follows: A high-level deep learning network is used to set the optimization process of total latency and total energy consumption as a Markov decision process; the state S is defined as follows in time slot n: The state includes the size of the sensing data, the selection strategy for the sensing mode, the computational resources allocated by the RSU to the two modes respectively, the remaining energy of the RSU, and the remaining bandwidth for processing; action space. This includes a computational resource allocation factor, which determines the computational resources allocated to RSU perception and vehicle-RSU cooperative perception. This factor is a continuous number and its value ranges from [value range missing]. In the reward design of high-level reinforcement learning networks, the optimization problem is to minimize latency and energy consumption; the reward design is the inverse of the optimization problem. The reward value is designed based on the optimized allocation of computing resources according to different modes.

Images