Resource allocation methods in multimodal semantics and bit coexistence communication

Abstract

This invention relates to the field of wireless communication technology in vehicle-to-everything (V2X) communication, and discloses a resource allocation method in multimodal semantic and bit-coexistence communication. This method first treats CAVs (Conducting Vehicles) on the road as semantic communication users and HDVs (Hardware-Defined Vehicles) as traditional bit users. The base station maximizes the semantic rate of CAVs and the bit rate of HDVs by optimizing beamforming, power allocation, and the number of semantic symbols. Simultaneously, by introducing a lane change prediction model to predict the lane change probability of each vehicle, more communication resources are allocated to lane-changing vehicles in time slots with high lane change probability, thereby improving the semantic rate of CAVs or the bit rate of HDVs. This invention can effectively allocate resources, thereby improving the communication performance of lane-changing vehicles in time slots with high lane change probability while ensuring the communication performance of other vehicles.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology in the Internet of Vehicles (IoV), and specifically to a resource allocation method for multimodal semantic and bit coexistence communication in hybrid intelligent transportation systems. Background Technology

[0002] Semantic communication is a key enabling technology for future sixth-generation (6G) networks. It also improves the efficiency and intelligence of vehicle-to-everything (V2X) communication. However, connected autonomous vehicles (CAVs) and human-driven vehicles (HDVs) are expected to coexist for a considerable period: CAVs, due to their significantly higher intelligence, can be equipped with semantic transceivers composed of neural networks to process and transmit semantic information in real time, while HDVs will continue to use traditional transceivers. It is important to note that a more suitable resource allocation scheme needs to be developed for both CAVs and HDVs, addressing their different requirements. For example, for CAVs, the focus should be on the quality of task completion, while for HDVs, the focus should be on the accuracy and speed of information transmission. Furthermore, vehicles require more communication resources when changing lanes. Therefore, allocating more resources to lane-changing vehicles in a timely manner to meet their communication needs during lane changes is also crucial.

[0003] Base stations can acquire additional information from other vehicles and transmit it to the target vehicle, thereby assisting in inter-vehicle cooperation and providing more references for vehicle decision-making during driving. However, CAVs and HDVs have different requirements, and the urgent needs of lane-changing vehicles must also be considered. This places higher demands on timely resource allocation.

[0004] However, there is currently no effective resource allocation method for multimodal semantics and bit coexistence communication in hybrid intelligent transportation systems. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides an effective method for resource allocation in hybrid intelligent transportation systems where semantic and bit-based communication coexist.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows:

[0007] A resource allocation method for multimodal semantics and bit coexistence communication in a hybrid intelligent transportation system includes the following steps:

[0008] 1) CAV and HDV enter the road and continue to move forward. The base station can obtain the location and channel information of each vehicle in real time in each time slot. Based on the location information of the target vehicle and the location information of the surrounding vehicles, the base station can predict the lane change probability of the target vehicle through a neural network model (Long Short-Term Memory Network LSTM model).

[0009] 2) CAV and HDV have different communication requirements: CAV focuses more on the quality of task completion, therefore semantic rate is introduced. This serves as the performance evaluation standard. For HDV, however, the transmission bit rate is used. As a performance evaluation standard. Furthermore, to avoid intermodal interference, in addition to channel estimation, the time slot for transmitting data will be divided into three sub-time slots, where the first sub-time slot is used to transmit bit information and point cloud semantic information, the second sub-time slot is used to transmit bit information and image semantic information, and the third sub-time slot is used to transmit bit information.

[0010] 3) As the vehicles continue to move, the base station will generate beams for each vehicle in real time to enhance the signal strength in the direction of the vehicle, and will also allocate power to each vehicle. In addition, for CAV, the semantic rate also depends on the number of semantic symbols used for transmission. Therefore, the base station will also optimize the number of semantic symbols for each CAV. Furthermore, the beamforming matrix and power allocation will also be optimized for sub-time slots.

[0011] 4) Based on semantic rate and bit rate, the base station optimizes the beamforming matrix, power allocation and the number of semantic symbols for each CAV in each time slot; at the same time, in order to give more communication resources to lane-changing vehicles, the prediction results of the lane-changing prediction model are also incorporated into the optimization objective.

[0012] Furthermore, in each time slot In the modeling of the base station optimization problem (1)-(7), the following is a summary:

[0013] ;

[0014] The objective function in optimization problem (1) means maximizing the weighted sum of the average HDV bit rate and CAV semantic rate by optimizing the beamforming matrix, power allocation, and semantic symbol number selection for each time slot. Among them, optimization problems (2) and (3) are the amplitude constraints of the beamforming vector, optimization problem (4) means that the transmit power does not exceed the maximum power limit, optimization problems (5) and (6) mean that the bit rate and semantic rate of HDV and CAV are not lower than the minimum limit, and optimization problem (7) means that the sum of CAV and semantic symbol number does not exceed the maximum number of transmitted symbols in one time slot.

[0015] in, Indicates the weighting parameter; and The weights of HDV and CAV for each time slot with respect to the lane change probability are represented by the following function:

[0016] ;

[0017] and This indicates the channel change probability for HDV and CAV in the current time slot; and These are the weight parameters of the function; and This indicates the bit rate and semantic rate for each time slot HDV and CAV; This represents the beamforming vector of the i-th sub-time slot for the b-th HDV; This represents the beamforming vector for the t-th CAV when transmitting point cloud semantic information; This represents the beamforming vector for the t-th CAV when transmitting semantic information of an image; This represents the power allocation of the i-th sub-time slot to the b-th HDV; This represents the power allocation for the t-th CAV when transmitting point cloud semantic information; This represents the power allocation for the t-th CAV when transmitting image semantic information; This represents the number of semantic symbols in the image and point cloud assigned to the t-th CAV; This indicates the maximum power limit; L indicates the maximum number of symbols transmitted per time slot.

[0018] Furthermore, the bit rate of the nth time slot in optimization problem (1) It can be represented as

[0019] ;

[0020] in, . This represents the signal dryness ratio from the i-th sub-timeslot base station to the b-th HDV signal.

[0021] ;

[0022] in, This represents the channel from the base station to the b-th HDV; This represents the variance of Gaussian white noise. , , These represent the power transmitted from the 1st, 2nd, and 3rd sub-time slot base stations to the j-th HDV, which are also the interference power of the j-th HDV to the b-th HDV. , , respectively, represent the transmission power of the base station when transmitting point cloud semantic information and image semantic information in the sub-time slots (the first and second sub-time slots), and are also the interference power of the i-th CAV to the b-th HDV. , , These represent the beamforming vectors designed for the j-th HDV by the 1st, 2nd, and 3rd sub-time slot base stations, respectively. , These represent the beamforming vectors designed by the base station for the i-th CAV in the sub-time slots (the first and second sub-time slots) for transmitting point cloud semantic information and image semantic information, respectively.

[0023] Furthermore, the semantic rate of the nth time slot in optimization problem (1) It can be represented as

[0024] ;

[0025] in, The weights represent the semantics of the image and the semantics of the point cloud. and This represents the semantic rate of the image and the semantic rate of the point cloud.

[0026] Point cloud semantic rate is a quality metric for point cloud reconstruction: chamfer distance (CD), while image semantic rate is a quality metric for image reconstruction: structural similarity (SSIM). CAV (Computer-Aided Visualization) can receive semantic information and reconstruct the original information; the reconstruction quality is judged by these two metrics. However, due to the black-box nature of neural networks, the semantic rates of both can only be obtained through curve fitting; the expression for the fitted curve is:

[0027] ;

[0028] in Represents the fitting parameters, which are derived from... Decide, Indicates the signal-to-noise ratio. The number of semantic symbols representing an image or point cloud; that is, the semantic rate is a number of symbols related to the number of semantic symbols. and The function.

[0029] Furthermore, the specific process of step 4) is as follows:

[0030] Step 4.1) At each time slot base station, the lane change prediction model is used to predict the lane change probability of all vehicles in the next time slot.

[0031] Step 4.2) Model the vehicle network using a Graph Neural Network (GNN). In this model, vehicles are treated as nodes in a graph, and the features of each node are the channel from the base station to each vehicle and the vehicle's current lane change probability. The model then aggregates the features of each node through a graph aggregation mechanism to form a new graph representation. After L layers of aggregation, the GNN outputs the aggregated feature matrix.

[0032] Step 4.3) The output of the GNN is transformed into a probability distribution for each optimization variable using a Multi-Head Policy Network (MHPN). Specifically, the MHPN consists of three heads, each corresponding to beamforming, power allocation, and semantic symbol count, respectively. The beamforming head and power allocation head are further divided into three sub-heads, each representing a sub-time slot. That is, in the optimization of each time slot, each sub-time slot will adopt a different beamforming and power allocation scheme.

[0033] Step 4.4) The probability distribution of each optimized variable in each sub-slot of the MHPN output is then obtained. The Proximal Policy Gradient (PPO) algorithm is then used to sample from the probability distribution to obtain the specific optimized variable values ​​for each slot. After obtaining the specific optimized variable values, the network parameters of PPO, GNN, and MHPN are updated according to the reward function. In summary, GNN is used for feature extraction, MHPN for feature mapping, and PPO for specific variable optimization. The three networks jointly optimize to maximize the weighted semantic rate and bit rate.

[0034] The design concept of this invention is as follows:

[0035] First, CAVs (Consumer Abilities) in the road are considered semantic communication users, and HDVs (Hardware-Defined Vehicles) are considered traditional bit users. The base station maximizes the semantic rate of CAVs and the bit rate of HDVs by optimizing beamforming, power allocation, and the number of semantic symbols. At the same time, a lane change prediction model is introduced to predict the lane change probability of each vehicle, and more communication resources are allocated to lane-changing vehicles in time slots with high lane change probability, thereby improving the semantic rate of CAVs or the bit rate of HDVs. This invention can effectively allocate resources, thereby improving the communication performance of lane-changing vehicles in time slots with high lane change probability while ensuring the communication performance of other vehicles.

[0036] The beneficial effects of this invention are as follows:

[0037] This invention fills the gap in existing technologies regarding resource allocation methods for multimodal semantic and bit coexistence communication in hybrid intelligent transportation systems. It provides a solution that considers beamforming, power allocation, and the number of semantic symbols. This method allocates resources through joint optimization of GNN, MHPN, and PPO, thereby improving the weighted bit rate and semantic rate. Attached Figure Description

[0038] Figure 1 It is a scene overview image;

[0039] Figure 2 This is a schematic diagram of the sub-timeslot allocation scheme and timeslot transmission;

[0040] Figure 3 It is the semantic rate of a CAV changing lanes in the lane change time slot in the road;

[0041] Figure 4 It refers to the change in bit rate of each HDV vehicle on the road as the base station's transmission power changes. Detailed Implementation

[0042] The present invention will now be further described with reference to the accompanying drawings.

[0043] like Figure 1 As shown, in a hybrid intelligent transportation system, a base station is deployed on the roadside, which includes both semantic transceivers and traditional transceivers. Multiple vehicles exist on the road served by this base station. and , and This represents a collection of HDVs and CAVs. CAVs or HDVs on the road can change lanes in real time. Simultaneously, the base station uses a neural network model to predict the probability of the target vehicle changing lanes at the next moment based on the target vehicle's real-time location information and the location information of surrounding vehicles. CAVs can transmit and receive semantic information, including point cloud semantic information and image semantic information, while HDVs can transmit and receive traditional bit signals. Other CAVs or HDVs can transmit information to the target vehicle through the base station.

[0044] like Figure 2 As shown, a resource allocation method for multimodal semantic and bit coexistence communication in a hybrid intelligent transportation system is implemented based on an existing wireless information transmission system. In this system, the base station acts as the information sender, and CAV and HDV act as receivers. Data transmission employs sub-time slot transmission, where the first sub-time slot is used to transmit bit information and point cloud semantic information, the second sub-time slot is used to transmit bit information and image semantic information, and the third sub-time slot is used to transmit bit information.

[0045] In this embodiment, the bit rate of the nth time slot It can be represented as

[0046] ;

[0047] in, . This represents the signal dryness ratio from the i-th sub-timeslot base station to the b-th HDV signal.

[0048] ;

[0049] in This represents the channel from the base station to the b-th HDV. This represents the variance of Gaussian white noise.

[0050] Furthermore, the semantic rate of the nth time slot in optimization problem (1) It can be represented as

[0051] ;

[0052] in, The weights represent the semantics of the image and the semantics of the point cloud. and This represents the semantic rate of the image and the semantic rate of the point cloud.

[0053] Point cloud semantic rate and image semantic rate are respectively the point cloud reconstruction quality metric: chamfer distance (CD), and the image reconstruction quality metric: structural similarity (SSIM). CAV can receive semantic information and reconstruct the original information; the reconstruction quality is judged by these two metrics. However, due to the black-box nature of neural networks, the semantic rates of both can only be obtained through curve fitting. The expression for the fitted curve is:

[0054] ;

[0055] in This represents the fitted parameters.

[0056] In this embodiment, the optimization of beamforming, power allocation, and semantic symbol number selection for each time slot n can be expressed as:

[0057] ;

[0058] This embodiment presents a resource allocation method for multimodal semantic and bit-coexisting communication in a hybrid intelligent transportation system. It addresses the shortcomings of existing methods that neglect joint semantic and bit optimization, as well as the communication resource requirements of lane-changing vehicles. By jointly optimizing the beamforming matrix, power allocation, and semantic symbol number selection, it improves the semantic rate of CAV and the bit rate of HDV.

[0059] In this embodiment, CAV and HDV move forward continuously in the road, and the base station is located on the roadside in the middle of the road. It can generate beamforming matrix and allocate power in real time according to the position of each vehicle, and select the number of semantic symbols for CAV.

[0060] like Figure 3 As shown in the figure, the resource allocation effect of the method of the present invention in the case of CAV lane change is shown. It can be seen from the figure that during the period when the probability of lane change is high, the method of the present invention effectively identifies the lane change behavior and assigns it a higher resource priority weight, thus significantly improving its semantic rate.

[0061] like Figure 4As shown in the figure, the bit rate of each HDV on the road changes with the change of base station transmit power. It can be seen from the figure that the method proposed in this invention exhibits a higher bit rate than other comparative schemes when the power increases. This demonstrates that the scheme proposed in this invention can not only adapt to the additional communication needs of lane-changing vehicles in a timely manner, but also ensure superior average system performance.

[0062] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered as limited to the specific forms stated in these embodiments. The scope of protection of this invention is also based on equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A resource allocation method in multimodal semantic and bit coexistence communication, based on a hybrid intelligent transportation system, wherein the system includes a base station, which acts as both a sender and receiver of semantic and bit information, and within the service range of the base station are T networked autonomous vehicles (CAVs) and B human-driven vehicles (HDVs); characterized in that, Includes the following steps: Step 1) The base station in each time slot The system acquires the location and channel information of each vehicle in real time. Based on the location information of the target vehicle and the location information of the surrounding vehicles, it predicts the lane change probability of the target vehicle through a neural network model. Step 2) Based on the different communication requirements of CAV and HDV, using semantic rate As a standard for evaluating CAV performance, bit rate As a standard for evaluating HDV performance; Step 3) The base station generates beams for each vehicle in real time, allocates power to each vehicle, and determines the number of semantic symbols for CAV; Step 4) The base station in each time slot The joint optimization of beamforming matrix, power allocation, and number of semantic symbols per CAV; at the same time, the lane change probability prediction results are incorporated into the optimization objective to allocate more communication resources to lane-changing vehicles; In each time slot In the modeling of the base station optimization problem (1)-(7), the following steps are taken: ； in, Indicates the weighting parameter; and This represents the weight of HDV and CAV for each time slot with respect to the probability of lane change. and This indicates the channel change probability of HDV and CAV in the current time slot; and These are the weight parameters of the function; and This indicates the bit rate and semantic rate for each time slot HDV and CAV; This represents the beamforming vector of the i-th sub-time slot for the b-th HDV; This represents the beamforming vector for the t-th CAV when transmitting point cloud semantic information; This represents the beamforming vector for the t-th CAV when transmitting semantic information of an image; This represents the power allocation of the i-th sub-time slot to the b-th HDV; This represents the power allocation for the t-th CAV when transmitting point cloud semantic information; This represents the power allocation for the t-th CAV when transmitting image semantic information; This represents the number of semantic symbols in the image and point cloud assigned to the t-th CAV; This indicates the maximum power limit; L indicates the maximum number of symbols transmitted per time slot.

2. The resource allocation method in multimodal semantic and bit coexistence communication as described in claim 1, characterized in that: Optimize the bit rate of the nth time slot in problem (1) Represented as: ； in, , This represents the signal-to-dryness ratio from the i-th sub-timeslot base station to the b-th HDV signal; where: ； in, This represents the channel from the base station to the b-th HDV. This represents the variance of Gaussian white noise; , , These represent the power transmitted from the 1st, 2nd, and 3rd sub-time slot base stations to the j-th HDV, respectively. , , representing the transmission power of the sub-timeslot base station transmitting point cloud semantic information and image semantic information to the i-th CAV signal, respectively. , , These represent the beamforming vectors designed for the j-th HDV by the 1st, 2nd, and 3rd sub-time slot base stations, respectively. , These represent the beamforming vectors designed for the i-th CAV by the sub-time slot base station transmitting point cloud semantic information and image semantic information, respectively. Furthermore, the semantic rate of the nth time slot in optimization problem (1) Represented as: ； in, The weights representing the semantics of the image and the point cloud are... and These represent the image semantic rate and the point cloud semantic rate, respectively. This represents the signal-to-dryness ratio at the t-th CAV during the transmission of image semantic data; This represents the signal-to-dryness ratio at the t-th CAV when transmitting point cloud semantic data.

3. The resource allocation method in multimodal semantics and bit coexistence communication as described in claim 1, characterized in that: The specific process of step 4) is as follows: Step 4.1) Predict the lane change probability of all vehicles in the next time slot in each time slot; Step 4.2) Model the vehicle network using a graph neural network (GNN): Treat vehicles as nodes in a graph. The features of each node are the channel from the base station to each vehicle and the lane change probability at the current time. Then, the features of each node are aggregated through a graph aggregation mechanism to form a new graph representation. After L layers of aggregation, the GNN outputs the aggregated feature matrix. Step 4.3) The output of the GNN is transformed into a probability distribution for each optimization variable using the Multi-Head Policy Network (MHPN). The MHPN is divided into three heads, each corresponding to beamforming, power allocation, and semantic symbol number, respectively. The beamforming head and power allocation head are further divided into three sub-heads, which represent three sub-time slots. The first sub-time slot is used to transmit bit information and point cloud semantic information, the second sub-time slot is used to transmit bit information and image semantic information, and the third sub-time slot is used to transmit bit information. Step 4.4) MHPN outputs the probability distribution of each optimization variable in each sub-slot, and then uses the near-end policy gradient algorithm (PPO) to sample from the probability distribution to obtain the specific optimization variable value for each slot; after obtaining the specific optimization variable value, the network parameters of PPO, GNN and MHPN are updated according to the reward function.

Images