High energy efficiency three-dimensional mobile terminal anti-interference communication method with size model cooperation

Abstract

The application provides a high-energy-efficiency three-dimensional mobile terminal anti-interference communication method based on a size model, and relates to the fields of wireless communication and artificial intelligence. As a transmitter, the mobile terminal adopts reinforcement learning combined with time delay and energy consumption information to select a transmission power, a channel, a trajectory and a feedback strategy, so that interference attack can be quickly defended. A large model deployed in a controller infers interference characteristics such as an interference distance, an interference machine speed and an interference type, and environmental characteristics such as an obstacle layout and a line-of-sight change condition based on image and spectrum information of the mobile terminal and the controller, and feeds back neural network weights from historical mobile terminals. The method can assist in quickly defending various types of interference attacks, avoiding obstacles and reducing time delay and terminal energy consumption, and supports applications such as environmental monitoring, embodied intelligence and industrial intelligent inspection.

Description

Technical Field

[0001] This invention relates to the fields of wireless communication and artificial intelligence, and in particular to a high-efficiency anti-interference communication method for three-dimensional mobile terminals that utilizes a combination of large and small models. Background Technology

[0002] Existing mobile terminal anti-interference communication technologies are mainly based on information such as channel state, received interference power, and bit error rate. They employ algorithms such as successive convex approximation, graph attention networks, and reinforcement learning to select strategies such as channel, transmit, and relay power to improve transmission rate and reduce energy consumption. For example, (1) based on channel state information, successive convex approximation and alternating optimization methods are used to adjust transmit power and trajectory to improve transmission rate, reduce transmit power and energy consumption, and defend against intelligent interference. (2) Based on information such as channel state and remaining battery power, transmit and relay power are selected to reduce bit error rate and interruption probability and defend against reactive interference. (3) Based on information such as received interference intensity and channel state, federated reinforcement learning algorithms are used to select the channel and power of the mobile terminal to improve transmission rate, reduce energy consumption, and defend against intelligent interference. (4) Based on channel state and received interference intensity, graph attention networks and deep deterministic strategy gradient algorithms are used to select the beamforming and trajectory of the mobile terminal, thereby improving transmission rate and defending against intelligent interference. (5) The control center deploys a large model based on information such as images from mobile terminals and other devices to infer environmental characteristics such as rainfall indicators, select data compression rate, transmission power and transmission channel to reduce latency and energy consumption.

[0003] However, the above-mentioned solutions only use large models to extract environmental features and select anti-interference strategies. When the mobile terminal communication environment and obstacle layout change rapidly, the existing intelligent anti-interference algorithms become slower in defending against various interference attacks. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in the aforementioned technologies. To this end, one objective of this invention is to propose a high-energy-efficiency three-dimensional mobile terminal anti-interference communication method using a large-scale model collaboration. The mobile terminal acts as a transmitter, employing information such as reinforcement learning speed, latency, and energy consumption to select transmission power, channel, trajectory, and feedback strategies, balancing the feedback overhead of the large model with the speed of interference defense. The large model deploying the controller infers interference and environmental characteristics based on image and spectrum information acquired by itself and the mobile terminal, and feeds back neural network weights from historical mobile terminals. This enables obstacle avoidance and reduces latency and mobile terminal energy consumption.

[0005] To achieve the above objectives, a first aspect of the present invention proposes a high-efficiency three-dimensional mobile terminal anti-interference communication method based on a large-scale model collaboration. The method includes the following steps: acquiring sensor data, communication link quality data, and environmental and interference features fed back by the large-scale model of the mobile terminal in the current time slot to construct a system state containing the mobile terminal's own state, communication link state, and environmental and interference information; inputting the system state into a neural network deployed on the mobile terminal to obtain the long-term utility value and long-term risk value of each candidate anti-interference strategy; constructing a modified Boltzmann distribution based on the long-term utility value and the long-term risk value, and selecting an anti-interference strategy that meets the safe distance requirement from the candidate anti-interference strategies based on the obstacle layout in the environmental features, wherein the anti-interference strategy includes transmit power, transmission channel, trajectory, and feedback strategy; adjusting the trajectory according to the anti-interference strategy, and then using the sensor data, mobile terminal neural network weights, and the feedback strategy to select... A predetermined transmission power and transmission channel are sent to the controller to receive feedback results returned by the controller. The feedback results include updated environmental characteristics, updated interference characteristics, and latency. Based on the latency, mobile terminal power consumption, and feedback overhead, immediate utility is calculated, and immediate risk is calculated based on whether the latency exceeds a preset threshold. The system state, the anti-interference strategy, the immediate utility, the immediate risk, and the system state of the next time slot are constructed as experience and stored in an experience pool. Small batches of experience are sampled from the experience pool to update the neural network weight parameters using stochastic gradient descent and backpropagation algorithms. When the feedback strategy requests neural network weights, the neural network weights of historical mobile terminals are fused based on task similarity until the neural network converges, resulting in an anti-interference communication strategy decision model. Thus, the mobile terminal deploys a small reinforcement learning model, which, based on speed and the environmental and interference characteristics extracted by the large model, quickly defends against various interference attacks and avoids obstacles.

[0006] In addition, the energy-efficient three-dimensional mobile terminal anti-interference communication method with large and small model collaboration proposed in the above embodiments of the present invention may also have the following additional technical features: Optionally, before acquiring the sensor data, communication link quality data, and environmental and interference characteristics of the mobile terminal in the current time slot, the process includes: dividing the communication process into continuous time slots, and obtaining a set of candidate anti-interference strategies based on the quantization order of the transmit power, the number of available channels, the dimension of the feedback strategy, and the order of the trajectory; constructing the neural network in the mobile terminal, and initializing the neural network discount rate, delay coefficient, energy consumption coefficient, feedback overhead coefficient, risk threshold, environmental characteristics, interference characteristics, network weights, delay, and experience pool.

[0007] Optionally, the neural network includes a Q network and an R network. The Q network is used to estimate the long-term utility of each candidate anti-interference strategy, and the R network is used to estimate the long-term risk of each candidate anti-interference strategy. The Q network and the R network have the same input layer dimension, and the number of neurons in their output layers is the total number of candidate anti-interference strategies.

[0008] Optionally, the sensor data includes images and their data volume acquired by the mobile terminal from the sensor, its remaining battery power, and generated large model prompt words; the communication link quality data includes the channel gain from the mobile terminal to the controller, the controller's estimated delay, the interfered channel, and the received interference intensity.

[0009] Optionally, the environmental features are inferred by the large model deployed on the controller based on images and spectrum information sent by itself and mobile devices, including line-of-sight changes and obstacle layout; the interference features are inferred by the large model, including jammer speed, interference distance, and interference type.

[0010] Optionally, constructing a modified Boltzmann distribution based on the long-term utility value and the long-term risk value includes: determining the selection probability of each candidate anti-interference strategy based on a comprehensive trade-off between the long-term utility value and the long-term risk value of each candidate anti-interference strategy.

[0011] Optionally, the feedback strategy includes a first feedback level and a second feedback level. The first feedback level is used to request the controller to return the environmental features, the interference features, and the latency. The second feedback level is used to additionally request the neural network weights of the historical mobile terminal based on the first feedback level.

[0012] Optionally, the instantaneous utility is obtained by weighting the latency, the mobile terminal power consumption, and the feedback overhead.

[0013] Optionally, the instantaneous risk is obtained by multiplying the amount by which the delay exceeds a preset delay threshold by the corresponding risk coefficient and then summing them up.

[0014] Optionally, fusing the neural network weights of historical mobile terminals based on task similarity includes: calculating the task similarity between the current mobile terminal and historical mobile terminals, determining the fusion coefficient of each external network weight based on the task similarity, and weighting and fusing the external network weights with the current mobile terminal network weights according to the fusion coefficient. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a high-efficiency three-dimensional mobile terminal anti-interference communication method based on a large-scale model collaboration according to an embodiment of the present invention. Figure 2A time delay diagram illustrating the high-efficiency anti-interference communication method for a three-dimensional mobile terminal based on a large-scale model collaboration according to an embodiment of the present invention; Figure 3 A schematic diagram illustrating the energy consumption of a high-energy-efficiency anti-interference communication method for a three-dimensional mobile terminal based on a large-scale model collaboration according to an embodiment of the present invention. Figure 4 This is a schematic diagram illustrating the efficiency of a high-efficiency three-dimensional mobile terminal anti-interference communication method based on a large-scale model collaboration according to an embodiment of the present invention. Detailed Implementation

[0016] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0017] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the invention to those skilled in the art.

[0018] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0019] refer to Figure 1 As shown, the energy-efficient 3D mobile terminal anti-interference communication method of large and small model collaboration according to an embodiment of the present invention includes the following steps: S101, acquire sensor data, communication link quality data and environmental and interference characteristics of the mobile terminal in the current time slot, and the feedback of the large model, in order to construct a system state that includes the mobile terminal's own state, communication link state and environmental and interference information.

[0020] As an example, sensor data includes images and their data volume acquired by the mobile terminal from the sensor, its remaining battery power, and generated large model prompt words; communication link quality data includes channel gain from the mobile terminal to the controller, controller estimated delay, interfered channels, and received interference intensity.

[0021] As an example, environmental features are inferred by a large model deployed on the controller based on images and spectrum information sent by itself and mobile devices, including changes in line-of-sight, center coordinates of surrounding obstacles, and obstacle layout such as maximum length, width, and height; interference features are inferred by the large model, including jammer speed, interference distance, and interference type.

[0022] Specifically, in time slots The mobile terminal acquires images from the sensor. Information such as data volume is used to estimate the amount of data. Get your remaining battery level Generate prompt words Estimate the channel gain from the mobile terminal to the controller. Obtain the delay estimated by the controller. Interferenced channel and received interference strength And the feedback from the large model, including changes in view distance. and obstacle layout Environmental characteristics jammer speed Interference distance and interference type Interference characteristics Thus, the state of the mobile terminal construction is obtained. :

[0023] As an example, before obtaining the environmental and interference characteristics fed back by the large model of sensor data and communication link quality data of the mobile terminal in the current time slot, the process includes: dividing the communication process into continuous time slots, and obtaining a set of candidate anti-interference strategies based on the quantization order of the transmit power, the number of available channels, the dimension of the feedback strategy, and the order of the trajectory; constructing a neural network in the mobile terminal, and initializing the neural network discount rate, latency coefficient, energy consumption coefficient, feedback overhead coefficient, risk threshold, environmental characteristics, interference characteristics, network weight, latency, and experience pool.

[0024] Specifically, the communication process between the mobile terminal and the controller is divided into continuous time slots, with each time slot having a length of T. Let the transmit power be... Quantified as Rank, among which For maximum transmit power, transmission channel bandwidth is The feedback strategy is... For example, 1 indicates a request to the controller for information such as latency and feedback from the large model regarding the distance and speed of the interfering machine, obstacle layout, and line-of-sight changes; 2 indicates an additional request to the large model for information such as neural network weights. (Mobile terminal trajectory) Including pitch angle Azimuth and speed Therefore, the optional anti-interference strategy is: .

[0025] Mobile terminal construction Network and The network consists of an input layer with M neurons, and two hidden layers, each containing M neurons. and The output layer contains neurons, and the output layer contains Each neuron corresponds to a neuron. First-order transmit power, One transmission channel, Dimensional feedback strategies and Candidate joint actions formed by the trajectories of the order. Network weight parameters are denoted as This is used to estimate the long-term utility of candidate anti-interference strategies. Network weight parameters are denoted as This is used to estimate the long-term risk of candidate anti-interference strategies. The neural network is initialized. Discount rate Neural Networks Discount rate Delay coefficient Energy consumption coefficient Feedback cost coefficient Risk threshold Initialize environment characteristics Interference characteristics Network weight and Delay Experience Pool Empty set, small batch size .

[0026] S102, the system state is input into the neural network deployed on the mobile terminal to obtain the long-term utility value and long-term risk value of each candidate anti-interference strategy.

[0027] As an example, the neural network includes a Q network and an R network. The Q network is used to estimate the long-term utility of each candidate anti-interference strategy, and the R network is used to estimate the long-term risk of each candidate anti-interference strategy. The input layer dimensions of the Q network and the R network are the same, and the number of neurons in the output layer is the total number of candidate anti-interference strategies.

[0028] S103. A modified Boltzmann distribution is constructed based on long-term utility value and long-term risk value. Based on the obstacle layout in the environmental characteristics, an anti-interference strategy that meets the safety constraints is selected from the candidate anti-interference strategies. The optimal anti-interference strategy includes transmission power, transmission channel, trajectory and feedback strategy.

[0029] As an example, a modified Boltzmann distribution is constructed based on long-term utility value and long-term risk value, including: determining the selection probability of each candidate anti-interference strategy based on a comprehensive trade-off between the long-term utility value and the long-term risk value of each candidate anti-interference strategy.

[0030] In other words, each candidate anti-interference strategy is assigned a different selection probability according to its utility value and risk value. After eliminating all candidate anti-interference strategies that violate security constraints, the anti-interference strategy is sampled and selected according to the probability distribution.

[0031] Specifically, the mobile terminal status Enter them separately Network and Network, output anti-interference strategy Calculating long-term utility value and risk value A modified Boltzmann distribution is constructed based on the long-term utility and risk values ​​of each candidate anti-interference strategy. Trajectory strategies are then selected based on the obstacle layout generated by the large model to ensure the mobile terminal maintains a safe distance from obstacles, while avoiding collisions between trajectories and obstacles between adjacent waypoints. The anti-interference strategy is then output. .

[0032] S104 After adjusting the trajectory according to the optimal anti-interference strategy, the sensor data, mobile terminal neural network weights and feedback strategy are sent to the controller through the selected transmission power and transmission channel to receive the feedback results returned by the controller. The feedback results include the updated environmental characteristics, the updated interference characteristics and the time delay.

[0033] As an example, the feedback strategy includes a first feedback level and a second feedback level. The first feedback level is used to request the controller to return environmental characteristics, interference characteristics, and latency. The second feedback level is used to additionally request the neural network weights of the historical mobile terminal based on the first feedback level.

[0034] Specifically, the mobile terminal adjusts its pitch, azimuth, and speed to reach the next position, using the selected transmission power. and transmission channel Image and its own neural network weights Information and feedback strategies The data is sent to the controller; the mobile terminal receives the feedback results generated by the controller via the control link. The feedback results include environmental characteristics. Interference characteristics and latency When feedback strategy At the same time, it also includes neural network weights from historical mobile terminals. .

[0035] S105 calculates the immediate utility based on latency, mobile terminal power consumption, and feedback overhead, and calculates the immediate risk based on whether the latency exceeds a preset threshold, so as to construct the system state, optimal anti-interference strategy, immediate utility, immediate risk, and the system state of the next time slot into an experience pool.

[0036] As an example, the instant utility is obtained by multiplying latency, mobile terminal power consumption, and feedback overhead as negative benefit terms by their respective weighting coefficients.

[0037] Specifically, the mobile terminal depends on the latency. Energy consumption and feedback overhead Calculate immediate utility : .

[0038] As an example, the immediate risk is determined by the amount by which the latency exceeds a preset latency threshold.

[0039] Specifically, calculate immediate risk. : .

[0040] S106. Sample small batches of experience from the experience pool to update the neural network weight parameters through stochastic gradient descent and backpropagation algorithms. When the feedback strategy is to request neural network weights, fuse the neural network weights of historical mobile terminals based on task similarity until the neural network converges to obtain the anti-interference communication strategy decision model.

[0041] As one embodiment, fusing the neural network weights of historical mobile terminals based on task similarity includes: calculating the task similarity between the current mobile terminal and historical mobile terminals, determining the fusion coefficient of each external network weight based on the task similarity, and weighting and fusing the external network weights with the current mobile terminal network weights according to the fusion coefficients.

[0042] Specifically, generating experience And store it in the experience pool : .

[0043] When the amount of experience in the experience pool is not less than the small batch size At that time, from the experience pool Random selection These experiences constitute a small batch set. Define the discount factor. and The weight parameters are updated using stochastic gradient descent and backpropagation algorithms. and ,make The network output approximates the long-term cumulative utility, making The network output approximates the long-term accumulated risk, as shown in the following formula:

[0044]

[0045] If the feedback strategy is At that time, based on task similarity Update network parameters:

[0046] Repeat the above steps until the following conditions are met: and The algorithm converges once the preset number of training time slots is reached.

[0047] Therefore, this application presents a high-energy-efficiency 3D mobile terminal anti-jamming communication method using a large-scale model-driven approach. The mobile terminal, acting as a transmitter, deploys reinforcement learning to select transmission power, channel, trajectory, and feedback strategies based on information such as latency and energy consumption, balancing communication performance, mobile terminal energy consumption, and feedback overhead. The controller deploys a Qwen3.5-27B large-scale model, generating environmental features such as obstacle layout and interference features such as jammer speed based on images and spectral features from itself and the mobile terminal, and feeding back neural network weights from historical mobile terminals. The mobile terminal adjusts its anti-jamming strategy distribution according to the obstacle layout to avoid collisions and quickly defend against jamming attacks.

[0048] As a specific embodiment, to better understand the above technical solution, the high-efficiency anti-interference communication method for 3D mobile terminals with coordinated large and small models includes the following steps: Step 1: Divide the communication process between the mobile terminal and the controller into continuous time slots, with each time slot having a length of 1 second. Record the transmit power. The quantization is of order 5, where, 100mW, transmission channel GHz, bandwidth MHz. The feedback strategy is... Here, 1 represents requesting information from the controller, including latency and feedback from the large model such as the distance and speed of the interfering machine, obstacle layout, and line-of-sight changes; 2 represents additionally requesting information from the large model, such as neural network weights from historical mobile terminals. Mobile terminal trajectory. Including pitch angle Azimuth and speed m / s. Therefore, the possible anti-interference strategy is: .

[0049] Step 2: Mobile Terminal Construction Network and The network consists of an input layer with 17 neurons, a second and third hidden layer with 64 and 128 neurons respectively, and an output layer with 2400 neurons, each corresponding to a candidate joint action composed of 5-order transmit power, 5 transmission channels, a 2-dimensional feedback strategy, and a 48-order trajectory. Network weight parameters are denoted as This is used to estimate the long-term utility of candidate anti-interference strategies. Network weight parameters are denoted as This is used to estimate the long-term risk of candidate anti-interference strategies. The neural network is initialized. Discount rate Neural Networks Discount rate Delay coefficient Energy consumption coefficient Feedback cost coefficient Risk threshold Initialize environment characteristics Interference characteristics Network weight and Delay Experience Pool Empty set, small batch size .

[0050] Step 3: In the time slot The mobile terminal acquires images from the sensor. Estimate its data volume Get your remaining battery level Generate 200B prompt words Estimate the channel gain from the mobile terminal to the controller. Obtain the delay estimated by the controller. Interferenced channel and received interference strength And the feedback from the large model, including changes in view distance. and obstacle layout Environmental characteristics jammer speed Interference distance and interference type Interference characteristics .

[0051] Step 4: Based on the mobile terminal construction status obtained in Step 3 :

[0052] Step 5: Set the status Enter them separately Network and Network, output anti-interference strategy Calculating long-term utility value and risk value .

[0053] Step 6: Construct a modified Boltzmann distribution based on the long-term utility and risk values ​​of each candidate anti-interference strategy, and select trajectory strategies based on the obstacle layout generated by the large model to ensure that the mobile terminal maintains a safe distance from obstacles, while avoiding collisions between trajectories and obstacles between adjacent waypoints, and output the anti-interference strategy. .

[0054] Step 7: The mobile terminal adjusts its pitch, azimuth, and speed to reach the next position and uses the selected transmission power. and transmission channel Combine the image and its own neural network weights Information and feedback strategies Send to the controller.

[0055] Step 8: The mobile terminal receives the feedback result generated by the controller through the control link. The feedback results include environmental characteristics. Interference characteristics and latency When feedback strategy At the same time, it also includes neural network weights from historical mobile terminals. .

[0056] Step 9: The mobile terminal determines the latency based on the time delay. Energy consumption and feedback overhead Calculate immediate utility :

[0057] Step 10: Calculate immediate risk :

[0058] Step 11: Generate Experience And store in the experience pool :

[0059] Step 12: When the amount of experience in the experience pool is not less than the small batch size of 64, retrieve the experience from the experience pool. 64 experiences were randomly selected to form a mini-batch set. Define the discount factor. and The weight parameters are updated using stochastic gradient descent and backpropagation algorithms. and ,make The network output approximates the long-term cumulative utility, making The network output approximates the long-term accumulated risk, as shown in the following formula:

[0060]

[0061] Step 13: If the feedback strategy is At that time, based on task similarity Update network parameters:

[0062] Step 14: Repeat steps 3-13 until the condition is met. and The algorithm converges once the preset number of training time slots is reached.

[0063] It should be noted that the converged neural network, and the anti-interference communication strategy decision model derived therefrom, can adaptively select the optimal transmission power, transmission channel, trajectory, and feedback strategy according to the current system state. Under the premise of meeting security constraints, it balances communication performance, mobile terminal energy consumption, and large model feedback overhead to reduce latency and energy consumption, and achieve reliable communication that resists intelligent interference.

[0064] Depend on Figures 2-4 It can be seen that the embodiments of this application can effectively reduce latency and energy consumption, while improving overall efficiency.

[0065] In summary, the high-efficiency 3D mobile terminal anti-interference communication method based on large-scale model collaboration according to embodiments of the present invention utilizes reinforcement learning combined with information such as channel gain, latency, and energy consumption to select transmission power, channel, trajectory, and feedback strategy. The large-scale model deployed on the controller infers interference characteristics such as interference distance, jammer speed, and interference type, as well as environmental characteristics such as obstacle layout and line-of-sight changes, based on image and spectrum information sent from itself and the mobile terminal, and feeds back neural network weights from historical mobile terminals. This method rapidly defends against various interference attacks, avoids obstacles, and reduces latency and mobile terminal energy consumption, supporting applications such as environmental monitoring, embodied intelligence, and industrial intelligent inspection.

[0066] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0067] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0070] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0071] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0072] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0073] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0074] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0075] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0076] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0077] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A high-efficiency anti-interference communication method for three-dimensional mobile terminals with coordinated large and small models, characterized in that, Includes the following steps: The system acquires sensor data, communication link quality data, and environmental and interference characteristics from the current time slot mobile terminal, in order to construct a system state that includes the mobile terminal's own state, communication link state, and environmental and interference information. The system state is input into the neural network deployed on the mobile terminal to obtain the long-term utility value and long-term risk value of each candidate anti-interference strategy. A modified Boltzmann distribution is constructed based on the long-term utility value and the long-term risk value. Based on the obstacle layout in the environmental features, the transmit power, transmission channel, trajectory, and feedback strategies that meet the safety constraints are selected from the candidate anti-interference strategies to obtain the optimal anti-interference strategy. After adjusting the trajectory according to the optimal anti-interference strategy, the sensor data, the neural network weights of the mobile terminal, and the feedback strategy are sent to the controller through the selected transmission power and transmission channel to receive the feedback results returned by the controller. The feedback results include the updated environmental characteristics, the updated interference characteristics, and the time delay. The instant utility is calculated based on the latency, mobile terminal power consumption and feedback overhead, and the instant risk is calculated based on whether the latency exceeds a preset threshold, so as to construct the system state, the optimal anti-interference strategy, the instant utility, the instant risk and the system state of the next time slot into an experience pool. Small batches of experience are sampled from the experience pool to update the neural network weight parameters through stochastic gradient descent and backpropagation algorithms. When the feedback strategy requests neural network weights, the neural network weights of historical mobile terminals are fused based on task similarity until the neural network converges, thus obtaining an anti-interference communication strategy decision model.

2. The high-efficiency three-dimensional mobile terminal anti-interference communication method with large and small model collaboration as described in claim 1, characterized in that, Before acquiring the sensor data, communication link quality data, and environmental and interference characteristics fed back by the large model for the current time slot mobile terminal, the following steps are included: The communication process is divided into continuous time slots, and a set of candidate anti-interference strategies is obtained based on the quantization order of the transmit power, the number of available channels, the dimension of the feedback strategy, and the trajectory order. The neural network is constructed on the mobile terminal, and the neural network discount rate, latency coefficient, energy consumption coefficient, feedback overhead coefficient, risk threshold, environmental characteristics, interference characteristics, network weight, latency, and experience pool are initialized.

3. The high-efficiency three-dimensional mobile terminal anti-interference communication method with large and small model collaboration as described in claim 1, characterized in that, The neural network includes a Q network and an R network. The Q network is used to estimate the long-term cumulative utility of each candidate anti-interference strategy, and the R network is used to estimate the long-term cumulative risk of each candidate anti-interference strategy. The Q network and the R network have the same input layer dimension, and the number of neurons in the output layer is the total number of candidate anti-interference strategies.

4. The high-efficiency three-dimensional mobile terminal anti-interference communication method with large and small model collaboration as described in claim 1, characterized in that, The sensor data includes images and their data volume acquired by the mobile terminal from the sensor, its remaining battery power, and generated large model prompt words; the communication link quality data includes the channel gain from the mobile terminal to the controller, the controller's estimated delay, the interfered channel, and the received interference intensity.

5. The high-efficiency three-dimensional mobile terminal anti-interference communication method with large and small model collaboration as described in claim 1, characterized in that, The environmental features are inferred from the large model deployed by the controller itself and from image and spectrum information from the mobile terminal, including line-of-sight changes and obstacle layout; the interference features are inferred from the large model, including jammer speed, interference distance, and interference type.

6. The high-efficiency three-dimensional mobile terminal anti-interference communication method for large and small model collaboration as described in claim 1, characterized in that, A modified Boltzmann distribution is constructed based on the long-term utility value and the long-term risk value, including: The selection probability of each candidate anti-interference strategy is determined by a comprehensive trade-off between the long-term utility and long-term risk of each candidate anti-interference strategy.

7. The high-efficiency three-dimensional mobile terminal anti-interference communication method with large and small model collaboration as described in claim 1, characterized in that, The feedback strategy includes a first feedback level and a second feedback level. The first feedback level is used to request the controller to return the environmental features, the interference features, and the latency. The second feedback level is used to additionally request the neural network weights of the historical mobile terminal based on the first feedback level.

8. The high-efficiency three-dimensional mobile terminal anti-interference communication method for large and small model collaboration as described in claim 1, characterized in that, The immediate utility is a weighted sum of the latency, the mobile terminal power consumption, and the feedback overhead.

9. The high-efficiency three-dimensional mobile terminal anti-interference communication method for large and small model collaboration as described in claim 1, characterized in that, The immediate risk is calculated by accumulating the amount by which the delay exceeds a preset delay threshold.

10. The high-efficiency three-dimensional mobile terminal anti-interference communication method with large and small model collaboration as described in claim 1, characterized in that, The neural network weights based on task similarity fusion of historical mobile terminal data include: Calculate the task similarity between the current mobile terminal and historical mobile terminals, determine the fusion coefficient of each external network weight based on the task similarity, and then perform weighted fusion of the external network weights and the current mobile terminal network weights according to the fusion coefficients.

Images