Dynamic resource allocation method for NCO-NOMA-assisted IRS-MEC system
By introducing a hybrid optimization method of the NCO-NOMA transmission scheme and DRL framework in the NOMA auxiliary IRS-MEC system, the shortcomings of existing systems in handling user tasks are solved, and more efficient resource allocation and energy consumption reduction are achieved.
Patent Information
- Application Number
- CN202510065993.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-16
- Publication Date
- 2025-04-25
AI Technical Summary
The existing NOMA assisted IRS-MEC system has shortcomings in handling user tasks heterogeneity, making it difficult to effectively support diversified application scenarios and reduce system energy consumption.
An NCO-NOMA transmission scheme is proposed, allowing different users in the same NOMA group to have independent transmission time according to their task requirements. By combining traditional optimization methods with a hybrid optimization method of the DRL framework, IRS phase shift, transmission time and computational frequency allocation of MEC servers are optimized to minimize energy consumption.
It significantly improves the flexibility and efficiency of resource allocation, can support diversified application scenarios more efficiently, and further reduces system energy consumption.
Smart Images
Figure CN119893592A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of MEC technology, and in particular to a dynamic resource allocation method for an NCO-NOMA assisted IRS-MEC system. Background Art
[0002] In recent years, with the rapid popularization of mobile Internet and smart devices, the demand for computing resources in mobile applications has exploded, especially in scenarios such as real-time computing, virtual reality and intelligent driving, where many computing tasks are highly sensitive to latency. To meet this challenge, the European Telecommunications Standards Institute proposed the technical concept of mobile edge computing (MEC). By deploying MEC servers close to the user side, computing tasks are transferred from the cloud to the edge, significantly reducing network latency and improving computing efficiency. At the same time, in order to further improve the performance of MEC systems, non-orthogonal multiple access (NOMA) technology has become a research hotspot. NOMA supports large-scale device connections by multiplexing multiple users in the same resource block, while effectively reducing system energy consumption. In addition, NOMA technology has strong compatibility and can be seamlessly integrated with emerging technologies such as intelligent reflecting surfaces (IRS). IRS is a programmable metasurface that can optimize the wireless propagation environment by dynamically adjusting the phase and amplitude of the reflected signal, bringing significant gains to the MEC system.
[0003] Currently, most of the research on NOMA-assisted IRS-MEC systems is based on the fully overlapping NOMA (CO-NOMA) transmission scheme. In this scheme, all users in the same NOMA group share the same transmission time. Although CO-NOMA performs well in scenarios where the user task data volume and latency requirements are similar, it has obvious shortcomings in dealing with the heterogeneity of user tasks. Summary of the invention
[0004] The main purpose of the present invention is to provide a dynamic resource allocation method for an NCO-NOMA-assisted IRS-MEC system to solve the problems existing in the prior art. A non-completely overlapping NOMA (NCO-NOMA) transmission scheme is proposed in the MEC system, which allows different users in the same NOMA group to have independent transmission time according to their task requirements, significantly improving the flexibility and efficiency of resource allocation. Compared with CO-NOMA, the NCO-NOMA scheme is more suitable for handling complex and heterogeneous service quality requirements, and can more efficiently support the diversified application scenarios of MEC users, while further reducing system energy consumption, showing broad application prospects.
[0005] In order to achieve the above object, the solution of the present invention is: A dynamic resource allocation method for an NCO-NOMA assisted IRS-MEC system is provided for a MEC system including a BS, an IRS and multiple user groups, wherein the BS is equipped with a MEC server, and the user offloads the computing task to the BS for processing through NOMA transmission, including the following steps: Step 1: Divide all users into NOMA groups, each containing only two users, represents the set of all NOMA groups, Indicates user groups; the system time is divided into time slots, Use to indicate time slots; the goal of the dynamic resource allocation method is to optimize the energy consumption of the MEC system. The energy consumption optimization problem is expressed as: ; in, It is a time slot The total energy consumption of completing the computing tasks of all user groups within a period; Step 2: Perform a hybrid optimization method that combines traditional optimization methods with the DRL framework, including the following sub-steps: Step 2.1 The energy consumption optimization problem is decomposed into the IRS phase shift and transmission time optimization sub-problems, the calculation frequency allocation sub-problem and the transmission power optimization sub-problem; Step 2.2 Definition , Represents user groups Users , The total effective gain to the BS is discussed separately in the transmit power optimization sub-problem. and Two cases; in the first case, the convexity of the optimization problem is proved and solved using the convex optimization toolbox; in the second case, the optimization problem is proved to belong to differential convex function programming and solved by the concave-convex algorithm; Step 2.3 In the computational frequency allocation subproblem, derive the closed-form expression of the optimal solution based on the monotonicity of the objective function; Step 3: For the IRS and transmission time optimization sub-problems, we model them as DRL problems by introducing several penalty mechanisms in the reward function. Specifically: Step 3.1 Define the state space Time slot Status It is defined as the set of effective channel gains, namely: ; in, Indicates time slot Internal User Channel gain with IRS, Indicates time slot The channel gain between the IRS and the BS, Indicates time slot Internal User Channel gain with BS; Step 3.2 Define the action space Time slot Actions in Defined as: ; in, Indicates that IRS is in time slot Neidi The phase shift of the reflective element, represents the total number of reflective elements, , Represents user groups Users , the transmission times during the shared transmission period and the transmission times during the individual transmission period; Step 3.3 Define the reward function After taking an action, the DRL agent will receive a reward, which is used to evaluate the consistency of the action selected by the agent with the optimization goal; The reward inside is defined as: ; in, represents the negative exponent of the energy consumption of the system under consideration, and ; When the slot When there is no solution to the subproblems of optimizing the transmit power and calculating the frequency allocation in and is the penalty value; specifically, when the transmit power optimization subproblem has no solution ,otherwise ; When there is no solution to the computational frequency allocation subproblem ,otherwise ; Step 4: Based on the SAC algorithm, obtain the approximately optimal real-time IRS phase shift and transmission time strategy.
[0006] In the step 1, The following constraints need to be met: Constraint 1. In any time slot In user group No. Users The total amount of task data Must be limited by uplink bandwidth and power; Constraint 2. See Figure 2 , in any time slot In user group Users and Transmit power , and The maximum transmission power limit is not exceeded. , where the superscript , Respectively represent the shared transmission period and the individual transmission period; Constraint 3. In any time slot In user group Users Transfer times during shared transfer periods Calculate latency with MEC server The sum does not exceed its maximum delay period , its users Transfer times during shared transfer periods and individual transfer periods , Calculate latency with MEC server The sum does not exceed its maximum delay period ; Constraint 4. In any time slot BS is assigned to user groups No. Users Computing resources The total cannot exceed the maximum calculation frequency of the MEC server .
[0007] The step 4 is specifically: Step 4.1 Given a discount factor , Soft Copy Factor , Maximum number of rounds and the number of time slots per round ; Step 4.2 Randomly initialize weight parameters , , , and , and clear the experience replay pool ; Step 4.3 SAC enters a training process consisting of two layers of nested loops, with a total training rounds, each round contains Time slots: At the beginning of each round of training, initialize the environment and obtain the initial state ; Then in each time slot In the process, the SAC agent is based on the current state Select actions by strategy ; After executing the action, the environment will feedback immediate rewards , and transfer the state to the next state ; At this time, the experience sample obtained in the current time slot Stored in the replay pool ; Next, from the playback pool Randomly extract small batches of data experience for training and update the weight parameters , , , and ; Step 4.4 After training is completed, the SAC algorithm outputs the optimal weight parameters of the policy network, thereby obtaining an approximately optimal real-time solution for IRS phase shift and transmission time.
[0008] After adopting the above technical solution, the present invention has the following technical effects: The key to the present invention is that when the IRS phase shift and transmission time are given, the transmission power optimization of each group of users is independent of the calculation frequency allocation of the MEC server; by decomposing the problem into IRS phase shift and transmission time optimization sub-problems, calculation frequency allocation sub-problems and transmission power optimization sub-problems, an efficient solution is achieved.
[0009] For the optimization of transmit power, the present invention discusses two cases: in the first case, the convexity of the optimization problem is proved and CVX is used to solve it; in the second case, the optimization problem is proved to be DC planning and CCCP is used to solve it; for the calculation of frequency allocation, a closed-form expression of the optimal solution is derived, which further reduces the computational complexity; on this basis, the present invention proposes a DRL framework combined with a penalty mechanism, introduces a penalty term in the reward function, and adopts the SAC algorithm to obtain the approximate optimal real-time strategy of IRS phase shift and transmission time, providing an efficient and reliable solution for dynamic resource optimization. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] Figure 1 This is a scenario diagram of the MEC system based on IRS assistance.
[0011] Figure 2 For user groups In time slot Schematic diagram of the task offloading process.
[0012] Figure 3 Block diagram of IRS-MEC system resource optimization for DRL-based NCO-NOMA. DETAILED DESCRIPTION
[0013] In order to further explain the technical solution of the present invention, the present invention is described in detail below through specific embodiments.
[0014] refer to Figure 1-3 As shown, the present invention discloses a dynamic resource allocation method for an NCO-NOMA assisted IRS-MEC system, see Figure 1 , the present invention mainly targets a MEC system including a base station (BS), an IRS and multiple user groups, wherein the BS is equipped with a MEC server, and the user offloads the computing task to the BS through NOMA transmission. The present invention minimizes the long-term energy consumption of the MEC system and improves the overall performance and energy efficiency of the system by jointly optimizing the transmission power and transmission time of the user, the computing frequency allocation of the MEC server and the IRS phase shift. Specifically, see Figure 3 , the present invention comprises the following steps: Step 1: Divide all users into NOMA groups, each containing only two users, represents the set of all NOMA groups, Indicates user groups; the system time is divided into time slots, Use to indicate time slots; the goal of the dynamic resource allocation method is to optimize the energy consumption of the MEC system (that is, to minimize the energy consumption), and the energy consumption optimization problem is expressed as: ; in, It is a time slot The total energy consumption of completing the computing tasks of all user groups within a period; Step 2: Perform a hybrid optimization method that combines traditional optimization methods with a deep reinforcement learning (DRL) framework, which specifically includes the following sub-steps: Step 2.1 Since the transmit power optimization of each group of users is independent of the calculation frequency allocation of the MEC server when the IRS phase shift and transmission time are given, the energy consumption optimization problem is decomposed into the IRS phase shift and transmission time optimization sub-problem, the calculation frequency allocation sub-problem and the transmit power optimization sub-problem; Step 2.2 Definition , Represents user groups Users , The total effective gain to the BS is discussed separately in the transmit power optimization sub-problem. and Two cases; in the first case, the convexity of the optimization problem is proved and solved using the Convex Optimization Toolbox (CVX); in the second case, the optimization problem is proved to belong to the Differential Convex Programming (DC) and solved by the Concave-Convex Programming (CCCP) algorithm; Step 2.3 In the computational frequency allocation subproblem, based on the monotonicity of the objective function, a closed-form expression of the optimal solution is derived, thereby reducing computational complexity and improving efficiency; Step 3: For the IRS and transmission time optimization sub-problems, we model them as DRL problems by introducing several penalty mechanisms in the reward function. Specifically: Step 3.1 Define the state space In the considered NCO-NOMA-assisted IRS-MEC system, the time slot Status It is defined as the set of effective channel gains, namely: ; in, Indicates time slot Internal User The channel gain with IRS, Indicates time slot The channel gain between the IRS and the BS, Indicates time slot Internal User Channel gain with BS; Step 3.2 Define the action space In order to obtain the optimal IRS phase shift and transmission time, the time slot Actions in Defined as: ; in, Indicates that IRS is in time slot Neidi The phase shift of the reflective element, represents the total number of reflective elements, , Represents user groups Users , the transmission times during the shared transmission period and the transmission times during the individual transmission period; Step 3.3 Define the reward function After taking an action, the DRL agent will be rewarded, which is used to evaluate the consistency of the action selected by the agent with the optimization goal; in the MEC system, in order to obtain the minimum energy consumption, the time slot The reward inside is defined as: ; in, represents the negative exponent of the energy consumption of the system under consideration, and ; When the slot When there is no solution to the subproblems of optimizing the transmit power and calculating the frequency allocation in and is the penalty value; specifically, when the transmit power optimization subproblem has no solution ,otherwise ; When there is no solution to the computational frequency allocation subproblem ,otherwise ; Step 4: Based on the SAC (soft actor-critic) algorithm, obtain the approximately optimal real-time IRS phase shift and transmission time strategy.
[0015] Through the above scheme, the key of the present invention is that when the IRS phase shift and transmission time are given, the transmission power optimization of each group of users is independent of the calculation frequency allocation of the MEC server; by decomposing the problem into IRS phase shift and transmission time optimization sub-problems, calculation frequency allocation sub-problems and transmission power optimization sub-problems, an efficient solution is achieved.
[0016] For the optimization of transmit power, the present invention discusses two cases: in the first case, the convexity of the optimization problem is proved and CVX is used to solve it; in the second case, the optimization problem is proved to be DC planning and CCCP is used to solve it; for the calculation of frequency allocation, a closed-form expression of the optimal solution is derived, which further reduces the computational complexity; on this basis, the present invention proposes a DRL framework combined with a penalty mechanism, introduces a penalty term in the reward function, and adopts the SAC algorithm to obtain the approximate optimal real-time strategy of IRS phase shift and transmission time, providing an efficient and reliable solution for dynamic resource optimization.
[0017] Specific embodiments of the present invention are shown below.
[0018] In step 1 above, The following constraints need to be met: Constraint 1. In any time slot In user group No. Users The total amount of task data Must be limited by uplink bandwidth and power; Constraint 2. See Figure 2 , in any time slot In user group Users and Transmit power , and The maximum transmission power limit is not exceeded. , where the superscript , Respectively represent the shared transmission period and the individual transmission period; Constraint 3. In any time slot In user group Users Transfer times during shared transfer periods Calculate latency with MEC server The sum does not exceed its maximum delay period , its users Transfer times during shared transfer periods and individual transfer periods , Calculate latency with MEC server The sum does not exceed its maximum delay period ; Constraint 4. In any time slot BS is assigned to user groups No. Users Computing resources The total cannot exceed the maximum calculation frequency of the MEC server .
[0019] The above step 4 is specifically: Step 4.1 Given a discount factor , Soft Copy Factor , Maximum number of rounds and the number of time slots per round ; Step 4.2 Randomly initialize weight parameters , , , and , and clear the experience replay pool ; Step 4.3 SAC enters a training process consisting of two layers of nested loops, with a total training rounds, each round contains Time slots: At the beginning of each round of training, initialize the environment and obtain the initial state ; Then in each time slot In the process, the SAC agent is based on the current state Select actions by strategy ; After executing the action, the environment will feedback immediate rewards , and transfer the state to the next state ; At this time, the experience sample obtained in the current time slot Stored in the replay pool ; Next, from the playback pool Randomly extract small batches of data experience for training and update the weight parameters , , , and ; Step 4.4 After training is completed, the SAC algorithm outputs the optimal weight parameters of the policy network, thereby obtaining an approximately optimal real-time solution for IRS phase shift and transmission time.
[0020] The above embodiments and drawings do not limit the product form and style of the present invention. Any appropriate changes or modifications made by ordinary technicians in the relevant technical field should be deemed to be within the patent scope of the present invention.
Claims
1. A dynamic resource allocation method for an NCO-NOMA assisted IRS-MEC system, for a MEC system comprising a BS, an IRS and multiple user groups, wherein the BS is equipped with a MEC server, and the user offloads the computing task to the BS for processing via NOMA transmission, characterized in that The following steps are involved: Step 1: Divide all users into NOMA groups, each containing only two users, represents the set of all NOMA groups, Indicates user groups; the system time is divided into time slots, Use to indicate time slots; the goal of the dynamic resource allocation method is to optimize the energy consumption of the MEC system. The energy consumption optimization problem is expressed as: ； in, It is a time slot The total energy consumption of completing the computing tasks of all user groups within a period; Step 2: Perform a hybrid optimization method that combines traditional optimization methods with the DRL framework, including the following sub-steps: Step 2.1 The energy consumption optimization problem is decomposed into the IRS phase shift and transmission time optimization sub-problems, the calculation frequency allocation sub-problem and the transmission power optimization sub-problem; Step 2.2 Definition , Respectively represent user groups Users , The total effective gain to the BS is discussed separately in the transmit power optimization sub-problem. and Two cases; in the first case, the convexity of the optimization problem is proved and solved using the convex optimization toolbox; in the second case, the optimization problem is proved to belong to differential convex function programming and solved by the concave-convex algorithm; Step 2.3 In the computational frequency allocation subproblem, derive the closed-form expression of the optimal solution based on the monotonicity of the objective function; Step 3: For the IRS and transmission time optimization sub-problems, we model them as DRL problems by introducing several penalty mechanisms in the reward function. Specifically: Step 3.1 Define the state space Time slot Status It is defined as the set of effective channel gains, namely: ； in, Indicates time slot Internal User Channel gain with IRS, Indicates time slot The channel gain between the IRS and the BS, Indicates time slot Internal User Channel gain with BS; Step 3.2 Define the action space Time slot Actions in Defined as: ； in, Indicates that IRS is in time slot Neidi The phase shift of the reflective element, represents the total number of reflective elements, , Represents user groups Users , the transmission times during the shared transmission period and the transmission times during the individual transmission period; Step 3.3 Define the reward function After taking an action, the DRL agent will receive a reward, which is used to evaluate the consistency of the action selected by the agent with the optimization goal; The reward inside is defined as: ； in, represents the negative exponent of the energy consumption of the system under consideration, and ; When the slot When there is no solution to the subproblems of optimizing the transmit power and calculating the frequency allocation in and is the penalty value; specifically, when the transmit power optimization subproblem has no solution ,otherwise ; When there is no solution to the computational frequency allocation subproblem ,otherwise ; Step 4: Based on the SAC algorithm, obtain the approximately optimal real-time IRS phase shift and transmission time strategy.
2. The dynamic resource allocation method of the NCO-NOMA assisted IRS-MEC system as claimed in claim 1, characterized in that In the step 1, The following constraints need to be met: Constraint 1. In any time slot In user group No. Users The total amount of task data Must be limited by uplink bandwidth and power; Constraint 2. Referring to Figure 2, in any time slot In user group Users and Transmit power , and The maximum transmission power limit is not exceeded. , where the superscript , Respectively represent the shared transmission period and the individual transmission period; Constraint 3. In any time slot In user group Users Transfer times during shared transfer periods Calculate latency with MEC server The sum does not exceed its maximum delay period , its users Transfer times during shared transfer periods and individual transfer periods , Calculate latency with MEC server The sum does not exceed its maximum delay period ; Constraint 4. In any time slot BS is assigned to user groups No. Users Computing resources The total cannot exceed the maximum calculation frequency of the MEC server .
3. The dynamic resource allocation method of the NCO-NOMA assisted IRS-MEC system as claimed in claim 1, characterized in that The step 4 is specifically: Step 4.1 Given a discount factor , Soft Copy Factor , Maximum number of rounds and the number of time slots per round ; Step 4.2 Randomly initialize weight parameters , , , and , and clear the experience replay pool ; Step 4.3 SAC enters a training process consisting of two layers of nested loops, with a total training rounds, each round contains Time slots: At the beginning of each round of training, initialize the environment and obtain the initial state ; Then in each time slot In the process, the SAC agent is based on the current state Select actions by strategy ; After executing the action, the environment will feedback immediate rewards , and transfer the state to the next state ; At this time, the experience sample obtained in the current time slot Stored in the replay pool ; Next, from the playback pool Randomly extract small batches of data experience for training and update the weight parameters , , , and ; Step 4.4 After training is completed, the SAC algorithm outputs the optimal weight parameters of the policy network, thereby obtaining an approximately optimal real-time solution for IRS phase shift and transmission time.