Joint optimization method for communication and computing resource allocation in non-completely overlapped NOMA (Non-Orthogonal Multiple Access) network with assistant-assisted computing

Abstract

The invention relates to a joint optimization method for communication and computing resource allocation in an incomplete overlapping NOMA network with assistants assisting in computing, and belongs to the technical field of MEC networks. A user needs to unload two heterogeneous tasks to two assistants through the incomplete overlapping NOMA in each time slot; under the condition of meeting the delay cut-off requirements of different tasks and the energy constraint of assistants, the energy consumption of the collaborative edge computing system of the non-completely overlapped NOMA network is minimized by jointly optimizing the computing resources of the users and the assistants, the user unloading data volume, the transmitting power, the transmission time and the selection of the assistants.

Description

Technical Field

[0001] The present invention belongs to the technical field of MEC networks, and in particular relates to a joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network with assistant-assisted computing. Background Art

[0002] Traditional mobile cloud computing (MCC) has the problem of long transmission distance from terminal devices to cloud centers, which makes it difficult to meet application scenarios with high real-time requirements. Driven by the Internet of Things and the fifth generation mobile communication technology (5G), mobile edge computing (MEC) has been proposed as a new network architecture. Its main feature is to push computing resources and services to the edge of the network, close to end users, thereby effectively coping with the challenges brought by massive computing needs. Considering that there are many idle devices with computing power in the actual network, in order to further alleviate the challenge of congestion in MEC networks, it is necessary to combine collaborative technology with edge computing offloading and study the energy consumption of collaborative edge computing systems that minimize non-completely overlapping non-orthogonal multiple access (NOMA). Summary of the invention

[0003] The purpose of the present invention is to provide a joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network with assistant-assisted computing, combining collaborative technology with edge computing offloading, and minimizing the energy consumption of the collaborative edge computing system of the non-completely overlapping NOMA network through joint optimization.

[0004] The present invention proposes a joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network with assistant-assisted computing. In each time slot, a user needs to offload two heterogeneous tasks to two assistants through non-completely overlapping NOMA respectively. Under the conditions of meeting the delay deadline requirements of different tasks and the energy constraints of the assistant, the computing resources of the user and the assistant, the amount of user offloaded data, the transmission power, the transmission time and the assistant selection are jointly optimized to minimize the energy consumption of the collaborative edge computing system of the non-completely overlapping NOMA network. Assuming that time is uniformly discretized into T time slots, the index is , the duration of each time slot is , each user uses a cellular network subchannel, and user tasks have heterogeneous delay requirements; suppose Represents a collection of users, Represents a collection of helpers; Representative User Two tasks; including the following steps: Step 1: Build a model for minimizing the long-term average energy consumption of the MEC system ; Step 2: Optimize computing resources, user transmission power, and offloaded data volume given the selection of assistants for each time slot and the allocation of user transmission time in the two transmission phases: The computing resources, user transmission power, and offload data volume optimization issues of different users are independent of each other. For Users Model ; Solve the optimal local calculation frequency and replace the optimal value in the corresponding constraint. Convert to model : prove and The transmission power and offload data volume optimization models in both cases can be expressed as corresponding convex models. When the time-varying channel model is adopted, the communication between the user and the assistant is in the time slot The normalized value of the channel gain is used to solve the convex model using the CVX toolbox to obtain the optimal transmit power and the optimal amount of offloaded data; Step 3: Assistant selection and transmission time optimization: Step 3.1: Formulate the assistant selection and transmission time optimization sub-problem as a DRL problem: (1) Define the state: in the time slot , the state It is expressed as the reward of the previous time slot, that is: ; (2) Define an action: introduce a new action index variable , representing the task Select the index value of the helper, which ranges from ; Select variables with a facilitator The mapping relationship is:

[0005] Time slot ,action It is expressed as follows:

[0006] (3) Designing a reward function: Adding a penalty mechanism to the reward function , when the helper selects an index value that satisfies mutual exclusivity, ,otherwise, Set the number of the same index value selected by the current assistant, and introduce a no-solution penalty mechanism , when the current action Enable users When there is no solution to the optimization problem of computing resources, user transmission power, and offloaded data volume, ,otherwise ; The optimization goal is to minimize the long-term average energy consumption of the MEC system, so the time slot The reward function in It is expressed as:

[0007] When the optimization sub-problems of computing resources, user transmission power, and offload data volume optimization have no solution, then Set to ", only when the model The average reward will be maximized when the objective function is minimized and all constraints are met; Step 3.2: Using action quantization, the SAC agent based on the improved soft actor-critic SAC algorithm obtains the real-time near-optimal transmission time and assistant selection decision for each time slot: The SAC algorithm uses maximum entropy regularization, which allows the SAC agent to maximize both reward and entropy to obtain an approximate optimal strategy. It is expressed as:

[0008] in, is the state-action distribution determined by the policy, represents the policy entropy, represents the discount factor, is a temperature parameter that balances the importance of entropy and reward; use Represents in state Follow the strategy Execute an action The soft Q value is used to evaluate the quality of the current strategy.

[0009] The step 1 is specifically: make , and Respectively represent the time slot In the task The total amount of data, the number of CPU cycles required to calculate the 1-nat task, and the maximum tolerable computing delay; assuming ,Task The transmission time is shorter; users In time slot There are two transmission phases. In the first transmission phase, the user The two tasks are transmitted simultaneously; in the second transmission phase, the user Mission Transmit separately, using and Respectively represent the time slot ,user duration of the first and second transmission phases; Indicates the task The transmit power in the first transmission phase, , Respectively represent tasks Transmitted power during the first and second transmission phases; make Indicates the assistant's choice, in the time slot When a helper With users Mission When connected, ,otherwise , because each helper At most one computation task can be executed, so there are: ; Each task is calculated by a helper:

[0010] Using the time-varying channel model, user With helpers In the time slot Channel gain for:

[0011] in, Indicates time slot ,user With helpers The distance between and are the path loss exponent and a random variable with unit mean exponential distribution; set up represents the power of additive white Gaussian noise, then the channel gain Normalized to: ; In the first transmission phase of task offloading, SIC is used for decoding. According to the downlink NOMA decoding principle, the channel with smaller gain is decoded first; , for users ,Task and tasks In the first transmission stage The rates are expressed as and ,Task In the second transmission phase The rate is expressed as , which is expressed as follows:

[0012] ; in, is an indicator variable. , ; Otherwise, ; make , , Respectively represent users Mission In the first transmission stage Uninstalls, users Mission In the first transmission stage Uninstalls, users Mission In the second transmission phase The amount of uninstallation, therefore: ,

[0013] ; The data volume constraints for task offloading are as follows:

[0014] In order to meet the user's maximum transmit power constraint ,have:

[0015] In time slot ,user Energy consumption of offloading tasks for:

[0016] The task The local calculation frequency is expressed as ,Task The amount of local computing tasks is ,Task The amount of local computing tasks is , then in the time slot ,Task The computation time for local execution is:

[0017] The maximum tolerable delay constraint for each task is:

[0018] user The constraint on the maximum local computation frequency is:

[0019] in, Indicates user The maximum local calculation frequency of In time slot ,user The local computing energy consumption is expressed as: in, Represents constants related to the user's hardware architecture; make Indicates time slot , the facilitator assigns the task The calculation frequency of the task , The delay of the unloading part is expressed as:

[0020]

[0021] make Indicates helper The maximum calculation frequency assigned by the assistant to the user does not exceed the maximum value , so we have:

[0022] In order to satisfy the maximum tolerable delay constraint of each user, :

[0023]

[0024] For Users , the assistant calculates the task and The energy consumption required for the unloading part is expressed as:

[0025] in, Represents constants related to the helper hardware architecture; make Indicates helper The maximum allowable energy consumption, so the energy constraint of the assistant is expressed as:

[0026] make and Represent the energy consumption weights of the user and the assistant respectively, where: ; Therefore, the MEC system is in the time slot Total energy consumption It is expressed as:

[0027] The long-term average energy consumption minimization problem is expressed as a model :

[0028] The step 2 is specifically: Step 2.1: Given the assistant selection and user transmission time allocation In the case of , the computing resources, user transmission power, and offload data volume optimization issues of different users are independent of each other. , for users ,make and Respectively represent tasks and tasks Selected facilitators, namely: ; Time slot For Users Model :

[0029] Step 2.2: Find the optimal local calculation frequency: variable , , , The optimal value of , , , The following conditions should be met:

[0030]

[0031]

[0032] For Users , in the time slot , its two tasks , The optimal calculation frequency assigned by the assistant and the optimal value of the local calculation frequency variable are expressed as:

[0033]

[0034]

[0035]

[0036] The variable , , , Replace with the optimal value in the corresponding constraint, the model Can be further transformed into a model :

[0037] Step 2.3: Proof and The transmit power and offload data volume optimization models in both cases can be expressed as corresponding convex models: First define ; Case 1, ,have:

[0038]

[0039] Will , , Re-expressed as:

[0040]

[0041]

[0042] The variable , , Substitute the expression into When the model Can be converted into a model equivalent to :

[0043] when , When given, the model The objective function and all constraints are convex, so the model It is a convex model; Case 2, ,have:

[0044]

[0045] Similarly, regain , :

[0046]

[0047] The variable , , Substitute the expression into When the model Can be converted into a model equivalent to :

[0048] when , When given, the model The objective function and all constraints are convex, so the model It is a convex model; Step 2.4: Use the CVX toolbox to solve the problem in step 2.3 when the assistant selects and transmission time , Given a convex model at time , the optimal transmit power and the optimal amount of offloaded data are obtained.

[0049] In the training phase of the SAC algorithm in step 3, the Actor outputs the probability distribution of the action, uses two critics to fit the soft Q value, and the Actor is composed of DNN Approximately, its parameters are , input state and output Gaussian distribution , subject to the mean , the variance is ; The two critics are respectively composed of the main DNN and target DNN Approximation, the parameters are , , , ; When the SAC agent interacts with the non-completely overlapping NOMA-MEC environment, the historical experience are stored in the buffered experience pool, and then the Actor and Critic DNNs extract a small batch of experience from the buffered experience pool. To train and update: The training method is to minimize ,Should It is expressed as:

[0050] in, is the target soft Q value, expressed as:

[0051] in, Indicates that in the next state According to the strategy The actions generated; Strategy The training method is to minimize ,Should It is expressed as:

[0052] Adopt automatic entropy adjustment mechanism to minimize To dynamically adjust , It is expressed as:

[0053] in, is the target entropy; at last, Update it with:

[0054] in, Represents the soft update factor, which is a positive number less than 1.

[0055] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for jointly optimizing the allocation of communication and computing resources in the non-completely overlapping NOMA network with assistant-assisted computing is implemented.

[0056] A computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the joint optimization method for allocating communication and computing resources in the above-mentioned non-completely overlapping NOMA network with assistant-assisted computing.

[0057] By adopting the technical solution of the present invention, under time-varying channels, considering that users have multiple tasks with heterogeneous delay requirements, the long-term energy consumption minimization problem of the collaborative edge computing system based on non-completely overlapping NOMA is solved. The user partially unloads the task to multiple assistants through non-completely overlapping NOMA. While meeting the delay deadline requirements of different tasks and the energy constraints of the assistants, the long-term energy consumption of the system to complete the task is minimized by jointly optimizing the computing resources of the user and assistant, the amount of data offloaded by the user, the transmission power, the transmission time and the selection of the assistant. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 A scene graph to which the present invention is applicable; Figure 2 This is a flow chart of the joint optimization process of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0059] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0060] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or scheme described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way. Embodiment 1 Embodiment 1 of the present invention provides a joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network with assistant-assisted computing, wherein a user needs to offload two heterogeneous tasks to two assistants through non-completely overlapping NOMA in each time slot, respectively, and minimizes the energy consumption of a collaborative edge computing system of a non-completely overlapping NOMA network by jointly optimizing the computing resources of the user and the assistant, the amount of data offloaded by the user, the transmission power, the transmission time, and the assistant selection under the conditions of meeting the delay deadline requirements of different tasks and the energy constraints of the assistant, including the following steps: Step 1: Build a model for the long-term average energy consumption minimization problem : Assume that the system time is evenly discretized into several time slots, indexed as , the duration of each time slot is , in each time slot, each user has multiple independent tasks to complete; assuming that the user is lightweight, it is necessary to offload part of the tasks to the assistant to obtain computational assistance, and each user uses a cellular network subchannel. Consider a more realistic scenario where user tasks have heterogeneous delay requirements. In this case, in order to improve spectrum efficiency, during the task offloading process, such as Figure 1 As shown in the figure, each user uses the non-completely overlapping NOMA technology to offload its multiple tasks to multiple assistants. And because the computing resources of the assistants are limited, it is assumed that each assistant can only perform one computing task at most. In addition, due to the complexity of computational analysis and other reasons, temporarily consider the case where each user has only two heterogeneous tasks, which are offloaded to two assistants respectively. It is assumed that for each NOMA group, a shorter transmission time is given to the task with tighter tolerable delay in each time slot; and it is assumed that the data volume of the task calculation result is very small, ignoring the corresponding download process from the assistant.

[0061] set up Represents a collection of users, Represents a collection of helpers; Representative User Two tasks; , and Respectively refers to the time slot In the task The total amount of data, the number of CPU cycles required to calculate the 1-nat task, and the maximum tolerable computing delay; without loss of generality, assume ,Task The transmission time will be shorter; therefore, users In time slot There are two transmission phases. In the first transmission phase, the user The two tasks are transmitted simultaneously; in the second transmission phase, only the user Mission Transmit separately, using and Respectively represent the time slot ,user duration of the first and second transmission phases; Indicates the task The transmit power in the first transmission phase, , Respectively represent tasks Transmitted power during the first and second transmission phases; make Indicates the assistant's choice, in the time slot When a helper With users Mission When connected, ,otherwise , because each helper At most one computation task can be executed, so there are: ; In addition, each user contains two independent tasks, which are offloaded to two assistants respectively, that is, each task is calculated by one assistant:

[0062] The present invention adopts a time-varying channel model, that is, user With helpers In the time slot Channel gain for:

[0063] in, Indicates time slot ,user With helpers The distance between and are the path loss exponent and a random variable with unit mean exponential distribution; set up represents the power of additive white Gaussian noise, then Normalized to: ; In the first transmission phase of task offloading, SIC is used for decoding. According to the downlink NOMA decoding principle, the channel with smaller gain is decoded first; , for users ,Task and tasks In the first transmission stage The rates are expressed as and ,Task In the second transmission phase The rate is expressed as , which is expressed as follows:

[0064] ; in, is an indicator variable. , ; Otherwise, ; make , , Respectively represent users Mission In the first transmission stage Uninstalls, users Mission In the first transmission stage Uninstalls, users Mission In the second transmission phase The amount of uninstallation, therefore:

[0065] ; The data volume constraints for task offloading are as follows:

[0066] In order to meet the user's maximum transmit power constraint ,have:

[0067] In time slot ,user Energy consumption of offloading tasks for:

[0068] The task The local calculation frequency is expressed as ,Task The amount of local computing tasks is ,Task The amount of local computing tasks is , then in the time slot ,Task The computation time for local execution is:

[0069] The maximum tolerable delay constraint for each task is:

[0070] Due to the limited local computing power of users, there are the following constraints:

[0071] in, Indicates user The maximum local calculation frequency of In time slot ,user The computational energy consumption is expressed as: in, Represents constants related to the user's hardware architecture; make Indicates time slot , the facilitator assigns the task The calculation frequency is therefore ,Task , The delay of the unloading part is expressed as:

[0072]

[0073] make Indicates helper The maximum calculation frequency assigned by the assistant to the user does not exceed the maximum value , so we have:

[0074] In order to satisfy the maximum tolerable delay constraint of each user, :

[0075]

[0076] For Users , the assistant calculates the task and The energy consumption required for the unloading part is expressed as:

[0077] in, Represents constants related to the helper hardware architecture; make Indicates helper The maximum allowable energy consumption, so the energy constraint of the assistant is expressed as:

[0078] make and Represent the energy consumption weights of the user and the assistant respectively, where: ; Therefore, the MEC system is in the time slot Total energy consumption It is expressed as:

[0079] The long-term average energy consumption minimization problem is expressed as the following model: :

[0080]

[0081]

[0082] Model It is a non-convex problem with many closely related variables. In addition, since the system parameters change in different time slots, the problem needs to be solved in real time. A hybrid method combining learning methods with optimization methods based on theoretical derivation is proposed to decompose the original optimization problem into two sub-problems: the optimization sub-problem of computing resources, user transmission power, and offloaded data volume; and the optimization sub-problem of transmission time and assistant selection. Next, solutions are given for these two problems in turn.

[0083] Step 2: In each given time slot, the assistant selects and user transmission time allocation Optimize computing resources, user transmission power, and offload data volume under the following circumstances: Step 2.1: Given the assistant selection and user transmission time allocation In the case of , the computing resources, user transmission power, and offload data volume optimization issues of different users are independent of each other. , for users ,make and Respectively represent tasks and tasks Selected facilitators, namely: ; Time slot For Users Model :

[0084]

[0085]

[0086] Step 2.2: Find the optimal local calculation frequency: variable , , , The optimal value of , , , The following conditions should be met:

[0087]

[0088]

[0089] The proof is as follows: Suppose there is a user , in the time slot , the assistant is assigned to two tasks , The optimal calculation frequency of , , the following conditions are met:

[0090]

[0091] Next, you can find helpers to assign to the two tasks , Other calculation frequencies , , which are smaller than , and satisfy the relevant constraints; obviously, the smaller , The value satisfies all relevant constraints, so , It is also a feasible solution and can further reduce the objective function value, which contradicts the assumption of the assistant's optimal calculation frequency allocation. , The proof process of is similar and will not be repeated here.

[0092] Therefore, for users , in the time slot , its two tasks , The optimal calculation frequency assigned to the assistant and the optimal value of the local calculation frequency variable can be expressed as:

[0093]

[0094]

[0095]

[0096] The variable , , , Replace with the optimal value in the corresponding constraint, the model Can be further transformed into a model :

[0097]

[0098]

[0099] Step 2.3: Proof and The transmit power and offload data volume optimization models in both cases can be expressed as corresponding convex models: First define ; Case 1, ,have:

[0100]

[0101] Will , , Re-expressed as:

[0102]

[0103]

[0104] The variable , , Substitute the expression into When the model Can be converted into a model equivalent to :

[0105]

[0106]

[0107] when , When given, the model The objective function and all constraints are convex, so the model It is a convex model; Case 2, ,have:

[0108]

[0109] Similarly, regain , :

[0110]

[0111] The variable , , Substitute the expression into When the model Can be converted into a model equivalent to :

[0112]

[0113]

[0114] when , When given, the model The objective function and all constraints are convex, so the model It is a convex model; Step 2.4: Use the CVX toolbox to solve the problem in step 2.3 when the assistant selects and transmission time , Given a convex model at time, the optimal transmission power and the optimal amount of offloaded data are obtained; Step 3: Assistant selection and transmission time optimization: Step 3.1: Formulate the assistant selection and transmission time optimization sub-problem as a DRL problem: (1) Define the state: in the time slot , the state It is expressed as the reward of the previous time slot to reduce the dimension of the state space, that is: ; (2) Define actions: In the assistant selection and transmission time optimization subproblem, the action is the assistant selection and transmission time allocation decision. Since the assistant selection of each task is mutually exclusive, the effective action space of the system will be greatly limited, and the assistant selection variable The presence of a large number of 0s will lead to a large sparse generation space, making it difficult to identify which actions are effective in this vast action space. Therefore, in order to solve this problem, a new action index variable is introduced , representing the task Select the index value of the helper, which ranges from ; and The mapping relationship is:

[0115] So, the time slot ,action It is expressed as follows:

[0116] (3) Design reward function: Considering that , the index values ​​selected by the assistants of each user are mutually exclusive, so a penalty mechanism is added to the reward function. , when the helper selects an index value that satisfies mutual exclusivity, ,otherwise, Set the number of the same index value selected by the current assistant, and introduce a no-solution penalty mechanism , when the current action Enable users When there is no solution to the optimization problem of computing resources, user transmission power, and offloaded data volume, ,otherwise ; The optimization goal is to minimize the long-term average energy consumption of the system, so the time slot The reward function in It is expressed as:

[0117] When the optimization sub-problems of computing resources, user transmission power, and offload data volume optimization have no solution, then Set to ", only when the model The average reward is maximized when the objective function is minimized and all constraints are satisfied; Step 3.2: Using action quantization, the SAC agent based on the improved Soft Actor-Critic (SAC) algorithm obtains the real-time near-optimal transmission time and assistant selection decision for each time slot: The SAC algorithm is a DRL algorithm based on the random strategy and offline strategy of the Actor-Critic architecture, which performs well in sampling efficiency and stability. In addition, the SAC algorithm uses maximum entropy regularization, which allows the SAC agent to maximize both reward and entropy to obtain an approximate optimal strategy. It is expressed as:

[0118] in, is the state-action distribution determined by the policy, represents the policy entropy, represents the discount factor, is a temperature parameter that balances the importance of entropy and reward; use Represents in state Follow the strategy Execute an action The soft Q value is used to evaluate the quality of the current strategy; The SAC algorithm consists of a training phase and a decision phase. In the training phase, the Actor outputs the probability distribution of the action. In order to reduce overfitting, two critics are used to fit the soft Q value. Due to the existence of a high-dimensional continuous action space, DNN is used to approximate the Actor and Critic. Specifically, the Actor is composed of DNN Approximately, its parameters are , input state and output Gaussian distribution , subject to the mean , the variance is ; The two critics are respectively composed of the main DNN and target DNN Approximation, the parameters are , , , ; When the SAC agent interacts with the non-completely overlapping NOMA-MEC environment, the historical experience are stored in the buffered experience pool, and then the Actor and Critic DNNs extract a small batch of experience from the buffered experience pool. To train and update: The training method is to minimize ,Should It is expressed as:

[0119] in, is the target soft Q value, expressed as:

[0120] in, Indicates that in the next state According to the strategy The actions generated; To improve strategy, strategy The training method is to minimize ,Should It is expressed as:

[0121] In order to improve the stability of the training phase, an automatic entropy adjustment mechanism is adopted to minimize To dynamically adjust , It is expressed as:

[0122] in, is the target entropy; at last, Update it with:

[0123] in, Represents the soft update factor, which is a positive number less than 1.

[0124] Embodiment 2 The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented by electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0125] Specifically, each step of the method embodiment in the embodiment of the present application can be completed by the hardware integrated logic circuit and / or software instructions in the processor, and the steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor, or a combination of hardware and software modules in the decoding processor. Optionally, the software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps in the above method embodiment in combination with its hardware.

[0126] Embodiment 2 of the present invention provides an electronic device, which may include: a memory and a processor, wherein the memory is used to store a computer program and transmit the program code to the processor. In other words, the processor can call and run the computer program from the memory to implement the method of embodiment 1 of the present application. For example, the processor can be used to execute the above method embodiment according to the instructions in the computer program. In some embodiments of the present application, the processor may include but is not limited to: a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0127] In some embodiments of the present application, the memory includes, but is not limited to, volatile memory and / or non-volatile memory. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus random access memory (Direct Rambus RAM, DR RAM).

[0128] In some embodiments of the present application, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the method provided by the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.

[0129] Embodiment 3 Embodiment 3 of the present invention further provides a computer storage medium on which a computer program is stored. When the computer program is executed by a computer, the computer is enabled to execute the method of the above method embodiment.

[0130] When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the first embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integration. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a digital video disc (digital video disc, DVD)), or a semiconductor medium (e.g., a solid state disk (solid state disk, SSD)), etc.

[0131] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network with assistant-assisted computing. In each time slot, the user needs to offload two heterogeneous tasks to two assistants through non-completely overlapping NOMA. Under the conditions of meeting the delay deadline requirements of different tasks and the energy constraints of the assistants, the energy consumption of the collaborative edge computing system in the non-completely overlapping NOMA network is minimized by jointly optimizing the computing resources of the user and the assistant, the amount of user offloaded data, the transmission power, the transmission time and the assistant selection. Assuming that time is uniformly discretized into T time slots, the index is , the duration of each time slot is , each user uses a cellular network subchannel, and user tasks have heterogeneous delay requirements; suppose Represents a collection of users, Represents a collection of helpers; Representative User Two tasks; it is characterized by The steps include: Step 1: Build a model for minimizing the long-term average energy consumption of the MEC system ; Step 2: Optimize computing resources, user transmission power, and offloaded data volume given the selection of assistants for each time slot and the allocation of user transmission time in the two transmission phases: The computing resources, user transmission power, and offload data volume optimization issues of different users are independent of each other. For Users Model ; Solve the optimal local calculation frequency and replace the optimal value in the corresponding constraint. Convert to model : prove and The transmission power and offload data volume optimization models in both cases can be expressed as corresponding convex models. When the time-varying channel model is adopted, the communication between the user and the assistant is in the time slot The normalized value of the channel gain is used to solve the convex model using the CVX toolbox to obtain the optimal transmit power and the optimal amount of offloaded data; Step 3: Assistant selection and transmission time optimization: Step 3.1: Formulate the assistant selection and transmission time optimization sub-problem as a DRL problem: (1) Define the state: in the time slot , the state It is expressed as the reward of the previous time slot, that is: ; (2) Define an action: introduce a new action index variable , representing the task Select the index value of the helper, which ranges from ; Select variables with a facilitator The mapping relationship is: Time slot ,action It is expressed as follows: (3) Designing a reward function: Adding a penalty mechanism to the reward function , when the helper selects an index value that satisfies mutual exclusivity, ,otherwise, Set the number of the same index value selected by the current assistant, and introduce a no-solution penalty mechanism , when the current action Enable users When there is no solution to the optimization problem of computing resources, user transmission power, and offloaded data volume, ,otherwise ; The optimization goal is to minimize the long-term average energy consumption of the MEC system, so the time slot The reward function in It is expressed as: When the optimization sub-problems of computing resources, user transmission power, and offload data volume optimization have no solution, then Set to " ", only when the model The average reward will be maximized when the objective function is minimized and all constraints are met; Step 3.2: Using action quantization, the SAC agent based on the improved soft actor-critic SAC algorithm obtains the real-time near-optimal transmission time and assistant selection decision for each time slot: The SAC algorithm uses maximum entropy regularization, which allows the SAC agent to maximize both reward and entropy to obtain an approximate optimal strategy. It is expressed as: in, is the state-action distribution determined by the policy, represents the policy entropy, represents the discount factor, is a temperature parameter that balances the importance of entropy and reward; use Represents in state Follow the strategy Execute an action The soft Q value is used to evaluate the quality of the current strategy.

2. A joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network for assistant-assisted computing according to claim 1, characterized in that The step 1 is specifically as follows: make , and Respectively represent the time slot In the task The total amount of data, the number of CPU cycles required to calculate the 1-nat task, and the maximum tolerable computing delay; assuming ,Task The transmission time is shorter; users In time slot There are two transmission phases. In the first transmission phase, the user The two tasks are transmitted simultaneously; in the second transmission phase, the user Mission Transmit separately, using and Respectively represent the time slot ,user duration of the first and second transmission phases; Indicates the task The transmit power in the first transmission phase, , Respectively represent tasks Transmitted power during the first and second transmission phases; make Indicates the assistant's choice, in the time slot When a helper With users Mission When connected, ,otherwise , because each helper At most one computation task can be executed, so there are: ； Each task is calculated by a helper: Using the time-varying channel model, user With helpers In the time slot Channel gain for: in, Indicates time slot ,user With helpers The distance between and are the path loss exponent and a random variable with unit mean exponential distribution; set up represents the power of additive white Gaussian noise, then the channel gain Normalized to: ; In the first transmission phase of task offloading, SIC is used for decoding. According to the downlink NOMA decoding principle, the channel with smaller gain is decoded first; , for users ,Task and tasks In the first transmission stage The rates are expressed as and ,Task In the second transmission phase The rate is expressed as , which is expressed as follows: ； in, is an indicator variable. , ; Otherwise, ; make , , Respectively represent users Mission In the first transmission stage Uninstalls, users Mission In the first transmission stage Uninstalls, users Mission In the second transmission phase The amount of uninstallation, therefore: ； The data volume constraints for task offloading are as follows: In order to meet the user's maximum transmit power constraint ,have: In time slot ,user Energy consumption of offloading tasks for: The task The local calculation frequency is expressed as ,Task The amount of local computing tasks is ,Task The amount of local computing tasks is , then in the time slot ,Task The computation time for local execution is: The maximum tolerable delay constraint for each task is: user The constraint on the maximum local computation frequency is: in, Indicates user The maximum local calculation frequency of In time slot ,user The local computing energy consumption is expressed as: in, Represents constants related to the user's hardware architecture; make Indicates time slot , the facilitator assigns the task The calculation frequency of the task , The delay of the unloading part is expressed as: make Indicates helper The maximum calculation frequency assigned by the assistant to the user does not exceed the maximum value , so we have: In order to satisfy the maximum tolerable delay constraint of each user, : For Users , the assistant calculates the task and The energy consumption required for the unloading part is expressed as: in, Represents constants related to the helper hardware architecture; make Indicates helper The maximum allowable energy consumption, so the energy constraint of the assistant is expressed as: make and Represent the energy consumption weights of the user and the assistant respectively, where: ； Therefore, the MEC system is in the time slot Total energy consumption It is expressed as: The long-term average energy consumption minimization problem is expressed as a model : 。 3. A joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network with assistant-assisted computing according to claim 2, characterized in that The step 2 is specifically: Step 2.1: Given the assistant selection and user transmission time allocation In the case of , the computing resources, user transmission power, and offload data volume optimization issues of different users are independent of each other. , for users ,make and Respectively represent tasks and tasks Selected facilitators, namely: ; Time slot For Users Model : Step 2.2: Find the optimal local calculation frequency: variable , , , The optimal value of , , , The following conditions should be met: For Users , in the time slot , its two tasks , The optimal calculation frequency assigned by the assistant and the optimal value of the local calculation frequency variable are expressed as: The variable , , , Replace with the optimal value in the corresponding constraint, the model Can be further transformed into a model : Step 2.3: Proof and The transmit power and offload data volume optimization models in both cases can be expressed as corresponding convex models: First define ; Case 1, ,have: Will , , Re-expressed as: The variable , , Substitute the expression into When the model Can be converted into a model equivalent to : when , Given The objective function and all constraints are convex, so the model It is a convex model; Case 2, ,have: Similarly, regain , : The variable , , Substitute the expression into When the model Can be converted into a model equivalent to : when , Given The objective function and all constraints are convex, so the model It is a convex model; Step 2.4: Use the CVX toolbox to solve the problem in step 2.3 when the assistant selects and transmission time , Given a convex model at time , the optimal transmit power and the optimal amount of offloaded data are obtained.

4. The method for joint optimization of communication and computing resource allocation in a non-completely overlapping NOMA network with assistant-assisted computing according to claim 3, characterized in that: In the training phase of the SAC algorithm in step 3, the Actor outputs the probability distribution of the action, uses two critics to fit the soft Q value, and the Actor is composed of DNN Approximately, its parameters are , input state and output Gaussian distribution , subject to the mean , the variance is ; The two critics are respectively composed of the main DNN and target DNN Approximation, the parameters are , , , ; When the SAC agent interacts with the non-completely overlapping NOMA-MEC environment, the historical experience are stored in the buffered experience pool, and then the Actor and Critic DNNs extract a small batch of experience from the buffered experience pool. To train and update: The training method is to minimize ,Should It is expressed as: in, is the target soft Q value, expressed as: in, Indicates that in the next state According to the strategy The actions generated; Strategy The training method is to minimize ,Should It is expressed as: Adopt automatic entropy adjustment mechanism to minimize To dynamically adjust , It is expressed as: in, is the target entropy; at last, Update it with: in, Represents the soft update factor, which is a positive number less than 1.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network for assistant-assisted computing as described in any one of claims 1 to 4.

6. A computer storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the joint optimization method for communication and computing resource allocation in a non-completely overlapping NOMA network for assistant-assisted computing as described in any one of claims 1 to 4.

Images