Perceptual communication computing joint optimization method and system in mobile edge computing system

Abstract

The invention discloses a sensing communication computing joint optimization method and system in a mobile edge computing system. The method comprises the following steps: initializing state information; performing problem reconstruction; the upper layer sensing duration is optimized; performing lower-layer unloading decision and resource allocation optimization; judging whether a preset convergence condition or the maximum number of iterations is reached; if a preset convergence condition or the maximum number of iterations is reached, outputting an optimal resource allocation scheme and a system bandwidth allocation scheme; otherwise, returning to execute the optimization of the upper-layer sensing duration. According to the method, the potential that the mobile edge computing system provides computing support for the sensing task based on the multi-modal data is fully mined, the requirement of the acquisition and processing process of the multi-modal data for time synchronization is considered, and the effectiveness and accuracy of sensing information utilization are ensured.

Description

Technical Field

[0001] The present application relates to the field of mobile communications, and in particular, to a perception, communication and computing joint optimization method and system in a mobile edge computing system. Background Art

[0002] With the widespread popularity of smart terminals and the rapid development of artificial intelligence technology, terminal devices can autonomously collect multimodal data through built-in sensors of various types and perform complex perception tasks. Using these multi-source data, perception tasks with environmental information as input can provide a more comprehensive situational understanding, thereby supporting complex decisions and operations. These multimodal data need to be fused at different levels such as data level, feature level and decision level to meet diverse business needs. However, whether it is processing complex data such as radar point cloud and image information, or running multimodal information fusion algorithms, they are faced with huge computing pressure and strict latency requirements. The local computing power and resource reserves of mobile terminals are limited, and cannot meet the reliability and effectiveness requirements of intelligent applications for real-time processing of massive environmental data. In this context, Mobile Edge Computing (MEC), as an emerging computing paradigm that is closer to the terminal, provides an effective solution to meet the low latency requirements of multimodal data processing. The MEC system deploys resources such as storage and computing to the edge of the network, significantly reducing the latency caused by data transmission to the cloud, improving task processing efficiency, and reducing the computing burden on the terminal side. Given that the computing power of edge servers is not as sufficient as that of traditional cloud centers, and the communication resources of the core network are limited, the key issue in the MEC system is how to efficiently offload computing tasks and allocate resources, while meeting resource constraints and optimizing system performance such as application request latency, accuracy or energy consumption. Task offloading and allocation in the MEC system mainly focus on the optimization of related indicators such as task completion latency, terminal energy consumption or the weighted sum of the two. It mainly uses convex optimization, metaheuristics or alternating iterations to jointly realize task offloading decisions, offloading ratios, bandwidth, transmission power, processor frequency and other system communication and computing resource allocation. With the evolution of computing requirements and algorithms, task processing is becoming increasingly complex, and the differentiated dependence on different types of processing units integrated on terminal devices is more significant. Heterogeneous computing environments include different types of processors such as CPU, GPU, NPU and FPGA, which have their own advantages in performance, power consumption and applicable task types. Effective management of heterogeneous computing resources can not only improve system performance, but also enhance the flexibility and adaptability of the system.

[0003] With the continuous development of artificial intelligence applications, the computing tasks that edge servers in MEC systems need to process present two characteristics: on the one hand, multimodal fusion of computing data, and on the other hand, heterogeneous collaboration of computing resources. Although the multimodal data processing requirements are very consistent with the diversity of heterogeneous computing resources, the existing methods are mainly focused on the fusion and utilization methods of multi-source data, focusing on performance improvement algorithms for specific services, and lack of detailed discussions on their underlying resource support. Specifically, the key issues that need to be solved in the joint optimization of perception, communication, and computing for multimodal data in MEC systems are as follows: 1) The existing methods lack MEC system design centered on perception tasks. On the one hand, the multimodal data that drives the perception task needs to be synchronized and aligned during the collection and processing stages to ensure the effectiveness and accuracy of the perception information utilization. On the other hand, the traditional optimization objectives such as task completion delay and terminal energy consumption should no longer be used as a single evaluation dimension, but the perception task performance should be considered comprehensively, which also increases the complexity of problem solving. 2) Existing methods mostly ignore the heterogeneity of computing resources in MEC systems. Based on the preferences of different modal data for different types of computing resources and the performance differences of these resources when processing various types of data, designing a fine-grained computing resource allocation model can further explore the potential for optimal configuration of computing resources, thereby improving the efficiency and flexibility of multimodal data processing.

[0004] Therefore, how to provide a calculation method that takes into account task processing efficiency and terminal energy consumption while meeting task delay requirements has become an urgent problem to be solved in this field. Summary of the invention

[0005] The present application proposes a perception, communication and computing joint optimization method in a mobile edge computing system, comprising the following steps: initializing state information; after completing the initialization of the state information, reconstructing the problem; after completing the problem reconstruction, optimizing the upper-layer perception duration; after completing the optimization of the upper-layer perception duration, optimizing the lower-layer unloading decision and resource allocation; judging whether a preset convergence condition is met or a maximum number of iterations is reached; if the preset convergence condition is met or the maximum number of iterations is reached, outputting the best resource allocation plan and the system bandwidth allocation plan; if the preset convergence condition is not met or the maximum number of iterations is reached, returning to perform the optimization of the upper-layer perception duration.

[0006] In the perception, communication and computing joint optimization method in the mobile edge computing system as described above, initialization of state information includes initialization of system parameters, initialization of mobile terminal position and initialization of corresponding perception task duration, offloading decision and resource allocation.

[0007] In the above-mentioned perception, communication and computing joint optimization method in the mobile edge computing system, before reconstructing the problem, it also includes determining the task s u,m Data volume u,m , expressed as: b u,m =fs u,m b m δ u,m , where δ u,m Represents the perception duration of the mobile terminal u modality m data, which must meet the synchronization constraints of multi-modal data collection b m Indicates the sampling rate; fs u,m Indicates the data bit depth of mode m.

[0008] In the above-mentioned joint optimization method of perception, communication and computing in the mobile edge computing system, the problem reconstruction includes: Restructured into a bilevel programming problem, where U = {1, ..., u, ..., U} represents U mobile terminals, M = {1, ..., M} represents the modal type of different data corresponding to each sensor, and task s u,m Data volume u,m , D u,m Indicates the mth mode of mobile terminal u in the corresponding task s u,m The processing delay, E u,m represents the energy consumption of the corresponding mobile terminal u, δ represents the perceived duration of the task, o represents the offloading decision variable of the task, B represents the bandwidth allocated to the task, and n represents the number of computing resources allocated to the task.

[0009] In the above-mentioned mobile edge computing system, the perception communication computing joint optimization method, the perception task s u,m The processing delay D u,m and the corresponding energy consumption E of mobile terminal u u,m It is expressed as:

[0010]

[0011] o u,m Represents tasks u,m The uninstall decision variables, Represents tasks u,m The communication delay, Represents tasks u,m The computational latency of offloading to edge servers, Represents tasks u,m The corresponding calculation delay is processed locally on the mobile terminal. Represents tasks u,m The communication energy consumption of the corresponding mobile terminal needs to be offloaded. Represents tasks u,m The computational energy consumption of mobile terminal u when processing locally on the mobile terminal.

[0012] A perception, communication and computing joint optimization system in a mobile edge computing system, specifically comprising: an initialization unit, a problem reconstruction unit, an upper-layer perception duration optimization unit, a lower-layer unloading decision and resource allocation optimization unit, a judgment unit and an output unit; the initialization unit is used to initialize state information; the problem reconstruction unit is used to reconstruct the problem; the upper-layer perception duration optimization unit is used to optimize the upper-layer perception duration; the lower-layer unloading decision and resource allocation optimization unit is used to perform lower-layer unloading decision and resource allocation optimization; the judgment unit is used to judge whether a preset convergence condition is reached or a maximum number of iterations is reached; if the preset convergence condition is reached or the maximum number of iterations is reached, the output unit outputs an optimal resource allocation plan and a system bandwidth allocation plan; if the preset convergence condition is not reached or the maximum number of iterations is reached, the execution of the upper-layer perception duration optimization unit is returned.

[0013] In the perception, communication and computing joint optimization system in the mobile edge computing system as described above, the initialization unit performs initialization of the state information, including initialization of system parameters, initialization of the mobile terminal position and initialization of the corresponding perception task duration, offloading decision and resource allocation.

[0014] In the perception, communication and computing joint optimization system in the mobile edge computing system as described above, before the problem reconstruction unit performs problem reconstruction, it also includes determining the task s u,m Data volume u,m , expressed as: b u,m =fs u,m b m δ u,m , where δ u,m Represents the perception duration of the mobile terminal u modality m data, which must meet the synchronization constraints of multi-modal data collection b m Indicates the sampling rate; fs u,m Indicates the data bit depth of mode m.

[0015] In the perception, communication and computing joint optimization system in the mobile edge computing system as described above, the problem reconstruction unit performs problem reconstruction including: Restructured into a bilevel programming problem, where U = {1, ..., u, ..., U} represents U mobile terminals, M = {1, ..., M} represents the modal type of different data corresponding to each sensor, and task s u,m Data volume u,m , D u,m Indicates the mth mode of mobile terminal u in the corresponding task su,m The processing delay, E u,m represents the energy consumption of the corresponding mobile terminal u, δ represents the perceived duration of the task, o represents the offloading decision variable of the task, B represents the bandwidth allocated to the task, and n represents the number of computing resources allocated to the task.

[0016] In the mobile edge computing system described above, the perception communication computing joint optimization system, the problem reconstruction unit, the perception task s u,m The processing delay D u,m and the corresponding energy consumption E of mobile terminal u u,m It is expressed as:

[0017]

[0018] o u,m Represents tasks u,m The uninstall decision variables, Represents tasks u,m The communication delay, Represents tasks u,m The computational latency of offloading to edge servers, Represents tasks u,m The corresponding calculation delay is processed locally on the mobile terminal. Represents tasks u,m The communication energy consumption of the corresponding mobile terminal needs to be offloaded. Represents tasks u,m The computational energy consumption of mobile terminal u when processing locally on the mobile terminal.

[0019] This application has the following beneficial effects:

[0020] (1) This application fully explores the potential of mobile edge computing systems to provide computing support for perception tasks based on multimodal data, takes into account the requirements for time synchronization in the collection and processing of multimodal data, and ensures the effectiveness and accuracy of the use of perception information.

[0021] (2) This application fully explores the potential of mobile edge computing systems for heterogeneous computing resource management and the natural compatibility of heterogeneous computing resources with multimodal data processing, further increasing task processing efficiency and flexibility.

[0022] (3) This application takes into account the computational intensity, delay sensitivity and resource constraints of the perception task, and jointly improves the perception performance and saves terminal energy overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0024] Figure 1 It is a flowchart of a perception, communication and computing joint optimization method in a mobile edge computing system provided in an embodiment of the present application;

[0025] Figure 2 It is a schematic diagram of the internal structure of the perception, communication and computing joint optimization system in the mobile edge computing system provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are 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 those skilled in the art without creative work are within the scope of protection of the present application.

[0027] This application proposes a joint perception-communication-computing method in a mobile edge computing system, which maximizes the perception task processing efficiency and minimizes the terminal energy consumption under the condition of meeting the maximum tolerable delay of the perception task through the joint allocation of perception duration, offloading decision, communication bandwidth and heterogeneous computing resources.

[0028] Consider a multi-terminal MEC system, including an edge server (ES) and U mobile terminals (MT), denoted as U = {1,…,u,…,U}. Each mobile terminal is equipped with multiple sensors, and the modal types of different data corresponding to each sensor are denoted as M = {1,...,M}. The system time is divided into a series of time slots of length Δt according to T = {1,…,t,…,T}. Assume that each sensor device has been synchronized in hardware, that is, they all start to collect data synchronously at the beginning of each time slot. The processing process of each modal data can be regarded as a computing task. Specifically, the mth modality of mobile terminal u in the corresponding task is denoted as s u,m , it is necessary to use the computing resources of the terminal local or edge server to perform the corresponding data processing. Assume that the system includes Q kinds of computing resources, denoted as Q = {1, ..., q, ..., Q}. The purpose of computing and processing the collected data can be to directly obtain usable decision results, or to perform subsequent (feature level or decision level) perception fusion, which is not specified in this application.

[0029] Tasks u,m The perceived duration of u,m . Define binary variable o u,m Represents tasks u,m The unloading decision variables, and there are o u,m ∈{0,1}, where o u,m =1 indicates task s u,m Offload to edge server processing; otherwise, o u,m =0, indicating that the task is processed locally at terminal u. u,m For tasks u,m allocated bandwidth; The terminal u is s u,m The number of allocated computing resources q, correspondingly, Indicates that the edge server is s u,m The number of allocated computing resources q.

[0030] Embodiment 1

[0031] like Figure 1 As shown, this embodiment provides a perception communication computing joint optimization method in a mobile edge computing system, which specifically includes the following steps:

[0032] Step S110: Initialize the status information.

[0033] Before the system is run, the system parameters are initialized, the mobile terminal position is initialized, and the corresponding perception task duration, offloading decision and resource allocation are initialized.

[0034] The task duration perception and offloading decision are initialized randomly, and the resource allocation is initialized using an equal sharing strategy.

[0035] Step S120: After completing the initialization of the state information, the problem is reconstructed.

[0036] In each time slot, the computation task s of mode m of MT u u,m can be recorded as a four-tuple s u,m = u,m ,c u,m ,λ u,m ,τ u,m >, where b u,m is the amount of data collected; c u,m Indicates the corresponding amount of calculation, that is, the number of floating-point operations required; λ u,m =[λ u,m,1 ,…,λ u,m,Q ] is a Q-dimensional vector, which indicates the proportion of the computing load of the task on various computing resources. τ u,m ​is the maximum tolerable delay of the task. Specifically, task s u,m The amount of data b u,m It is expressed as:

[0037] b u,m =fs u,m b m δ u,m (1)

[0038] where δ u,m Represents the perception duration of MT u modal m data, which must meet the synchronization constraints of multimodal data acquisition b m Indicates the sampling rate; fs u,m Indicates the data bit depth of mode m.

[0039] The collected data of various modes usually rely on different processing algorithms, and the computational intensity of processing data of different modes is different. u,m The amount of computation required c u,m for:

[0040]

[0041] in, Indicates the computational intensity of mode m, that is, the number of floating-point operations required to process each bit of data.

[0042] Based on the above task generation process, the amount of data and computation of each task are dynamically determined by the perception duration and are related to the inherent properties of each modality.

[0043] Assuming that the channel state and the terminal position remain unchanged within a time slot, the data rate r of the m-mode data transmitted from MT u to ES is u,m for:

[0044]

[0045] Among them, B u,m For tasks u,m The allocated bandwidth must satisfy the constraints is the transmission power of MT u; |h u 2 represents the multipath fading coefficient; l u is the distance between MT u and ES, and [X u ,Y u ] is the current position coordinate of MT u; α represents the path loss index, σ 2 is the channel noise power.

[0046] Therefore, if task su,m Need to offload processing, corresponding communication delay for:

[0047]

[0048] The corresponding communication energy consumption of the mobile terminal It is expressed as:

[0049]

[0050] Mobile terminals and edge servers use their configured computing resources CR S u and CR S ES Provide corresponding computing support for the tasks to be processed. u,m Processing locally on the mobile terminal, the corresponding computing delay for:

[0051]

[0052] Where, the numerator represents the computational load of CR q; is MT u for s u,m The number of allocated CR qs must satisfy the constraints and s u,m The overall computing latency is the maximum value of the computing time of various computing resources. for:

[0053]

[0054] Among them, q is the computational energy efficiency of CR q, that is, the number of floating-point operations that can be performed per unit of energy (Watt).

[0055] If the task s u,m Offload to edge server processing and calculate latency for:

[0056]

[0057] Similarly, the numerator represents the computational load of CR q; Indicates ES is s u,m The number of allocated CR qs must satisfy the constraints and The corresponding computing energy consumption of the mobile terminal It is expressed as:

[0058]

[0059] in, It is the static power of MT Tu, which indicates the overhead of other basic functions of the terminal equipment such as maintaining the network, listening to signals, and staying powered even when not actively processing tasks.

[0060] o u,m Represents tasks u,m The unloading decision variables, and there are o u,m ∈{0,1}, where o u,m =1 indicates task s u,m Offload to ES for processing; otherwise, o u,m = 0. Therefore, based on the offloading decision result, the perception task s u,m The processing delay D u,m and the corresponding MT u energy consumption E u,m It is expressed as:

[0061]

[0062]

[0063] Among them, task data collection and processing need to be completed within one time slot: And the processing delay needs to meet the maximum tolerable delay constraint of the task and time alignment constraints for multimodal data processing

[0064] In order to simultaneously improve the accuracy of task execution and reduce completion delay, this embodiment defines a perception processing efficiency index χ u To express the trade-off between task performance and completion delay. Specifically, the function is defined as the ratio of the total amount of data collected by the mobile terminal u in each mode to its maximum task completion delay in a time slot, χ u , expressed as:

[0065]

[0066] In this embodiment, under the condition of satisfying the above constraints, the perception duration δ=[δ u,m ] U×M , uninstall decision o = [o u,m ] U×M , transmission bandwidth B = [B u,m ] U×M and local computing resource allocation on mobile terminals and computing resource allocation to edge servers (denoted as n={n u ,n ES}), maximizing the processing efficiency of the perception task while minimizing the terminal energy consumption:

[0067]

[0068] In the problem reconstruction, this embodiment divides the optimization process into two levels, corresponding to task generation and task processing, respectively. The multi-objective optimization problem (13) is reconstructed into a two-level planning problem, in which the upper-level problem is committed to maximizing the processing efficiency of the perception task, and this goal is constrained by the resource allocation decision of the lower level. Correspondingly, the lower-level resource allocation target value depends on the perception duration parameter set by the upper level. The two-level planning problem is expressed as:

[0069]

[0070] in, and Respectively represent the given perception duration, s u,m The task completion delay and energy consumption of MT u are functions of resource allocation variables. It is the allocation vector of Q heterogeneous computing resources of MT u or edge server for any u,m.

[0071] Step S130: After completing the problem reconstruction, optimize the upper layer perception duration.

[0072] In this embodiment, through a secondary transformation, the original form of the multi-proportional sum maximization problem is converted into a concave function maximization problem that can be solved by a classical convex optimization method, and it is iterated alternately with the update process of the introduced auxiliary variable y until the perception task processing efficiency stabilizes within a certain range (meeting the convergence condition), thereby outputting the current perception duration optimization result.

[0073] Based on the above, step S130 includes the following sub-steps:

[0074] Step S1301: taking the initialized system parameters, the initialized offloading decision and the resource allocation scheme as input, and initializing the upper layer iteration counter and auxiliary variables.

[0075] Step S1302: Obtain an equivalent problem using quadratic transformation and solve it using convex optimization.

[0076] Step S1303: Update auxiliary variables.

[0077] Among them, the auxiliary variable y u The update formula is:

[0078]

[0079] Step S1304: Determine whether a preset convergence condition is met or a maximum number of iterations is reached.

[0080] If the iteration does not meet the preset condition and is less than the maximum number of iterations, the number of iterations is increased by 1 and the process returns to step S1302.

[0081] If the iteration meets the preset condition or reaches the maximum number of iterations, step S1405 is executed.

[0082] Step S1305: output the current perception duration optimization result.

[0083] Step S140: After the optimization of the upper layer perception duration is completed, the lower layer unloading decision and resource allocation optimization are performed.

[0084] For each offloading decision combination, variable relaxation and branch-and-bound techniques are used to solve the terminal local computing resource allocation problem and the edge server computing resource allocation problem; in addition, the bandwidth allocation problem itself is a convex optimization problem, which can be solved using classical convex optimization methods. The offloading decision combination update and resource allocation based on linear integer programming are solved alternately until the terminal energy consumption stabilizes within a certain range (meets the convergence condition). The current offloading decision and resource allocation results are output.

[0085] Based on the above, step S140 includes the following sub-steps:

[0086] Step S1401: taking the initialization offloading decision and the perception duration optimization result as input, initializing the lower layer iteration counter.

[0087] Step S1402: Utilize variable relaxation, branch and bound, and convex optimization toolboxes to solve local computing resource allocation, edge server computing resource allocation, and bandwidth allocation under the current offloading decision combination.

[0088] Step S1403: using linear integer programming to solve an unloading decision plan based on current resource allocation.

[0089] Step S1404: Determine whether the preset convergence condition is reached or the maximum number of iterations is reached

[0090] If the iteration does not meet the preset condition and is less than the maximum number of iterations, the number of iterations is increased by 1 and the process returns to step S1402.

[0091] If the iteration meets the preset condition or reaches the maximum number of iterations, step S1405 is executed.

[0092] Step S1405: Output the current offloading decision and resource allocation result.

[0093] Step S150: Determine whether a preset convergence condition is reached or a maximum number of iterations is reached.

[0094] If the preset convergence condition is reached or the maximum number of iterations is reached, step S160 is executed; otherwise, the number of iterations is increased by 1 and the process returns to step S130.

[0095] If the number of system iterations is greater than the maximum allowed number, the iteration is terminated; otherwise, the next iteration process is entered after updating the parameters of each variable.

[0096] Step S160: Output the optimal resource allocation scheme and system bandwidth allocation scheme.

[0097] After an iteration is completed, the perception task processing efficiency and terminal energy consumption, perception duration, terminal local and edge server computing resource allocation plan, and system bandwidth allocation plan are updated. If the perception task processing efficiency and terminal energy consumption are stable within a certain range (meeting the convergence condition) in multiple consecutive iterations, the iteration is terminated and the best result is output.

[0098] The above method is deployed in the global control unit of the edge server, and the global control unit executes the above method. The mobile terminal in the system will upload its own resource configuration, computing requirements, current location and other information to the global control unit. After receiving the above information, the control unit will perform joint optimization of perception duration, unloading decision, and resource allocation, and broadcast the results of the system output to each terminal in the system to perform the corresponding multimodal data collection and processing perception tasks.

[0099] Embodiment 2

[0100] like Figure 2 As shown, this embodiment provides a perception communication computing joint optimization system in a mobile edge computing system, wherein the system is deployed in a global control unit of an edge server, thereby executing the perception communication computing joint optimization method in the mobile edge computing system in embodiment one.

[0101] The mobile terminals in the system will upload their own resource configuration, computing requirements, current location and other information to the global control unit. After receiving the above information, the control unit will perform joint optimization of perception duration, offloading decision and resource allocation according to the various units contained in the system in the control unit, and broadcast the system output results to each terminal in the system to perform the corresponding multimodal data collection and processing perception tasks.

[0102] The perception, communication and computing joint optimization system in the mobile edge computing system specifically includes: an initialization unit 210, a problem reconstruction unit 220, an upper-layer perception duration optimization unit 230, a lower-layer unloading decision and resource allocation optimization unit 240, a judgment unit 250 and an output unit 260.

[0103] The initialization unit 210 is used to initialize the state information.

[0104] Before the system is run, the system parameters are initialized, the mobile terminal position is initialized, and the corresponding perception task duration, offloading decision and resource allocation are initialized.

[0105] The task duration perception and offloading decision are initialized randomly, and the resource allocation is initialized using an equal sharing strategy.

[0106] The question reconstruction unit 220 is used to perform question reconstruction.

[0107] In each time slot, the computation task s of mode m of MT u u,m can be recorded as a four-tuple s u,m = u,m ,c u,m ,λ u,m ,τ u,m >, where b u,m is the amount of data collected; c u,m Indicates the corresponding amount of calculation, that is, the number of floating-point operations required; λ u,m =[λ u,m,1 ,…,λ u,m,Q ] is a Q-dimensional vector, which indicates the proportion of the computing load of the task on various computing resources. τ u,m is the maximum tolerable delay of the task. Specifically, task s u,m The amount of data b u,m It is expressed as:

[0108] b u,m =fs u,m b m δ u,m (1)

[0109] where δ u,m Represents the perception duration of MT u modal m data, which must meet the synchronization constraints of multimodal data acquisition b m Indicates the sampling rate; fs u,m Indicates the data bit depth of mode m.

[0110] The collected data of various modes usually rely on different processing algorithms, and the computational intensity of processing data of different modes is different. u,m The amount of computation required c u,m for:

[0111]

[0112] in, Indicates the computational intensity of mode m, that is, the number of floating-point operations required to process each bit of data.​

[0113] Based on the above task generation process, the amount of data and computation of each task are dynamically determined by the perception duration and are related to the inherent properties of each modality.

[0114] Assuming that the channel state and the terminal position remain unchanged within a time slot, the data rate r of the m-mode data transmitted from MT u to ES is u,m for:

[0115]

[0116] Among them, B u,m For tasks u,m The allocated bandwidth must satisfy the constraints is the transmit power of MTu; |h u 2 represents the multipath fading coefficient; l u is the distance between MT u and ES, and [X u ,Y u ] is the current position coordinate of MT u; α represents the path loss index, σ 2 is the channel noise power.

[0117] Therefore, if task s u,m Need to offload processing, corresponding communication delay for:

[0118]

[0119] The corresponding communication energy consumption of the mobile terminal It is expressed as:

[0120]

[0121] Mobile terminals and edge servers use their configured computing resources CR S u and CR S ES Provide corresponding computing support for the tasks to be processed. u,m Processing locally on the mobile terminal, the corresponding computing delay for:

[0122]

[0123] Where, the numerator represents the computational load of CR q; is MT u for s u,m The number of allocated CR qs must satisfy the constraints and s u,mThe overall computing latency is the maximum value of the computing time of various computing resources. for:

[0124]

[0125] Among them, q is the computational energy efficiency of CR q, that is, the number of floating-point operations that can be performed per unit of energy (Watt).

[0126] If the task s u,m Offload to edge server processing and calculate latency for:

[0127]

[0128] Similarly, the numerator represents the computational load of CR q; Indicates ES is s u,m The number of allocated CR qs must satisfy the constraints and The corresponding computing energy consumption of the mobile terminal It is expressed as:

[0129]

[0130] in, It is the static power of MT Tu, which indicates the overhead of other basic functions of the terminal equipment such as maintaining the network, listening to signals, and staying powered even when not actively processing tasks.

[0131] o u,m Represents tasks u,m The unloading decision variables, and there are o u,m ∈{0,1}, where o u,m =1 indicates task s u,m Offload to ES for processing; otherwise, o u,m = 0. Therefore, based on the offloading decision result, the perception task s u,m The processing delay D u,m and the corresponding MT u energy consumption E u,m It is expressed as:

[0132]

[0133] Among them, task data collection and processing need to be completed within one time slot: And the processing delay needs to meet the maximum tolerable delay constraint of the task and time alignment constraints for multimodal data processing

[0134] In order to simultaneously improve the accuracy of task execution and reduce completion delay, this embodiment defines a perception processing efficiency index χ u To express the trade-off between task performance and completion delay. Specifically, the function is defined as the ratio of the total amount of data collected by the mobile terminal u in each mode to its maximum task completion delay in a time slot, χ u , expressed as:

[0135]

[0136] In this embodiment, under the condition of satisfying the above constraints, the perception duration δ=[δ u,m ] U×M , uninstall decision o = [o u,m ] U×M , transmission bandwidth B = [B u,m ] U×M and local computing resource allocation on mobile terminals and computing resource allocation to edge servers (denoted as n={n u ,n ES}), maximizing the processing efficiency of the sensing task while minimizing the terminal energy consumption:

[0137]

[0138] In the problem reconstruction, this embodiment divides the optimization process into two levels, corresponding to task generation and task processing, respectively. The multi-objective optimization problem (13) is reconstructed into a two-level planning problem, in which the upper-level problem is committed to maximizing the processing efficiency of the perception task, and this goal is constrained by the lower-level resource allocation decision. Correspondingly, the lower-level resource allocation target value depends on the perception duration parameter set by the upper level. The two-level planning problem is expressed as:

[0139]

[0140] in, and Respectively represent the given perception duration, s u,m The task completion delay and energy consumption of MT u are functions of resource allocation variables. It is the allocation vector of Q heterogeneous computing resources of MT u or edge server for any u,m.

[0141] The upper layer perception duration optimization unit 230 is used to optimize the upper layer perception duration.

[0142] In this embodiment, through a secondary transformation, the original form of the multi-proportional sum maximization problem is converted into a concave function maximization problem that can be solved by a classical convex optimization method, and it is iterated alternately with the update process of the introduced auxiliary variable y until the perception task processing efficiency stabilizes within a certain range (meeting the convergence condition), thereby outputting the current perception duration optimization result.

[0143] Based on the above, the upper layer perception duration optimization unit 230 performs the following sub-steps:

[0144] Step T1: Take the initialized system parameters, initialized offloading decisions and resource allocation scheme as input to initialize the upper layer iteration counter and auxiliary variables.

[0145] Step T2: Use quadratic transformation to obtain an equivalent problem and solve it using convex optimization.

[0146] Step T3: Update auxiliary variables.

[0147] Among them, the auxiliary variable y u The update formula is:

[0148]

[0149] Step T4: Determine whether the preset convergence condition is reached or the maximum number of iterations is reached.

[0150] If the iteration does not meet the preset condition and is less than the maximum number of iterations, the number of iterations is increased by 1 and the process returns to step T2.

[0151] If the iteration reaches the preset condition or reaches the maximum number of iterations, step T5 is executed.

[0152] Step T5: Output the current perception duration optimization result.

[0153] The lower layer offloading decision and resource allocation optimization unit 240 is used to perform lower layer offloading decision and resource allocation optimization.

[0154] For each offloading decision combination, variable relaxation and branch-and-bound techniques are used to solve the terminal local computing resource allocation problem and the edge server computing resource allocation problem; in addition, the bandwidth allocation problem itself is a convex optimization problem, which can be solved using classical convex optimization methods. The offloading decision combination update and resource allocation based on linear integer programming are solved alternately until the terminal energy consumption stabilizes within a certain range (meets the convergence condition). The current offloading decision and resource allocation results are output.

[0155] Based on the above, the lower layer offloading decision and resource allocation optimization unit 240 performs the following sub-steps:

[0156] Step P1: Taking the initial offloading decision and the perceived duration optimization result as input, initialize the lower layer iteration counter.

[0157] Step P2: Utilize variable relaxation, branch and bound, and convex optimization toolboxes to solve the local computing resource allocation, edge server computing resource allocation, and bandwidth allocation under the current offloading decision combination.

[0158] Step P3: Use linear integer programming to solve the unloading decision plan based on the current resource allocation.

[0159] Step P4: Determine whether the preset convergence condition is reached or the maximum number of iterations is reached

[0160] If the iteration does not meet the preset condition and is less than the maximum number of iterations, the number of iterations is increased by 1 and the process returns to step P2.

[0161] If the iteration reaches the preset condition or reaches the maximum number of iterations, step P5 is executed.

[0162] Step P5: Output the current offloading decision and resource allocation results.

[0163] The judging unit 250 is used to judge whether a preset convergence condition is met or a maximum number of iterations is reached.

[0164] If the preset convergence condition is reached or the maximum number of iterations is reached, the output unit 260 is executed, otherwise the number of iterations is increased by 1 and the process returns to execute the upper layer perception duration optimization unit 230 .

[0165] If the number of system iterations is greater than the maximum allowed number, the iteration is terminated; otherwise, the next iteration process is entered after updating the parameters of each variable.

[0166] The output unit is used to output the best resource allocation scheme and the system bandwidth allocation scheme.

[0167] After an iteration is completed, the perception task processing efficiency and terminal energy consumption, perception duration, terminal local and edge server computing resource allocation plan, and system bandwidth allocation plan are updated. If the perception task processing efficiency and terminal energy consumption are stable within a certain range (meeting the convergence condition) in multiple consecutive iterations, the iteration is terminated and the best result is output.

[0168] This application has the following beneficial effects:

[0169] (1) This application fully explores the potential of mobile edge computing systems to provide computing support for perception tasks based on multimodal data, takes into account the requirements for time synchronization in the collection and processing of multimodal data, and ensures the effectiveness and accuracy of the use of perception information.

[0170] (2) This application fully explores the potential of mobile edge computing systems for heterogeneous computing resource management and the natural compatibility of heterogeneous computing resources with multimodal data processing, further increasing task processing efficiency and flexibility.

[0171] (3) This application takes into account the computational intensity, delay sensitivity and resource constraints of the perception task, and jointly improves the perception performance and saves terminal energy overhead.

[0172] Although the present application is described with reference to examples, this is for illustrative purposes only and is not intended to limit the present application, and changes, additions and / or deletions to the embodiments may be made without departing from the scope of the present application.

[0173] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A joint optimization method for perception, communication and computing in a mobile edge computing system, characterized in that: The following steps are involved: Initialize status information; After completing the initialization of the status information, reconstruct the problem; After completing the problem reconstruction, optimize the upper layer perception duration; After the optimization of the upper layer perception duration is completed, the lower layer offloading decision and resource allocation optimization are carried out; Determine whether the preset convergence condition is reached or the maximum number of iterations is reached; If the preset convergence condition is reached or the maximum number of iterations is reached, the optimal resource allocation plan and system bandwidth allocation plan are output; If the preset convergence condition is not met or the maximum number of iterations is reached, the optimization of the upper layer perception duration is returned to be performed.

2. The perception, communication and computing joint optimization method in the mobile edge computing system according to claim 1, characterized in that: Initializing the state information includes initializing system parameters, initializing the location of the mobile terminal and initializing the corresponding sensing task duration, offloading decision and resource allocation.

3. The perception, communication and computing joint optimization method in the mobile edge computing system according to claim 1, characterized in that: Before reconstructing the problem, it also includes determining the tasks u,m Data volume u,m , expressed as: b u,m =fs u,m b m δ u,m where δ u,m Represents the perception duration of the mobile terminal u modality m data, which must meet the synchronization constraints of multi-modal data collection b m Indicates the sampling rate; fs u,m Indicates the data bit depth of mode m.

4. The perception, communication and computing joint optimization method in the mobile edge computing system according to claim 3, characterized in that: The problem reconstruction includes converting the multi-objective optimization problem Restructured into a bilevel programming problem, where U = {1, ..., u, ..., U} represents U mobile terminals, M = {1, ..., M} represents the modal type of different data corresponding to each sensor, and task s u,m Data volume u,m , D u,m Indicates the mth mode of mobile terminal u in the corresponding task s u,m The processing delay, E u,m represents the energy consumption of the corresponding mobile terminal u, δ represents the perceived duration of the task, o represents the offloading decision variable of the task, B represents the bandwidth allocated to the task, and n represents the number of computing resources allocated to the task.

5. The perception, communication and computing joint optimization method in the mobile edge computing system according to claim 4, characterized in that: Perception tasks u,m The processing delay D u,m and the corresponding energy consumption E of mobile terminal u u,m It is expressed as: o u,m Represents tasks u,m The uninstall decision variables, Represents tasks u,m The communication delay, Represents tasks u,m The computational latency of offloading to edge servers, Represents tasks u,m The corresponding calculation delay is processed locally on the mobile terminal. Represents tasks u,m The communication energy consumption of the corresponding mobile terminal needs to be offloaded. Represents tasks u,m The computational energy consumption of mobile terminal u when processing locally on the mobile terminal.

6. A perception, communication and computing joint optimization system in a mobile edge computing system, characterized in that: Specifically include: Initialization unit, problem reconstruction unit, upper layer perception duration optimization unit, lower layer offloading decision and resource allocation optimization unit, judgment unit and output unit; An initialization unit, used for initializing state information; A problem reconstruction unit, used for problem reconstruction; An upper layer perception duration optimization unit, used for optimizing the upper layer perception duration; A lower-layer offloading decision and resource allocation optimization unit, used for making lower-layer offloading decisions and optimizing resource allocation; A judgment unit, used to judge whether a preset convergence condition is reached or a maximum number of iterations is reached; If the preset convergence condition is reached or the maximum number of iterations is reached, the output unit outputs the optimal resource allocation scheme and the system bandwidth allocation scheme; If the preset convergence condition is not met or the maximum number of iterations is reached, the process returns to the upper layer perception duration optimization unit for execution.

7. The perception, communication and computing joint optimization system in the mobile edge computing system according to claim 6, characterized in that: The initialization unit performs initialization of the state information including initialization of system parameters, initialization of the mobile terminal position and initialization of the corresponding sensing task duration, offloading decision and resource allocation.

8. The perception, communication and computing joint optimization system in the mobile edge computing system according to claim 7, characterized in that: Before the problem reconstruction unit conducts problem reconstruction, it also includes determining the task s u,m Data volume u,m , expressed as: b u,m =fs u,m b m δ u,m where δ u,m Represents the perception duration of the mobile terminal u modality m data, which must meet the synchronization constraints of multi-modal data collection b m Indicates the sampling rate; fs u,m Indicates the data bit depth of mode m.

9. The perception, communication and computing joint optimization system in the mobile edge computing system according to claim 8, characterized in that: The problem reconstruction unit performs problem reconstruction including converting the multi-objective optimization problem Restructured into a bilevel programming problem, where U = {1, ..., u, ..., U} represents U mobile terminals, M = {1, ..., M} represents the modal type of different data corresponding to each sensor, and task s u,m Data volume u,m , D u,m Indicates the mth mode of mobile terminal u in the corresponding task s u,m The processing delay, E u,m represents the energy consumption of the corresponding mobile terminal u, δ represents the perceived duration of the task, o represents the offloading decision variable of the task, B represents the bandwidth allocated to the task, and n represents the number of computing resources allocated to the task.

10. The perception, communication and computing joint optimization system in the mobile edge computing system according to claim 9, characterized in that: Problem reconstruction unit ends, perception task u,m The processing delay D u,m and the corresponding energy consumption E of mobile terminal u u,m It is expressed as: o u,m Represents tasks u,m The uninstall decision variables, Represents tasks u,m The communication delay, Represents tasks u,m The computational latency of offloading to edge servers, Represents tasks u,m The corresponding calculation delay is processed locally on the mobile terminal. Represents tasks u,m The communication energy consumption of the corresponding mobile terminal needs to be offloaded. Represents tasks u,m The computational energy consumption of mobile terminal u when processing locally on the mobile terminal.

Images