Unloading strategy distributed optimization method under sympathetic calculation fusion network

By building a three-layer ISCC architecture and ADMM optimization method, the problem of excessive computing burden of terminal devices in ISAC technology is solved, efficient execution of perceptual tasks and reduced latency, and the computing resources of multiple MEC and MCC servers are used to improve perception Task execution efficiency.

CN118900432AInactive Publication Date: 2024-11-05BEIJING INST OF TECH

Patent Information

Application Number
CN202410822391.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-25
Publication Date
2024-11-05
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

ISAC technology has significant challenges in computing complexity and delay in the perception process, resulting in excessive computing burden on end devices with limited resources and inability to effectively process high-precision sensing signals.

Method used

Build a three-layer ISCC architecture consisting of one MCC server, multiple MEC servers and multiple ISAC terminals. By calculating the terminal-aware task execution delay and power consumption under different offload schemes, unload decision variables are introduced, and all terminal-aware tasks are proposed. The total execution delay is the optimization problem of the objective function. The ADMM method is iteratively calculated to obtain the unload decision variables, and the unloading scheme is optimized to reduce latency and power consumption.

Benefits of technology

It significantly reduces the average execution delay and algorithm running time of perceived tasks, makes full use of the computing resources of MEC servers and MCC servers, and improves the execution efficiency of perceived tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118900432A_ABST
    Figure CN118900432A_ABST
Patent Text Reader

Abstract

The invention provides an unloading strategy distributed optimization method under a communication sensing calculation fusion network, and belongs to the technical field of communication sensing calculation integration, and the method comprises the steps: an MCC server respectively calculates the execution time delay and terminal power consumption of a terminal sensing task under different unloading schemes; an unloading decision variable is introduced, and the total execution time of each terminal perception task and a terminal total power consumption expression are obtained; proposing an optimization problem taking the total execution time delay of the perception tasks of all terminals as a target function; performing iterative calculation by different MEC servers and MCC servers to obtain a relaxation solution of the unloading decision variable; a relaxation solution of the unloading decision variable is recovered by using a relaxation continuity and expansion-based method, and different terminals are informed of an unloading scheme by the MEC server; and when sensing the target, the terminal unloads the sensing data at the previous moment to the MEC server or the MCC server for processing by using the ISAC signal according to the received unloading scheme. According to the method, the average execution delay of the sensing task and the algorithm running time are remarkably reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of communication-awareness-computing integration, and specifically relates to a distributed optimization method for offloading strategies in a communication-awareness-computing fusion network. Background Art

[0002] The sixth generation (6G) wireless network will promote the development of many emerging applications such as drones, connected vehicles, digital twins, and augmented reality. These applications require wireless networks to quickly provide high-precision perception capabilities and high-quality communication performance. Integrated Sensing and Communication (ISAC) technology can simultaneously transmit data and wireless perception on the same spectrum, improving the efficiency of hardware and spectrum utilization, and becoming the key to achieving efficient and flexible operation of future networks.

[0003] Although ISAC technology has many advantages, it also faces significant challenges in terms of computational complexity and latency. A large number of echo signals are generated during the perception process, and some advanced perception signal processing methods, such as compressed sensing, spatial spectrum estimation, and continuous wavelet transform, are usually very complex and require a lot of computing resources. This imposes a serious computational burden on resource-limited ISAC terminals, resulting in non-negligible latency in processing perception signals. Therefore, for computationally intensive and latency-sensitive perception tasks, a more efficient execution scheme is urgently needed.

[0004] Mobile edge computing (MEC) technology provides a feasible and effective way to solve the terminal computing burden in ISAC technology, allowing the terminal to offload the perception task to the edge of the network (such as the base station) for execution, thereby effectively reducing the computing burden of the terminal device. A large amount of research work has been focused on using MEC technology to solve the computing problems in ISAC systems. Q. Liu et al. proposed a joint computing offloading and resource allocation strategy for vehicle-to-everything (V2X) networks assisted by Integrated Sensing, Communication and Computation (ISCC) technology, which improved the energy efficiency of completing the perception data fusion task while ensuring the execution delay requirements. N. Huang et al. studied the trajectory optimization and power allocation problems in ISCC-assisted UAV systems, with the aim of minimizing the energy consumption and perception data collection time of the UAV while meeting the radar sensing accuracy. However, most previous works only considered a two-layer ISCC architecture consisting of one MEC server, and did not make full use of multiple potential MEC servers and mobile cloud computing (MCC) servers with more computing resources. Summary of the invention

[0005] The purpose of this invention is to propose a distributed optimization method for offloading strategy in a synaesthesia-computing fusion network, which makes full use of the computing resources of MEC servers and MCC servers and significantly reduces the average execution delay of perception tasks and the algorithm running time.

[0006] The present invention is achieved through the following technical solutions:

[0007] A distributed optimization method for offloading strategy in a synaesthesia-computing fusion network includes the following steps:

[0008] Step S1: construct a three-layer ISCC architecture consisting of an MCC server, multiple MEC servers and multiple ISAC terminals. The MCC server combines the channel information of different terminals and the amount of perception task data to calculate the execution delay and terminal power consumption of the terminal perception task under different offloading schemes. The different offloading schemes include local execution mode, offloading to MEC server execution mode and offloading to MCC server execution mode.

[0009] Step S2: Based on the execution delay and terminal power consumption of the terminal perception task under different offloading schemes obtained in step S1, an offloading decision variable is introduced to represent the offloading scheme corresponding to the terminal perception task, and the total execution time of each terminal perception task and the total power consumption expression of the terminal are obtained;

[0010] Step S3: Based on the total execution time of each terminal perception task and the total power consumption expression of the terminal in step S2, considering the maximum power of the terminal, the computing power of the MEC server, and the constraints of the offloading decision variables, an optimization problem with the total execution delay of the perception tasks of all terminals as the objective function is proposed;

[0011] Step S4: Processing the optimization problem in step S3 based on variable relaxation and ADMM method, and obtaining the relaxation solution of the offloading decision variables by iterative calculation by different MEC servers and MCC servers;

[0012] Step S5: Use the relaxation continuation and expansion-based method to restore the relaxed solution of the offloading decision variable obtained in step S4 to a binary 0-1 variable, and the MEC server notifies the offloading plan to different terminals;

[0013] Step S6: While sensing the target, the terminal uses the ISAC signal to offload the sensing data of the previous moment to the MEC server or MCC server for processing according to the received offloading scheme.

[0014] Furthermore, the step S1 specifically includes:

[0015] Step S11: Set the number of terminals and MEC servers in the three-layer ISCC architecture respectively, and form corresponding sets respectively. The number of MEC servers is L, which is equipped in the base station and provides computing offloading services for the terminals in its coverage area. The set is represented as The number of terminals is K, which can be expressed as a set:

[0016] Step S12: respectively by formula and Calculate the execution delay of the perception task in the local execution mode respectively and terminal power consumption in, represents the CPU frequency of terminal k, β represents the computational complexity of the perception task, and Z k represents the data volume of the perception task of terminal k, w k represents the antenna gain of terminal k, and η represents the power coefficient related to the terminal CPU chip architecture;

[0017] Step S13: By formula and Calculate the execution delay of the perception task in the MEC server execution mode respectively and terminal power consumption Among them, R lk represents the uplink transmission rate from terminal k to MEC server l, which is calculated according to the formula Calculation, B represents the signal bandwidth, Hlk represents the uplink and downlink channel power gain from terminal k to MEC server l, p k represents the uplink transmission power of terminal k, σ 2 represents the noise power, represents the perception task offloading delay from terminal k to MEC server l, Indicates the perceived task computation delay on the MEC server. The CPU frequency allocated to terminal k by MEC server l;

[0018] Step S14: by formula and Calculate the execution delay of the perception task in the MCC server execution mode respectively and terminal power consumption in, is the transmission delay of the perception task from the MEC server to the MCC server, Calculate latency for perceived tasks on the MCC server. is the CPU frequency assigned to terminal k by the MCC server, r f It is the transmission rate between MEC server and MCC server.

[0019] Furthermore, the step S2 specifically includes:

[0020] Step S21: Introducing binary unloading decision variables and when a lk =0,b lk = 0, the sensing task of terminal k is performed locally at the terminal; when a lk =1,b lk = 0, the perception task of terminal k is offloaded to MEC server l for execution; when a lk =0,b lk =1, the sensing task of terminal k is offloaded to the MCC server for execution;

[0021] Step S22: Based on the binary offloading decision variables introduced and the execution delay and power consumption of the terminal k's perceived task under different offloading schemes obtained in step S1, the formula and The total execution time of the terminal k perception task is obtained respectively And the total power consumption expression of the terminal

[0022] Furthermore, the optimization problem in step S3 is expressed as in, is the constraint expression of the offloading decision variable, which constrains each terminal to only choose one offloading solution. is the MEC server computing capacity constraint, G l is the computing power of MEC server l, is the terminal power constraint, P th The maximum power that each terminal can use for sensing, communication, and computing.

[0023] Furthermore, the step S4 specifically includes:

[0024] Step S41: Unloading binary decision variables Relaxation is 0≤a lk ≤1,0≤b lk ≤1, The continuous variable shown in step S3 is transformed into the following The first optimization problem shown;

[0025] Step S42: Express the objective function of the first optimization problem in step S41 as a decomposition form for different MEC servers.

[0026] Step S43: Introduce ω lk =a lk and ω lk =b lk As the local offloading decision variable of MEC server 1, the first optimization problem in step S42 is converted into The second optimization problem is shown in which, for all and all ω l ,ω l and a lk ,b lk The feasible sets of and express,

[0027] Step S44: Based on the ADMM method, the augmented Lagrangian function representation of the second optimization problem in step S43 is obtained: By combining linear terms and quadratic terms, the augmented Lagrangian function is simplified to in, and is the Lagrange multiplier associated with the two consistency constraints of the second optimization problem, ρ is the penalty factor,

[0028] Step S45: Given and The feasible solution that satisfies the constraints is taken as the initial value. The superscripts it and it+1 are introduced to represent the it-th iteration and it+1-th iteration of the variable. The initial value of it is 0.

[0029] Step S46: Known and Each MEC server independently calculates its own sub-problem To update the optimal local unloading decision variables Use interior point method to solve subproblems;

[0030] Step S47: Known and MCC server computing third optimization problem To update the optimal global unloading decision variables Introducing Lagrange multipliers and The Lagrangian function of the third optimization problem is Based on the Carlo-Kuhn-Tucker condition, a lk and b lk The optimal closed-form solutions are

[0031]

[0032] Step S48: Known and MCC Server Computing To update the optimal Lagrange multiplier, based on the ADMM method and gradient descent method, By formula and Calculate, where v is the iteration step size;

[0033] Step S49: Based on the results from steps S46 to S48 Judgment|a (it +1) -a it | and |b (it+1) -b it |Is it less than the set convergence threshold ε0 or whether the current number of iterations it has reached the maximum number of iterations iter max If yes, then output the relaxed solution of the unloading decision variable and jump to step S5; if no, set it=it+1 and Input to step S46 to start a new iterative operation.

[0034] Furthermore, the step S5 specifically includes:

[0035] Step S51: Set the collection The initial value of k is 1;

[0036] Step S52: for a given k, according to and Find out what makes alk and b lk The largest indexes l1 and l2;

[0037] Step S53: Based on l1 and l2 obtained in step S52, and renew

[0038] Step S54: Determine the uninstall decision variable after the update Does it meet the constraints of the optimization problem in step S3? If so, let Go to step S55; if not, directly go to step S55;

[0039] Step S55: determine whether k is equal to K. If so, output the binary unloading decision variable The MEC server notifies different terminals of the offloading plan; if not, set k=k+1 and jump to step S52.

[0040] The present invention has the following beneficial effects:

[0041] 1. The present invention first constructs a three-layer ISCC architecture consisting of an MCC server, multiple MEC servers and multiple ISAC terminals. The MCC server calculates the execution delay and terminal power consumption of the terminal perception task under different offloading schemes respectively. Then, the offloading decision variable is introduced to define the offloading scheme corresponding to the terminal perception task, and the total execution time of each terminal perception task and the total terminal power consumption expression are obtained. Secondly, an optimization problem with the total execution delay of the perception tasks of all terminals as the objective function is proposed. Then, the relaxation solution of the offloading decision variable is calculated, and the relaxation solution is restored to a binary 0-1 variable. The MEC server notifies the offloading scheme to different terminals. While perceiving the target, the terminal uses the ISAC signal to offload the perception data of the previous moment to the MEC server or MCC server for processing according to the received offloading scheme. In this way, the computing resources of the MEC server and the MCC server are fully utilized, thereby significantly reducing the average execution delay of the perception task. In addition, the present invention uses a distributed algorithm based on ADMM to optimize the offloading scheme, which can significantly reduce the algorithm running time. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present invention is further described in detail below with reference to the accompanying drawings.

[0043] Figure 1 It is a flow chart of the present invention.

[0044] Figure 2 The present invention is a flow chart of processing the optimization problem based on variable relaxation and distributed ADMM method to obtain the relaxation solution of the unloading decision variables.

[0045] Figure 3 The present invention is a flowchart of restoring the unloading decision variables to binary 0-1 variables using a relaxed continuation and an expansion-based method.

[0046] Figure 4 This is a simulation result diagram comparing the average execution delay of the perception task of the present invention and the comparison method as the signal bandwidth changes.

[0047] Figure 5 This is a simulation result diagram comparing the average CPU running time of the present invention and the comparison method with the change of the number of terminals. DETAILED DESCRIPTION

[0048] like Figures 1 to 3 As shown in FIG. 1 , the distributed optimization method of the offloading strategy under the synaesthesia fusion network includes the following steps:

[0049] Step S1: construct a three-layer ISCC architecture consisting of an MCC server, multiple MEC servers and multiple ISAC terminals. The MCC server combines the channel information of different terminals and the amount of perception task data to calculate the execution delay and terminal power consumption of the terminal perception task under different offloading schemes. The different offloading schemes include local execution mode, offloading to MEC server execution mode and offloading to MCC server execution mode.

[0050] The specific steps include:

[0051] Step S11: Set the number of terminals and MEC servers in the three-layer ISCC architecture respectively, and form corresponding sets respectively. The number of MEC servers is L=3, which is represented by the set: They are located at (0m, 0m), (300m, 0m) and (150m, 150m) respectively, receiving ISAC signals from the terminals and performing sensing tasks or forwarding sensing tasks to the MCC server for execution. The number of ISAC user terminals is K = 9, which can be represented by the set Randomly distributed in a 1km*1km two-dimensional area, receiving echo signals to perform radar detection on the perceived target, while performing the perception task or using the preset ISAC waveform to offload the perception task to the MEC server or MCC server. The perceived target of each terminal is randomly located, and its direction and distance are uniformly distributed on [0°, 180°] and [30m, 70m] respectively. All terminals operate in the same frequency band, and all communication channels follow the Rayleigh distribution;

[0052] Step S12: Calculate the execution delay of the sensing task in the local execution mode by using formula (1) and formula (2) respectively. and terminal power consumption

[0053]

[0054] in, represents the CPU frequency of terminal k (unit: CPU cycle / s), β represents the computational complexity of the perception task (unit: CPU cycle / bit), and Z k represents the data volume of the perception task of terminal k, w k represents the antenna gain of terminal k, and η represents the power coefficient related to the terminal CPU chip architecture;

[0055] Specifically in this embodiment, the CPU frequency f of terminal k i L is 0.5G CPU cycles / s, the CPU cycle β required to process 1 bit of data is 200 CPU cycles / bit, and the data volume Z of the terminal k perception task k is 5MB, and the power factor η related to the terminal CPU chip architecture is 10 -28 ;

[0056] Step S13: Calculate the execution delay of the sensing task in the MEC server execution mode by using formula (4) and formula (5) respectively: and terminal power consumption

[0057]

[0058] Among them, R lk represents the uplink transmission rate from terminal k to MEC server l, which is calculated according to formula (3):

[0059]

[0060] Where B represents the signal bandwidth, H lk represents the uplink and downlink channel power gain from terminal k to MEC server l, p k represents the uplink transmission power of terminal k, σ 2 represents the noise power, represents the perception task offloading delay from terminal k to MEC server l, Indicates the perceived task computation delay on the MEC server. The CPU frequency allocated to terminal k by MEC server l;

[0061] Specifically in this embodiment, the signal bandwidth B is 10 MHz, and the noise power σ 2 is -174dBm / Hz, and the CPU frequency allocated to terminal k by MEC server l is 3G CPU cycles / s;

[0062] Step S14: Calculate the execution delay of the sensing task in the MCC server execution mode by using formula (6) and formula (7) respectively: and terminal power consumption

[0063]

[0064] in, is the transmission delay of the perception task from the MEC server to the MCC server, Calculate latency for perceived tasks on the MCC server. is the CPU frequency assigned to terminal k by the MCC server, r f is the transmission rate between the MEC server and the MCC server;

[0065] Specifically in this embodiment, the computing resources allocated by the MCC server to terminal k The transmission rate between MEC server and MCC server is 10G CPU cycle / s. f It is 10Mbit / s.

[0066] So far, the calculation of the execution delay of the perception task and the terminal power consumption under different offloading schemes has been completed.

[0067] Step S2: Based on the execution delay and terminal power consumption of the terminal perception task under different offloading schemes obtained in step S1, an offloading decision variable is introduced to represent the offloading scheme corresponding to the terminal perception task, and the total execution time of each terminal perception task and the total power consumption expression of the terminal are obtained;

[0068] The specific steps include:

[0069] Step S21: Introducing binary unloading decision variables and Represents the perceived task offloading scheme: when a lk =0,b lk = 0, the sensing task of terminal k is performed locally at the terminal; when a lk =1,b lk = 0, the perception task of terminal k is offloaded to MEC server l for execution; when a lk =0,b lk =1, the sensing task of terminal k is offloaded to the MCC server for execution;

[0070] Step S22: According to the binary offloading decision variables introduced, and the execution delay and terminal power consumption of the terminal k perception task under different offloading schemes obtained in step S1, the total execution time of the terminal k perception task is obtained by formula (8) and formula (9) respectively: And the total power consumption expression of the terminal

[0071]

[0072] At this point, the derivation of the total execution time of each terminal perception task and the total power consumption expression of the terminal has been completed.

[0073] Step S3: Based on the total execution time of each terminal perception task and the total power consumption expression of the terminal in step S2, considering the maximum power of the terminal, the computing power of the MEC server, and the constraints of the offloading decision variables, an optimization problem with the total execution delay of the perception tasks of all terminals as the objective function is proposed;

[0074] The specific steps include:

[0075] Step S31: Set the maximum power that each terminal can use for sensing, communication and calculation to P th , we get the terminal power constraint expression (10):

[0076]

[0077] Specifically in this embodiment, the maximum power P of the terminal th 30dBm;

[0078] Step S32: Set the computing capacity of MEC server 1 to G l , we get the MEC server computing capacity constraint expression (11):

[0079]

[0080] Specifically in this embodiment, the computing capacity G of MEC server 1 l 9G CPU cycles / s;

[0081] Step S33: Each terminal can only select one uninstallation strategy, and the uninstallation decision variable constraint expression (12) is obtained:

[0082]

[0083] Step S34: According to the three constraint expressions in steps S31 to S33, an optimization problem with the total execution delay of the sensing tasks of all terminals as the objective function is obtained, as shown in formula (13):

[0084]

[0085] At this point, the construction of the optimization problem with the total execution delay of the perception tasks of all terminals as the objective function has been completed.

[0086] Step S4: Processing the optimization problem in step S3 based on variable relaxation and ADMM (alternating direction method of multipliers), and obtaining a relaxation solution of the offloading decision variable by iterative calculation by different MEC servers and MCC servers;

[0087] The specific steps include:

[0088] Step S41: Unloading binary decision variables The relaxation is a continuous variable as shown in formula (14):

[0089]

[0090] Then the optimization problem in step S3 is transformed into the first optimization problem as shown in formula (15):

[0091]

[0092] Step S42: Express the objective function of the first optimization problem in step S41 as a decomposition form for different MEC servers as shown in formula (16):

[0093]

[0094] Step S43: Introduce ω lk =a lk and As the local offloading decision variable of MEC server l, for all and all ω l , and a lk ,b lk The feasible sets of and express, and They are given by formula (17) and formula (18) respectively:

[0095]

[0096] Then the first optimization problem in step S42 is converted into the second optimization problem as shown in formula (19):

[0097]

[0098] Step S44: Based on the ADMM method, the augmented Lagrangian function of the second optimization problem in step S43 is obtained, as expressed in formula (20):

[0099]

[0100] By combining the linear term and the quadratic term, the augmented Lagrangian function is simplified as shown in formula (21):

[0101]

[0102] in, and is the Lagrange multiplier associated with the two consistency constraints of the second optimization problem, ρ is the penalty factor,

[0103] Step S45: Given and The feasible solution that satisfies the constraints is taken as the initial value. The superscripts it and it+1 are introduced to represent the it-th iteration and it+1-th iteration of the variable. The initial value of it is 0.

[0104] Step S46: Known and Each MEC server independently calculates its own sub-problem to update the optimal local offloading decision variables This sub-problem is shown in formula (22):

[0105]

[0106] This subproblem is a convex problem, and the interior point method is used to solve the subproblem;

[0107] Step S47: Known and The MCC server calculates the third optimization problem to update the optimal global offloading decision variables The third optimization problem is shown in formula (23):

[0108]

[0109] Introducing Lagrange multipliers and The Lagrangian function of the third optimization problem is given by formula (24):

[0110]

[0111] Based on the Karush-Kuhn-Tucker (KKT) condition, a lk and b lk The optimal closed-form solutions of are given by formula (25) and formula (26) respectively:

[0112]

[0113] Step S48: Known and The MCC server calculates formula (27) to update the optimal Lagrange multiplier

[0114]

[0115] Based on the ADMM method and gradient descent method, Calculated by formula (28) and formula (29):

[0116]

[0117] Where v is the iteration step size;

[0118] Step S49: Based on the results from steps S46 to S48 Judgment|a (it +1) -a it | and |b (it+1) -b it |Is it less than the set convergence threshold ε0 or whether the current number of iterations it has reached the maximum number of iterations iter max If so, then output the relaxed solution of the unloading decision variable (i.e. a lk and b lk The optimal closed-form solution of ), jump to step S5; if not, let it = it + 1, and Input to step S46 to start a new iterative operation.

[0119] Specifically in this embodiment, the convergence threshold ε0 is 0.01, and the maximum number of iterations iter max is 20.

[0120] So far, the relaxed solution of the unloading decision variables of this embodiment has been obtained.

[0121] Step S5: Use the relaxation continuation and expansion-based method to restore the relaxed solution of the offloading decision variable obtained in step S4 to a binary 0-1 variable, and the MEC server notifies the offloading plan to different terminals;

[0122] The specific steps include:

[0123] Step S51: Set the collection The initial value of k is 1;

[0124] Step S52: for a given k, find the solution that makes a lk and b lk The largest indexes l1 and l2:

[0125]

[0126] Step S53: Based on l1 and l2 obtained in step S52, update according to formula (32) and formula (33):

[0127]

[0128] Step S54: Determine the uninstall decision variable after the update Does it meet the constraints of the optimization problem in step S3? If so, let Go to step S55; if not, directly go to step S55;

[0129] Step S55: determine whether k is equal to K. If so, output the binary unloading decision variable The MEC server notifies different terminals of the offloading plan; if not, set k=k+1 and jump to step S52.

[0130] At this point, the recovery of binary offloading decision variables is completed, and the offloading plan for each terminal is obtained.

[0131] Step S6: While sensing the target, the terminal uses the ISAC signal to offload the sensing data of the previous moment to the MEC server or MCC server for processing according to the received offloading scheme.

[0132] Figure 4 In the figure, the horizontal axis is the signal bandwidth B, which takes values ​​of 5MHz, 10MHz, 15MHz, 20MHz, 25MHz, 30MHz, 35MHz, 40MHz, and 45MHz respectively; the vertical axis is the average execution delay of the perception task. The simulation experiment compared and analyzed three perception task execution schemes: 1) Local execution method: The perception data collected by each terminal is processed locally; 2) Edge-Terminal (ET) ISCC method: Considering the two-layer ISCC architecture, the terminal chooses to offload the sensing task to the MEC server or local processing; 3) Distributed Cloud-Edge-Terminal (DCET) ISCC method, that is, the present invention. Figure 4 It can be seen that as the signal bandwidth increases, the average execution delay also increases. The present invention can obtain a lower perception task execution delay than the other two schemes.

[0133] Figure 5The horizontal axis is the number of terminals k, which takes values ​​of 3, 6, 9, 12, 15, and 18 respectively; the vertical axis is the average CPU running time of the two perception task execution schemes on MATLAB R2022. The simulation experiment compared and analyzed the two perception task execution schemes: 1) Centralized Cloud-Edge-Terminal (CCET) ISCC method: All channel state information is collected by the MCC server, the MCC server executes the global offloading strategy, and then distributes the results to the base station and the terminal; 2) DCET-ISCC method: that is, the present invention. Figure 5 It can be seen that the CPU running time of the present invention is less than that of the CCET-ISCC method, and the benefit increases significantly with the increase of the number of terminals. Therefore, the present invention has a lower computational complexity.

[0134] The above description is only a preferred embodiment of the present invention, and therefore cannot be used to limit the scope of implementation of the present invention. That is, equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the specification should still fall within the scope covered by the patent of the present invention.

Claims

1. A distributed optimization method for offloading strategy in a synaesthesia-computing fusion network. Features: The steps include: Step S1: construct a three-layer ISCC architecture consisting of an MCC server, multiple MEC servers and multiple ISAC terminals. The MCC server combines the channel information of different terminals and the amount of perception task data to calculate the execution delay and terminal power consumption of the terminal perception task under different offloading schemes. The different offloading schemes include local execution mode, offloading to MEC server execution mode and offloading to MCC server execution mode. Step S2: Based on the execution delay and terminal power consumption of the terminal perception task under different offloading schemes obtained in step S1, an offloading decision variable is introduced to represent the offloading scheme corresponding to the terminal perception task, and the total execution time of each terminal perception task and the total power consumption expression of the terminal are obtained; Step S3: Based on the total execution time of each terminal perception task and the total power consumption expression of the terminal in step S2, considering the maximum power of the terminal, the computing power of the MEC server, and the constraints of the offloading decision variables, an optimization problem with the total execution delay of the perception tasks of all terminals as the objective function is proposed; Step S4: Processing the optimization problem in step S3 based on variable relaxation and ADMM method, and obtaining the relaxation solution of the offloading decision variables by iterative calculation by different MEC servers and MCC servers; Step S5: Use the relaxation continuation and expansion-based method to restore the relaxed solution of the offloading decision variable obtained in step S4 to a binary 0-1 variable, and the MEC server notifies the offloading plan to different terminals; Step S6: While sensing the target, the terminal uses the ISAC signal to offload the sensing data of the previous moment to the MEC server or MCC server for processing according to the received offloading scheme.

2. According to the distributed optimization method of offloading strategy in the synaesthesia-computing fusion network of claim 1, Features: The step S1 specifically includes: Step S11: Set the number of terminals and MEC servers in the three-layer ISCC architecture respectively, and form corresponding sets respectively. The number of MEC servers is L, which is equipped in the base station and provides computing offloading services for the terminals in its coverage area. The set is represented as The number of terminals is K, which can be expressed as a set: Step S12: respectively by formula and Calculate the execution delay of the perception task in the local execution mode respectively and terminal power consumption in, represents the CPU frequency of terminal k, β represents the computational complexity of the perception task, and Z k represents the data volume of the perception task of terminal k, w k represents the antenna gain of terminal k, and η represents the power coefficient related to the terminal CPU chip architecture; Step S13: By formula and Calculate the execution delay of the perception task in the MEC server execution mode respectively and terminal power consumption Among them, R lk represents the uplink transmission rate from terminal k to MEC server l, which is calculated according to the formula Calculation, B represents the signal bandwidth, H lk represents the uplink and downlink channel power gain from terminal k to MEC server l, p k represents the uplink transmission power of terminal k, σ 2 represents the noise power, represents the perception task offloading delay from terminal k to MEC server l, Indicates the perceived task computation delay on the MEC server. The CPU frequency allocated to terminal k by MEC server l; Step S14: by formula and Calculate the execution delay of the perception task in the MCC server execution mode respectively and terminal power consumption in, is the transmission delay of the perception task from the MEC server to the MCC server, Calculate latency for perceived tasks on the MCC server. is the CPU frequency assigned to terminal k by the MCC server, r f It is the transmission rate between MEC server and MCC server.

3. According to claim 2, a distributed optimization method for offloading strategy in a synaesthesia-computing fusion network, Features: The step S2 specifically includes: Step S21: Introducing binary unloading decision variables and when a lk =0,b lk = 0, the sensing task of terminal k is performed locally at the terminal; when a lk =1,b lk = 0, the perception task of terminal k is offloaded to MEC server l for execution; when a lk =0,b lk =1, the sensing task of terminal k is offloaded to the MCC server for execution; Step S22: Based on the binary offloading decision variables introduced and the execution delay and power consumption of the terminal k's perceived task under different offloading schemes obtained in step S1, the formula and The total execution time of the terminal k perception task is obtained respectively And the total power consumption expression of the terminal 4. According to claim 3, a distributed optimization method for offloading strategy in a synaesthesia-computing fusion network, Features: The optimization problem in step S3 is expressed as in, is the constraint expression of the offloading decision variable, which constrains each terminal to only choose one offloading solution. is the MEC server computing capacity constraint, G l is the computing power of MEC server l, is the terminal power constraint, P th The maximum power that each terminal can use for sensing, communication, and computing.

5. According to claim 4, a distributed optimization method for offloading strategy in a synaesthesia-computing fusion network, Features: The step S4 specifically includes: Step S41: Unloading binary decision variables Relaxation The continuous variable shown in step S3 is transformed into the following The first optimization problem shown; Step S42: Express the objective function of the first optimization problem in step S41 as a decomposition form for different MEC servers. Step S43: Introduce ω lk =a lk and As the local offloading decision variable of MEC server 1, the first optimization problem in step S42 is converted into The second optimization problem is shown in which, for all and all and a lk ,b lk The feasible sets of and express, Step S44: Based on the ADMM method, the augmented Lagrangian function representation of the second optimization problem in step S43 is obtained: By combining linear terms and quadratic terms, the augmented Lagrangian function is simplified to in, and is the Lagrange multiplier associated with the two consistency constraints of the second optimization problem, ρ is the penalty factor, Step S45: Given and The feasible solution that satisfies the constraints is taken as the initial value. The superscripts it and it+1 are introduced to represent the it-th iteration and it+1-th iteration of the variable. The initial value of it is 0. Step S46: Known and Each MEC server independently computes its own sub-problem To update the optimal local unloading decision variables Use interior point method to solve subproblems; Step S47: Known and MCC server computing third optimization problem To update the optimal global unloading decision variables Introducing Lagrange multipliers and The Lagrangian function of the third optimization problem is Based on the Carlo-Kuhn-Tucker condition, a lk and b lk The optimal closed-form solutions are Step S48: Known and MCC Server Computing To update the optimal Lagrange multiplier, based on the ADMM method and gradient descent method, By formula and Calculate, where v is the iteration step size; Step S49: Based on the results from steps S46 to S48 Judgment|a (it+1) -a it | and |b (it+1) -b it |Is it less than the set convergence threshold ε 0 Or whether the current number of iterations it has reached the maximum number of iterations iter max If yes, then output the relaxed solution of the unloading decision variable and jump to step S5; if no, set it=it+1 and Input to step S46 to start a new iterative operation.

6. According to claim 5, a distributed optimization method for offloading strategy in a synaesthesia-computing fusion network, Features: The step S5 specifically includes: Step S51: Set the collection The initial value of k is 1; Step S52: for a given k, according to and Find out what makes a lk and b lk The largest index l 1 and l 2 ; Step S53: based on the l obtained in step S52 1 and l 2 ,according to and renew Step S54: Determine the uninstall decision variable after the update Does it meet the constraints of the optimization problem in step S3? If so, let Go to step S55; if not, directly go to step S55; Step S55: determine whether k is equal to K. If so, output the binary unloading decision variable The MEC server notifies different terminals of the offloading plan; if not, set k=k+1 and jump to step S52.

Citation Information

Patent Citations

  • Distributed unloading method under small cell integrated mobile edge computing

    CN109121151A

  • Time delay minimization method under cooperative unloading mechanism in mobile edge computing

    CN111556143A

  • Joint optimization method and system in cloud side-end collaborative system, and storage medium

    CN116761218A

Cited By

  • Resource allocation method and system in communication, inductance and calculation integrated Internet of Vehicles system

    CN119562356A