User association method based on online learning

Abstract

The application discloses a user association method based on online learning, comprising the following steps: defining and initializing various data of base stations and VR users; calculating and updating the estimation value of the confidence upper bound at time t according to the association times g m,u (t-1) of the base station m and the user u at time t-1; selecting the optimal compression ratio of the base station m for the user u at time t according to the optimal transmission time required by the base station m to the user u and the optimal GPU clock frequency allocated by the base station m to the user u; calculating the base station m associated with the user u at time t according to the calculation, and calculating the user association A(t); calculating and updating the data amount Q m (t+1) stored in the MEC server corresponding to the base station m at time t+1; updating the association times g m,u (t) of the base station m and the user u at time t; and iteratively updating according to the above steps to realize the dynamic user association based on online learning. The method realizes the balance between communication and calculation delay, reduces the calculation complexity of the algorithm, and has good convergence.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of wireless virtual reality, and particularly relates to a user association method based on online learning. BACKGROUND

[0002] Virtual Reality (VR) has attracted a lot of research attention from both academia and industry due to its ability to provide immersive experiences and support a wide range of applications, and has been considered as one of the key enablers for various applications in education, tourism, and gaming. However, traditional VR-centric research usually focuses on optical performance, ergonomics, and mathematical models, as the motivation for applications is usually process-oriented, and the challenges of communication and computation in VR are usually overlooked. With the rapid development of 5G or beyond 5G, unprecedented breakthroughs have been made in VR wireless technology. Immersive VR experiences based on wireless need to deliver massive data with ultra-low latency, and due to the need for ultra-high transmission rate, wireless bandwidth and computing bottlenecks are caused.

[0003] To solve this problem, researchers have made various attempts, among which the solution represented by Multi-access Edge Computing (MEC) has been widely studied. In the MEC system, computing capability and network control are executed at the network edge, providing computing resources for running computationally intensive applications on resource-limited mobile devices, thus improving the computing experience, reducing latency and energy consumption. In order to provide high-quality VR content, although MEC-enabled wireless VR systems have been studied from 4G to 6G, most of the existing research mainly focuses on energy consumption of MEC-based VR or only analyzes VR applications in a single MEC server scenario.

[0004] However, VR applications, such as VR games, are more inclined to delay than energy consumption. Interactions between multiple users can be considered real-time. The large field of view (FOV) frame formed after VR rendering poses a challenge to real-time VR applications. In addition, latency experience and energy consumption are important indicators of wireless VR services. In the multi-MEC service scenario, it is a good method to combine transmission delay and MEC service selection to achieve a balance between transmission delay and MEC service selection. Finally, due to the change of time due to the change of channel conditions and the change of available computing resources, VR users need to select appropriate MEC servers according to real-time communication and computing resources.

[0005] Therefore, how to perform online resource allocation to achieve a balance between latency and communication is the key to solving dynamic problems. SUMMARY

[0006] In order to solve the above problems existing in the prior art, the application provides a user association method based on online learning.

[0007] A user association method based on online learning comprises the following steps:

[0008] Step 1: defining and initializing various data of base stations and VR users, including setting loop variables of base stations m and users u, total time slots T and data volume Q(t) of each base station, wherein t is a time slot index and t is set to 1, and initializing the association times g M (-1) of base stations m and users u. m,u

[0009] Step 2: calculating and updating the estimation value of the confidence upper bound at the moment according to the association times g m,u (t-1) of base stations m and users u at the moment t-1.

[0010] Step 3: according to allocating the optimal compression ratio of base stations m for users u at the moment t the optimal transmission time required by base stations m to users u and the optimal GPU clock frequency allocated by base stations m to users u.

[0011] Step 4: according to calculating the base stations m(u, t) associated with users u at the moment t, and calculating the user association A(t) = a 1, (t), 1, (t), …, a M,U (t)}, wherein U is the total number of VR users, M is the total number of base stations, a m,u (t) represents that the VR user u is connected to the base station m.

[0012] Step 5: calculating and updating the data volume Q m (t+1) stored in the MEC server corresponding to the base station m at the moment t+1.

[0013] Step 6: updating the association times g m,u (t) of base stations m and users u at the moment t and setting t = t+1.

[0014] Step 7: iteratively updating according to the operations of steps 2-6 to realize dynamic user association based on online learning.

[0015] The application has the following beneficial effects:

[0016] ​The user association method based on online learning provided by the application has the advantages that: on one hand, a joint optimization strategy including user association, GPU cycle frequency, compression ratio and transmission time is designed; on the other hand, an efficient algorithm without prior knowledge is developed to solve the optimization variables, so that the balance between communication and calculation delay is realized, the calculation complexity of the algorithm is reduced, and good convergence is achieved.

[0017] The application will be further described in detail below with reference to the drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of the user association method based on online learning provided by the embodiment of the application;

[0019] Figure 2 is an application scenario diagram provided by the embodiment of the application;

[0020] Figure 3 is a convergence diagram of the compression ratio allocation algorithm provided by the embodiment of the application;

[0021] Figure 4 is a convergence diagram of the interactive algorithm of the compression ratio and transmission time provided by the embodiment of the application;

[0022] Figure 5 is a convergence trend diagram of the user association method based on online learning provided by the embodiment of the application;

[0023] Figure 6 is a diagram showing the relationship between the average delay and the minimum compression ratio C min provided by the embodiment of the application;

[0024] Figure 7 is a diagram showing the relationship between the average time delay of different algorithms and the number of MEC servers M provided by the embodiment of the application;

[0025] Figure 8 is a diagram showing the relationship between the average energy consumption of different algorithms and the number of VR users u provided by the embodiment of the application;

[0026] Figure 9 is a diagram showing the relationship between the average energy consumption of different algorithms and the number of VR users u under the maximum CPU cycle frequency of MEC servers provided by the embodiment of the application;

[0027] Figure 10 is a diagram showing the influence of different lengths of time slots τ on the average delay of different algorithms provided by the embodiment of the application. DETAILED DESCRIPTION

[0028] The application will be described in further detail below with reference to specific embodiments, but the embodiments of the application are not limited thereto.

[0029] Please refer to Figure 1 , Figure 1 is a flowchart of a user association method based on online learning provided by an embodiment of the application, which includes:

[0030] Step 1: defining and initializing various data of base stations and VR users.

[0031] The user association method based on online learning provided by the embodiment can be applied to the scenario shown in Figure 2 , wherein Figure 2 represents a wireless VR system supporting MEC, which is composed of U VR users and M base stations (BSs) equipped with MEC servers, that is, MEC servers and base stations appear in pairs, and one-to-one correspondence. Therefore, in the description of the application, the base station can also be understood as the corresponding server. In the system, each VR user is playing a VR interactive game.

[0032] Due to the limited computing resources and battery of VR devices, it is difficult to independently render high-resolution game frames under ultra-low delay. The most common method is to offload a large number of rendering tasks to MEC servers for execution, thereby reducing the computing burden of the device. In addition, other computing tasks after rendering are collectively referred to as post-processing, and these tasks can also be executed on the MEC server. The main purpose of post-processing is to make the rendering result more suitable for display on the VR device. Through the post-processing of the MEC server, the FoV frame transmitted to the user is larger than the original data FoV frame, which is easy to cause excessive pressure on wireless communication.

[0033] To solve this problem, the embodiment reduces communication overhead by compressing the transmitted FOV frame. It is assumed that the system operates in time slots, where each time slot length is set to τ; introduce respectively represent the set of VR users and BSs.

[0034] Specifically, set the cycle time t, t∈T, where T is the total time slot set, denoted as T={1,2,...,};

[0035] Set the loop variable VR user u (hereinafter referred to as user u), wherein is the user set and

[0036] Set the loop variable base station m (which can also be understood as server m), wherein is the MEC server (base station) set and

[0037] Set the base station data volume Q(t) = {Q1(t), Q2(t),..., Q M (t)} where t is the time slot index and set t = 1.

[0038] Initialize the number of times g m,u (-1) = 0 associated with the base station m and the VR user u.

[0039] Step 2: Calculate and update the estimate of the confidence upper bound at time t according to the number of times g m,u (t-1) associated with the base station m and the user u at time t-1.

[0040] In this embodiment, after the initialization is completed, the loop is entered.

[0041] When it is determined that the base station m and the user u are associated, that is, g m,u (t-1) > 0 is satisfied, calculate and update the estimate of the confidence upper bound according to the minimum CPU cycle frequency f min , the CPU frequency of the base station m available for processing the subsequent tasks of the user u at time t. The estimate of the post-processing time of the base station m to the VR user u The sample average of Calculate and update the estimate of the confidence upper bound

[0042] Specifically, The calculation formula is as follows:

[0043]

[0044] Where, is the estimate of the post-processing time of the base station m to the user u at time t-1 The sample average of, which can be calculated by the following formula:

[0045]

[0046]

[0047] When the base station m and the user u are not associated, that is, g m,u (t-1) > 0 is not satisfied, set

[0048] Step 3: According to Distribute the optimal compression ratio selected by the base station m for the user u at time t The optimal transmission time required by the base station m to the user u And the optimal GPU clock frequency allocated by the base station m for the user u

[0049] 31) Establish an optimization model to convert the optimization problem of into solving Φ(A(t)), whose model expression is:

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] where V represents the control parameter, τ is the time slot length, represents the rendering delay of user u served by base station m at time t, represents the post-processing calculation time of user u served by base station m at time t, represents the transmission time of base station m to user u at time t, Q m (t) represents the data amount currently saved by MEC server m at time t, represents the compression ratio set by MEC server m for VR user u at time t, F m,u (t) represents the FoV original image data size of VR user u on MEC server m at time t, k (t) represents the total energy consumption of game frame rendered by user u at time t, represents the maximum energy consumed by the base station to provide services for user u, represents the GPU clock frequency allocated by base station m for user u at time t, represents the maximum calculation speed of CPU in the base station, C min represents the minimum ratio between compressed data and original data size, d u (t) represents the total delay of game frame rendered by user u at time t, (C3) represents the energy constraint when the base station provides services for user u, (C4) represents the calculation speed constraint when presenting, (C5) is the data compression constraint, and (C6) is the delay constraint for user u to enjoy VR games;

[0056] 32) Solve the optimization model to obtain the optimal compression ratio selected by base station m for user u at time t the optimal transmission time required by base station m to user u

[0057] Specifically, the embodiment solves the optimization model according to the following steps:

[0058] Initialize GPU cycle frequency f r (o, t); initialize iteration index o = 0, set algorithm accuracy ι > 0, and allocate compression ratio C R (o, t) and transmission time d Tr (o, t);

[0059] According to C R (o-1, t) and d Tr (o-1, t), get f r (o, t);

[0060] According to f r (o, t) allocate compression ratio C R (o, t) and transmission time d Tr (o, t);

[0061] According to C R (o, t), d Tr (o, t), f r (o, t) solve the above equivalent optimization problem, and calculate Φ o (A(t));

[0062] When Φ o (A(t)) satisfies the preset condition |Φ o (A(t))-Φ o-1 (A(t))|≤ι, stop iteration o, and output the optimal and

[0063] Further, in the embodiment, the interactive algorithm according to the allocation of the compression ratio and the transmission time is allocated, specifically as follows:

[0064] a. Allocate transmission time:

[0065] Initialize q, q max , γ(t, 0), m(t, 0), and η;

[0066] Calculate the transmission time according to the following formula:

[0067]

[0068] Where W(x) represents the principal branch of the Lambert W function, which maintains ze z =x when z = W(x), and the subscript m(u) represents that the VR user u is associated to the base station m, q represents an iteration index, q maxdenotes the maximum number of iterations, η denotes the accuracy of the algorithm, m(t) = {m1(t), m2(t), …, m U U (t)} are the Lagrange multipliers of the constraint conditions C(3) and C(6) respectively, where U is the total number of VR users, h m,u (t) denotes the channel gain from base station m to user u at time t, B denotes the bandwidth allocated to the VR users, and N0 denotes the power spectral density of base station m.

[0069] b, the allocated compression ratio;

[0070] The updates γ, m are calculated according to the following formula:

[0071] m u (q+1) = m u (q) - x(q)Δμ u (q)

[0072] γ u (q+1) = γ u (q) - y(q)Δγ u (q)

[0073] where x(q) and y(q) denote the step size of the subgradient method, Δμ u (q) and Δγ u (q) denote the subgradient, whose expression is as follows:

[0074]

[0075]

[0076] where ε m(u), and χ m(u), are introduced variables, and denote the effective switched capacitance of the MEC server, denotes the CPU cycle frequency allocated by base station m to user m for data compression, θ denotes the ratio of the size of the FoV frame to the size of the corresponding FoV original image data, F m(u), (t) denotes the size of the FoV original image data of VR user u associated with MEC server m at time t, ∈ is a positive number, B denotes the bandwidth allocated to the user, and q denotes the iteration index, q max denotes the maximum number of iterations, is the exponential function of the compression ratio.

[0077] When the condition q ≤ q max ​or || μ(q+1) - μ(q) ||2 < η and || γ(q+1) - γ(q) ||2 < η, the algorithm ends;

[0078] Let

[0079] By the qth iteration of allocation, to achieve the allocation of compression ratio.

[0080] Specifically, the allocation of allocation includes:

[0081] Set C u,min = min , C u,max = 1;

[0082] Set iteration index l = 0, the maximum number of iterations l max , iteration accuracy is ζ;

[0083] When the condition l ≤ l max , iteration is performed on l, and the following calculations are performed:

[0084] S1: Let

[0085] S2: Calculate The formula is as follows:

[0086]

[0087] Where a1, a2 and a3 are variables, represented as:

[0088]

[0089]

[0090] a3 = Q m ( u )(t) θF m ( u ) ,u (t)

[0091] Indicates the effective switching capacitor of the base station, Indicates the CPU cycle frequency allocated by the base station m to the user u for data compression, and V is a control parameter;

[0092] S3: When the condition is met, then The algorithm terminates;

[0093] Otherwise, if , then

[0094] Otherwise, let Return to step S1 until l≤l max .

[0095] Step 4: According to Calculate the base station m(u, t) associated with user u at time t, and calculate the user association A(t).

[0096] Specifically, the calculation method of m(u, t) and A(t) is as follows:

[0097]

[0098]

[0099]

[0100] A(t)=a m,u (t))

[0101] Where, and are equivalent expressions of and in constraint conditions (C3) and (C6), is an introduced variable, represents the optimal rendering delay of user u served by base station m at time t, represents the optimal post-processing calculation time of user u served by base station m at time t, represents the optimal transmission time of base station m to user u at time t, Q m (t) represents the data amount stored in the MEC server corresponding to the base station m at time t, represents the optimal compression ratio selected by base station m for user u at time t, and θ represents the ratio of the size of the FoV frame to the size of the corresponding FoV original image data, f min represents the minimum CPU cycle frequency, represents the estimate of the upper confidence bound, and V is a control parameter, represents the number of CPUs required by MEC server m to process 1 bit of data.

[0102] Step 5: Calculate the updated data amount Q m (t+1) stored in the MEC server corresponding to the base station m at time t+1.

[0103] Specifically, the updated data amount Q m (t+1) stored in the MEC server corresponding to the base station m at time t+1 is updated as follows:

[0104]

[0105] wherein, and A m (t) respectively represent the rate of data arrival and departure of server m at time t.

[0106] Step 6: updating the association times of base station m and user u at time t m,u (t) and let t = t + 1.

[0107] Specifically, the calculation formula of the association times g m,u (t) of base station m and user u at time t is as follows:

[0108]

[0109] wherein, a m,u (t) is a user association variable, a m,u (t) = 1 indicates that user u obtains FoV from base station m at time t; and a m,u (t) = 0 indicates that user u does not obtain FoV from base station m at time t.

[0110] It should be noted that after updating g m,u (t), g m,u (t) and a m,u (n) are used to update g Specific formula can be referred to step 2.

[0111] Step 7: iteratively updating according to the operations of steps 2-6 to realize dynamic user association based on online learning.

[0112] The present application proposes a random optimization framework for MEC wireless virtual reality system. In the developed framework, we jointly optimize user association, rendering GPU cycle frequency, compression ratio and transmission time. In order to achieve effective resource allocation and minimum average latency, we develop an efficient algorithm without prior knowledge. At the same time, the optimization variables are decoupled, which reduces the computational complexity of the algorithm. And the algorithm has good convergence.

[0113] In addition, the joint optimization scheme proposed in the present application shows better performance in average latency than the prior art.

[0114] Embodiment two

[0115] The beneficial effects of the user association method based on online learning provided in the above embodiment one are further verified and explained by simulation test.

[0116] Figure 3 is a convergence diagram of the compression ratio allocation algorithm provided in the present embodiment. As shown in​Figure 3 In this experiment, two VR users (i.e. user 3 and user 8) were randomly selected to represent the convergence of the algorithm. From Figure 3 It can be seen that different users have different convergence trends. However, the final F(C R ) value converges to 0. At the same time, it can be seen that its convergence speed is very fast.

[0117] Figure 4 is the convergence diagram of the interactive algorithm of compression ratio and transmission time allocation provided by the embodiment of the application. It can be seen that it has a very fast convergence speed, which is about the 50th step.

[0118] Figure 5 is the convergence trend diagram of the user association method based on online learning provided by the embodiment of the application. It can be observed that it converges at 200 steps.

[0119] It can be seen from Figures 3-5 that the algorithm designed by the application has the ability to flexibly process the problem to be solved.

[0120] Figure 6 is the relationship diagram between the average delay and the minimum compression ratio C min provided by the embodiment of the application. From Figure 6 it can be seen that when the time slot duration τ = 1 and the CPU cycle frequency f c ∈ [0.1, 1], the relationship between the average delay and the minimum compression is: as the minimum compression ratio C min increases, the average delay always shows a trend of first increasing and then decreasing. Obviously, this phenomenon shows that it is necessary to balance the trade-off between compression and transmission to minimize the average delay.

[0121] Figure 7 is the relationship diagram between the average delay of different algorithms and the number of EMC servers M provided by the embodiment of the application. Among them, JAFDA, OAFO and OLYAP are existing algorithms, and Proposed is the algorithm proposed by the application. From Figure 7 it can be seen that the average delay decreases with the increase of the number of MEC servers M. At the same time, by arbitrarily specifying the number of MEC servers M, it can be seen that the performance of the scheme proposed by the application is the best among all the comparison schemes. Therefore, it can be concluded that the joint optimization of the application can effectively improve the performance of the system.

[0122] Figure 8is a schematic diagram of the relationship between the average energy consumption of different algorithms and the number of VR users u provided by the embodiment of the present application. For the VR system, the more users join, the more FoV frames the system processes, thereby causing the overall energy consumption of the system to increase accordingly. Therefore, for a certain number of MEC servers, the average energy consumption of the system increases with the increase of the number of VR users. As can be seen from the figure, the performance of the proposed scheme is the best.

[0123] Figure 9 is a schematic diagram of the relationship between the average energy consumption of different algorithms and the maximum CPU cycle frequency of MEC servers provided by the embodiment of the present application. The performance of the method of the present application under different maximum computing speeds is verified. Because the processing speed of different MEC servers can affect user association. Therefore, the verification is performed. With the increase of the maximum computing speed , the average delay of all schemes decreases. In addition, as can be seen from Figure 9 , when increases to a certain extent, the value of the average delay decreases very little, or even almost unchanged. It is indicated that because the system task amount is fixed, the improvement of the processing speed has little effect on the system performance.

[0124] Figure 10 demonstrates the effect of different algorithms using different lengths of time slots τ on the average delay. As can be seen from Figure 10 , the average delay becomes more significant with the increase of the time slot length. Therefore, selecting a suitable time slot length has a profound impact on the performance of the system.

[0125] The simulation results show that the method provided by the present application achieves the purpose of effective resource allocation and minimum average delay, and the algorithm has good convergence. The joint optimization scheme in the algorithm performs better in terms of average delay than the existing scheme.

[0126] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in the present specification.

[0127] Although the application has been described in connection with specific embodiments thereof, it will be understood that it is capable of further modifications and this application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains. It is intended to cover and embrace all adaptations or modifications of the application. The terminology used herein for the purpose of describing particular embodiments is not intended to be limiting of the application. This application is intended to cover all such modifications and alternatives within the scope of the application including combinations of aspects of the application with other items not described.

[0128] The above description is further detailed in connection with specific preferred embodiments of the application, and it is not to be construed that the specific implementation of the application is limited to these descriptions. For those skilled in the art, without departing from the concept of the application, a number of simple deductions or substitutions can be made, and all of them should be considered as falling within the protection scope of the application.

Claims

1. A user association method based on online learning, characterized in that: include: Step 1: Define and initialize the data of the base station and VR user, including setting the loop variables base station m and user u, total time slot T and the data volume of each base station Q(t) = {Q1(t),Q2(t),...,Q M (t)}, where t is the time slot index and t=1 is set, and the number of associations g between base station m and user u is initialized m,u (-1); Step 2: According to the number of associations g between base station m and user u at time t-1 m,u (t-1) Calculate and update the estimate of the confidence upper bound at time t Step 3: According to Assign the optimal compression ratio selected by base station m for user u at time t The optimal transmission time required from base station m to user u And the optimal GPU clock frequency assigned by base station m to user u Step 4: According to Calculate the base station m(u,t)* associated with user u at time t, and calculate the user association A(t) = {a 1,1 (t),a 1,2 (t),…,a M,U (t)}; where U is the total number of VR users, M is the total number of base stations, and a m,u (t) indicates that user u is connected to base station m; Step 5: Calculate and update the amount of data Q stored in the MEC server corresponding to base station m at time t+1 m (t+1); Step 6: Update the number of associations g between base station m and user u at time t m,u (t) and let t = t + 1; Step 7: Perform iterative updates according to the operations in steps 2-6 to achieve dynamic user association based on online learning.

2. The user association method based on online learning according to claim 1, characterized in that Step 2 includes: When the number of associations meets g m,u When (t-1)>0, calculate and update the upper confidence bound estimate according to the following formula in, is the estimated post-processing time of base station m for user u at time t-1 The sample mean of Otherwise, let 3. The user association method based on online learning according to claim 2, characterized in that: Step 3 includes: 31) Establish an optimization model to The optimization problem is transformed into solving Φ(A(t)), and its model expression is: Among them, V represents the control parameter, τ is the time slot length, represents the rendering delay of user u served by base station m at time t, represents the post-processing computation time of user u served by base station m at time t, represents the transmission time from base station m to user u at time t, F represents the compression ratio set by MEC server m for VR user u at time t, m,u (t) represents the size of the original FoV image data of VR user u on MEC server m at time t, e k (t) represents the total energy consumption of the game frame rendered by user u at time t, represents the maximum energy consumed by the base station to provide services to user u, represents the GPU clock frequency allocated by base station m to user u at time t, Indicates the maximum computing speed of the CPU in the base station, C min Indicates the minimum ratio between the size of compressed data and the original data, d u (t) represents the total delay of the game frame rendered by user u at time t, (C3) represents the energy constraint when the base station provides services to user u, (C4) represents the computational speed constraint during rendering, (C5) is the data compression constraint, and (C6) is the delay constraint for user u to enjoy the VR game; 32) Solve the optimization model to obtain the optimal compression ratio selected by base station m for user u at time t The optimal transmission time required from base station m to user u And the optimal GPU clock frequency assigned by base station m to user u 4. The user association method based on online learning according to claim 3, characterized in that In step 32), solving the optimization model includes: Initialize GPU cycle frequency f r (o, t); Initialize the iteration index o = 0, set the algorithm precision ι> 0, and assign the compression ratio C R (o, t) and transmission time d Tr (o, t); According to C R (o-1, t) and d Tr (o-1, t) gets f r (o,t); According to f r (o,t) distribution compression ratio C R (o, t) and transmission time d Tr (o, t); According to C R (o, t), d Tr (o, t), f r (o,t) Solve the above equivalent optimization problem and calculate the Φ of the oth iteration o (A(t)); When Φ o (A(t)) satisfies the preset condition |Φ o (A(t))-Φ o-1 When (A(t))|≤ι, stop iterating o and output the optimal as well as 5. The user association method based on online learning according to claim 4, characterized in that: The transmission time d is allocated as follows Tr (o, t) and compression ratio C R (o, t): a. Allocate transmission time: Initialize q, q max , γ (t, 0), μ (t, 0) and η; Calculate the transmission time as follows: Where W(x) represents the main branch of the LambertW function, which maintains ze when z = W(x) z = x, the subscript m(u) indicates that user u is associated with base station m, q indicates the iteration index, q max represents the maximum number of iterations, η represents the algorithm accuracy, μ(t)={μ1(t),μ2(t),…,μ U (t)} and γ(t)={γ1(t), γ2(t),…,γ U (t)} are the Lagrange multipliers of constraint C(3) and constraint C(6), U is the total number of VR users, h m,u (t) represents the channel gain from base station m to user u at time t, B represents the bandwidth allocated to VR user, and N0 represents the power spectral density of base station m; b. Allocate compression ratio; Calculate and update γ(t) and μ(t) according to the following formula: m u (t,q+1)=μ u (q)-x(q)Dm u (t,q) c u (t,q+1)=γ u (q)-y(q)Δγ u (t,q) Where x(q) and y(q) represent the step size of the subgradient method, Δμ u (t,q) and Δγ u (t,q) represents the subgradient, which is expressed as follows: Among them, ε m(u),u and χ m(u),u is the variable introduced, and Indicates the effective switching capacitance of the MEC server, represents the CPU cycle frequency allocated by base station m to user m for data compression, θ represents the ratio of the size of the FoV frame to the size of the corresponding FoV original image data, and F m(u),u (t) represents the size of the original FoV image data of the VR user u associated with the MEC server m at time t, ∈ is a positive constant, B represents the bandwidth allocated to the user, q represents the iteration index, q max represents the maximum number of iterations, is an exponential function of the compression ratio; When the condition q≤q is satisfied max Or when ‖μ(t,q+1)-μ(t,q)‖2<η and ‖γ(t,q+1)-γ(t,q)‖2<η, the algorithm ends; make By the qth iteration Distribution is performed to achieve distribution of compression ratio.

6. The user association method based on online learning according to claim 5, characterized in that: For the qth iteration Values ​​are assigned in the following ways, including: Setting C u,min =C min ,C u,max =1; Set the iteration index l = 0, the maximum number of iterations l max , the iteration accuracy is ζ; When the condition l≤l is satisfied max When , l is iterated and the following calculation is performed: S1: Order S2: Calculation The formula is as follows: Among them, a1, a2 and a3 are variables, expressed as: a3=Q m(u) (t)θF m(u),u (t) represents the effective switching capacitance of the base station, It represents the CPU cycle frequency allocated by base station m to user u for data compression, and V is a control parameter; S3: When the conditions are met When The algorithm terminates; Otherwise, if When Otherwise, let Return to step S1 until l≤l max .

7. The user association method based on online learning according to claim 6, characterized in that: In step 4, m(u,t)* and A(t) are calculated as follows: A(t)=(a m,u (t)) in, and Is the constraint (C3) and constraint (C6) and The equivalent expression of is the variable introduced, represents the optimal rendering delay of user u served by base station m at time t, represents the optimal post-processing computation time of user u served by base station m at time t, represents the optimal transmission time from base station m to user u at time t, Q m (t) represents the amount of data stored in the MEC server corresponding to base station m at time t, represents the optimal compression ratio selected by base station m for user u at time t, θ represents the ratio of the size of the FoV frame to the size of the corresponding FoV original image data, and f min Indicates the minimum CPU cycle frequency, represents the estimate of the confidence upper bound, V is the control parameter, Indicates the number of CPUs required by MEC server m to process 1 bit of data.

8. The user association method based on online learning according to claim 7, characterized in that: In step 5, the amount of data Q stored in the MEC server corresponding to base station m at time t+1 is m (t+1) The update method is as follows: in, and A m (t) denote the data arrival and departure rates of server m at time t, respectively.

9. The user association method based on online learning according to claim 1, characterized in that: In step 6, the number of associations g between base station m and user u at time t is m,u The calculation formula for (t) is: Among them, a m,u (t) is the user-associated variable, a m,u (t) = 1 means that user u obtains FoV from base station m at time t; a m,u (t)=0 indicates that user u does not obtain FoV from base station m at time t.

Images