Cellular-free uplink communication system edge task unloading method based on rate division multiple access assisted by intelligent reflecting surface

Abstract

The invention discloses an edge task unloading method of a cellular-free uplink communication system based on rate division multiple access assisted by an intelligent reflecting surface, and relates to a wireless communication and mobile edge computing (MEC) technology. The method aims at optimizing calculation and communication resource allocation of the MEC system, reducing task unloading time delay and improving system throughput. By introducing the RIS technology, the wireless channel environment is dynamically adjusted, the signal propagation quality is enhanced, meanwhile, the RSMA technology is combined to optimize data transmission, the spectrum utilization rate is improved, and the multi-user interference is reduced. A block coordinate descent (BCD) optimization framework is adopted, a UE-AP matching algorithm based on channel state information (CSI) and computing resource availability is provided, an efficient user access strategy is achieved, an RIS phase shift matrix, AP beam forming, a task unloading factor and a power distribution strategy are jointly optimized, task processing delay is minimized, and the user access efficiency is improved. And the computing resource utilization rate and communication quality are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication; in particular, it relates to a method for offloading edge tasks in a cell-free uplink communication system based on intelligent reflecting surface-assisted rate-splitting multiple access. Background Art

[0002] With the rapid development of wireless communication technology, the sixth-generation mobile communication network is required to support ultra-reliable and ultra-low latency services. However, traditional cellular communication systems face challenges such as spectrum resource shortage and interference management. Inter-cell interference can lead to poor service quality for users at the cell edge, thereby affecting the performance of the entire system. The cell-free (CF) massive multiple-input multiple-output (mMIMO) system, as a distributed and networked technology, can well solve the above problems. In this architecture, multiple randomly deployed AP points share the same resource block and jointly serve a small number of users, greatly improving the performance of edge users.

[0003] However, in multi-user cell-free uplink communication, interference management among users and efficient utilization of spectrum resources are still key issues to be urgently solved. Traditional multiple access technologies cannot fully utilize spectrum resources in the face of high-density users and complex environments, resulting in a decline in communication performance. To support joint transmission of multiple services, the rate-splitting multiple access (RSMA) technology is considered a very promising implementation method. RSMA can partially decode interference and treat part of the interference as noise. With this ability, RSMA has higher spectrum and energy efficiency than space-division multiple access (SDMA) or non-orthogonal multiple access (NOMA) in multi-user deployments. The intelligent reflecting surface (RIS), as a key technology for 6G, has the advantages of low cost, low complexity, and easy deployment. By dynamically adjusting the phase and amplitude of the reflected signal, it can significantly enhance the propagation path of wireless signals, thereby improving communication quality and system capacity. The research on combining RIS and RSMA technologies for cell-free communication systems can, while maintaining good performance, alleviate the system's demand for CSI.

[0004] In a CF-mMIMO network, due to the distributed deployment of wireless AP points, each access point node can be regarded as a small server. Computational tasks can be assigned to each node through mobile edge computing (MEC), thereby extending the data processing ability from the cloud to the edge of the network. When terminal devices execute computationally intensive tasks, the workload can be offloaded from devices with scarce resources and energy to computing nodes with sufficient resources and energy, thereby increasing the available resources and energy of mobile devices. To fully leverage the computational advantages of MEC, an efficient multi-user uplink transmission scheme needs to be designed to achieve simultaneous offloading of massive data. Therefore, based on a RIS-assisted cell-free uplink RSMA-MEC system, the present invention proposes an edge task offloading method for a cell-free uplink communication system based on intelligent reflecting surface-assisted rate-splitting multiple access. Summary of the Invention

[0005] To solve the above problems, the present invention discloses an edge task offloading method for a cell-free uplink communication system based on intelligent reflecting surface-assisted rate-splitting multiple access, which jointly optimizes beamforming design and infinite resource scheduling to achieve dynamic allocation of communication and computing resources and ensure minimization of task offloading delay. This method takes into account the physical constraints and complexity requirements in the actual system, ensuring the practicality and feasibility of the solution.

[0006] An edge task offloading method for a cell-free uplink communication system based on intelligent reflecting surface-assisted rate-splitting multiple access includes the following steps:

[0007] Step S1: First, establish a RIS-assisted cell-free uplink RSMA-MEC system model, including a communication model and a task offloading model, which contains multiple distributed access points (APs), intelligent reflecting surfaces (RISs), and multiple user equipments (UEs). Construct a UE-AP matching factor matrix and design a matching strategy based on channel state information (CSI) and AP computing resource availability, enabling UEs to adaptively select the optimal AP for communication.

[0008] Step S2: Adopt rate-splitting multiple access (RSMA) technology to split the uplink transmission task into multiple sub-messages at the user side and allocate different transmission powers. At the AP side, use successive interference cancellation (SIC) technology for decoding. By optimizing power allocation and decoding order, improve the achievable rate of sub-messages. Introduce RIS to dynamically optimize passive beamforming, adjust the channel environment, improve signal quality, reduce interference, and optimize the communication performance of edge task offloading.

[0009] Step S3: Establish a task offloading model, where each UE needs to execute computational tasks. Some tasks are computed locally, and some tasks are offloaded through the uplink to the MEC server on the AP side with which it communicates for computing.

[0010] Step S4, establish an optimization problem, and jointly optimize the UE-AP matching, task offloading factor, RIS phase shift matrix, AP receive beamforming, and decoding order to minimize the task processing delay, improve the system throughput and energy efficiency;

[0011] Step S5, adopt the block coordinate descent (BCD) optimization framework for solution, alternately optimize each variable, so that the system iteratively converges to the optimal solution, and improve the utilization rate of computing and communication resources.

[0012] The beneficial effects of the present invention are as follows:

[0013] 1. Combining the dynamic optimization of passive beamforming by RIS and the efficient signal processing ability of RSMA, it improves the spectral efficiency and communication reliability of the system, reduces multi-user interference, and enhances the task scheduling ability of edge computing.

[0014] 2. The proposed optimization method takes into account the computing resource allocation and the change of wireless channel state, enables the system to adaptively adjust under different load conditions, and thus effectively supports the application requirements of large-scale Internet of Things (IoT) and intelligent wireless networks. Description of the Drawings

[0015] Figure 1 is the scene structure diagram of the embodiment of the present invention.

[0016] Figure 2 is the time allocation diagram of task offloading and computing process. Detailed Embodiment

[0017] The present invention will be further clarified below in conjunction with the drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. It should be noted that the words "front", "rear", "left", "right", "up" and "down" used in the following description refer to the directions in the drawings, and the words "inside" and "outside" refer to the directions towards or away from the geometric center of a specific component respectively.

[0018] This embodiment takes Figure 1 as the research scenario. This scenario consists of three parts: multiple APs, RIS, and multiple users. It adopts a cell-free network architecture. By introducing RIS technology, the reflection path of the signal is dynamically adjusted to enhance the communication link quality. Combining with RSMA technology, by splitting the user transmission signal and performing multi-user joint decoding, the spectral efficiency and robustness of the system are significantly improved. On this basis, this embodiment proposes an edge task offloading method for a cell-free uplink communication system based on intelligent reflecting surface-assisted rate splitting multiple access, and the specific steps are as follows:

[0019] Step S1: First, establish a RIS-assisted cell-free uplink RSMA-MEC system model, including a communication model and a task offloading model, which contains multiple distributed access points (APs), intelligent reflecting surfaces (RISs), and multiple user equipments (UEs). Construct a UE-AP matching factor matrix, and design a matching strategy based on channel state information (CSI) and AP computing resource availability, enabling UEs to adaptively select the optimal AP for communication;

[0020] Step S2: Adopt rate-splitting multiple access (RSMA) technology to split the uplink transmission task into multiple sub-messages at the user side and allocate different transmission powers. At the AP side, use successive interference cancellation (SIC) technology for decoding. By optimizing power allocation and decoding order, improve the achievable rate of sub-messages. Introduce RIS to dynamically optimize passive beamforming, adjust the channel environment, improve signal quality, reduce interference, and optimize the communication performance of edge task offloading;

[0021] Step S3: Establish a task offloading model, where each UE needs to execute computing tasks. Part of the tasks are computed locally, and part of the tasks are offloaded to the MEC server on the AP side for computing through the uplink.

[0022] Step S4: Establish an optimization problem, jointly optimize UE-AP matching, task offloading factor, RIS phase shift matrix, AP receive beamforming, and decoding order to minimize the task processing delay and improve system throughput and energy efficiency;

[0023] Step S5: Adopt a block coordinate descent (BCD) optimization framework for solving, alternately optimize each variable, making the system iteratively converge to the optimal solution, and improving the utilization rate of computing and communication resources.

[0024] First, in Step S1, establish a system model, considering a CF-mMIMO network architecture consisting of M AP points and K single-antenna user equipments (UEs), and each AP point is equipped with N r receive antennas. The central processing unit (CPU) is connected to all APs through a reliable backhaul link to exchange network information and conduct unified coordination. Our research focus is on the uplink (UL), where each user adopts RSMA technology for access and adaptively selects several optimal AP points to establish communication. Define to represent the set of APs, to represent the set of users, to represent the matching factor matrix between users and AP points, and d m,k ∈{0,1}. If d m,k =1, it indicates that the k-th user successfully matches with the m-th AP and establishes a data transmission link; otherwise, d m,k =0. Denote the set of AP points selected by the k-th user. Denote the set of users served by the m-th AP point. In the UL RSMA system, user k splits the transmitted message into J sub-messages, allocates different powers, and after superposition coding, transmits them to the AP point within the same time slot and frequency slot. The transmitted signal can be expressed as: Where, Denote the set of each user's sub-message, s k,j Denote the j-th sub-message of user k, and satisfy p k,j Denote the transmit power allocated to the sub-message s k,j . The total transmit power of user equipment k Is constrained by the maximum transmit power P max , that is

[0025] In the complex urban environment, the direct link is easily blocked by obstacles such as buildings and trees. To improve the communication quality, we deploy a passive RIS with N reflection units between the AP and the user, and bypass the obstacles by establishing an indirect link of UE-RIS-AP. Let Be the phase shift matrix of the RIS, where The reflection coefficient of the n-th unit of the RIS is given by , and the reflection phase satisfies φ n ∈[0, 2π). The signal received by the m-th AP point can be expressed as: Where x k Denote the message sent from the UE to the AP, n m Denote the additive white Gaussian noise, and the mean is zero and the variance is σ 2 , that is Is the receive beamforming vector at the AP. Denote the channel gain of the direct link between the k-th user and the m-th AP point. Denote the channel gain between the RIS and the m-th AP point, Denote the channel gain between the k-th user and the RIS, H R,m Φh k,R Denote the indirect LOS link of UE-RIS-AP. For simplicity of representation, define Denote the equivalent channel from the k-th user to the m-th AP point. Assume that in this scenario, the AP point and the RIS have complete knowledge of the CSI of all the involved channels.

[0026] Step S2, at the receiving end, the AP point adopts the successive interference cancellation (SIC) technique and decodes the total Strip message. Assume that the m-th AP decodes the sub-message s k,j in the order denoted by and the set

[0027] represents the decoding order of all sub-messages received at the m-th AP. Ω m represents the set of all possible decoding orders of sub-messages at the m-th AP. Then the set ζ m belongs to the large set Ω m . Each time an AP decodes a sub-message, it uses the SIC technique to eliminate it from the entire data stream and then continues to decode the remaining content. Therefore, for the sub-message s k,j , all sub-messages with decoding orders earlier than it have been eliminated, and the remaining undecoded sub-messages continue to exist as interference. The achievable rate of decoding the sub-message s k,j is:

[0028]

[0029] where B is the channel bandwidth, represents the set of all sub-messages with decoding orders greater than s k,j at the m-th AP, that is At the m-th AP, the message from user is transmitted through J streams. Therefore, the achievable rate of the entire message is:

[0030] Step S3, establish a task offloading model. In the above UL RSMA system, the CPU responsible for managing all APs has cloud computing capabilities. An independent MEC server is deployed on the side of each AP. The physical locations of the AP and its MEC server are almost the same and are connected by high-speed optical fibers. Therefore, the communication delay between the two can be ignored. Assume that each user has a divisible compute-intensive task to process. Due to the limited computing power of the user equipment itself, the data processing time is relatively long. To meet the requirements of 6G low latency, users offload part of their computing tasks to the corresponding MEC servers for processing through the uplink communication link. In this paper, we assume that the user equipment can perform offloading and local computing simultaneously, and the MEC server starts computing immediately after the offloading is completed, that is, there is no queue delay on the MEC server.

[0031] As Figure 2 shown, the entire computing cycle can be divided into four time periods.

[0032] The first time period: 0 - T 1 The time period determines the RSMA transmit power allocation factor and the task offloading factor. Then, the second time period: at T 1 ~T 2During this period, the user processes some tasks locally, and divides the remaining tasks into multiple subtasks and offloads the data to the MEC server deployed by the matching AP via the uplink; Third period: T 2 ~T 3 In this period, each MEC server executes edge computing tasks.

[0033] In the second and third periods, that is, T 1 ~T 3 , the user equipment synchronously processes local computing tasks.

[0034] Fourth period: T 3 ~T 4 During this time period, the task calculation results are downloaded to the user side.

[0035] In the above four periods, in the first stage, the CPU determines the allocation factor and performs unified scheduling. Since the CPU has very powerful computing capabilities, the algorithm processing delay can be ignored. In the fourth stage, the amount of data of the processed calculation results is usually small, and the download delay can also be ignored. Therefore, for user k, the delay of the entire process can be approximated as T 1 ~T 3 , and we denote this period as t k , that is: where represents the delay of user k for local computing, represents the maximum delay for offloading the task to the MEC server and performing calculations. When user k allocates tasks to multiple servers, we use to represent the total delay consumed by user k for offloading and calculating tasks to the m-th MEC server, represent the task transmission delay and task calculation delay respectively, then there is For user k, its computing task contains a total of D k bits of data. Define the task offloading factor η k =(η k,0 ,η k,1 ,…,η k,M ),η k,m ∈[0,1], where η k,0 represents the proportion of the data volume calculated locally by user k, and η k,1 ,…,η k,M respectively represent the proportion of the subtask data volume offloaded by user k to the m-th MEC server. If then no communication link is established between this AP point and user k, and data cannot be transmitted either. At this time, there must be η k,m =0. Obviously, in order to ensure that the computing tasks can be completed in full, the task offloading factor should satisfy the relational expression During the task offloading phase, the amount of data offloaded by user k to the m-th MEC server it is connected to is η k,m D k bits. Assuming the data can be sent at the maximum transmission rate, the time taken to offload this part of the task is: Let represent the local computing power of the k-th user and the computing power of the m-th MEC server respectively, with the unit of cycles per second. C k represents the number of cycles required to compute one unit bit. Then the latencies of local computing and edge computing can be expressed as: The time for all K users in the system to complete the task is the task processing latency of the entire RSMA-MEC system, denoted as

[0036] Step S4, establish an optimization problem, jointly optimize UE-AP matching, user-side sub-message power allocation and task offloading factor, RIS phase shift matrix, AP-side receive beamforming and decoding order, so as to minimize the task processing latency of the RIS-assisted cell-free uplink RSMA-MEC system. The optimization problem can be expressed as:

[0037]

[0038] where C1 represents the constraint on the user's transmission power; C2 and C3 respectively represent the constant modulus constraints of RIS passive beamforming and AP-side receive beamforming; C4 represents selecting the sub-message decoding order from all possible decoding order sets; C5 represents the value range of the elements of the UE-AP matching matrix; C6 represents that the user rate needs to meet the minimum QoS requirement; the non-negative exponent α k,j represents the splitting ratio of the sub-message, and Constraint C7 represents the fairness constraint of the user's sub-message, making the sub-message rate satisfy C8 and C9 represent the constraints on the task offloading factor, and the sub-task allocation should ensure that the tasks of each user are completely processed.

[0039] Step S5, solve it based on the optimization framework of the block coordinate descent (BCD) algorithm. First, propose a UE-AP matching algorithm based on CSI and computing resource availability, then propose a decoding order design method based on the UE-AP channel gain and the user sub-message splitting ratio, then use the successive convex optimization algorithm to optimize the user sub-message power allocation, and finally design an alternating Riemannian manifold (ARM) algorithm to optimize the active and passive beamforming.

[0040] P1 is a min - max optimization problem, and the piecewise form of \(t\) makes the problem difficult to solve directly. To linearize the optimization problem, we introduce an auxiliary variable \(T\), which is not less than the longest task processing time of all users in the system.

[0041]

[0042] s.t. \(C1,C2,C3,C4,C5,C6,C7,C8,C9\)

[0043]

[0044] The discrete variable \(d\) m,k is a 0 - 1 problem, and there is a high coupling between the variables \(\varPhi\) and \(v\). Therefore, we need to decouple the original problem into multiple sub - problems and use the block coordinate descent (BCD) method for alternating optimization.

[0045] In the RIS - assisted cell - free RSMA - MEC system, reasonable UE - AP matching is crucial for optimizing system performance. In the complex environment of multi - user and multi - AP described in this paper. Due to the differences in the channel responses between each user and different APs, different UE - AP matching strategies will have different impacts on system performance, throughput, and task transmission delay. In addition, although the computing power of the MEC server on the AP side is much higher than that of the user's local, if too many users are simultaneously connected and offload tasks to it, it is easy to cause data congestion and queuing delay, which will have an adverse impact on the system task processing delay. Therefore, when performing UE - AP matching, it is necessary to comprehensively consider two factors, channel gain and computing resources, and make a reasonable access.

[0046] When solving the UE - AP matching problem, if the exhaustive search method is used, although the optimal matching can be found, a large amount of redundant calculation will lead to a sharp increase in complexity, low efficiency, and it is not applicable to large - scale systems; while the fixed allocation scheme lacks flexibility and cannot change dynamically according to the current channel state and computing resources, and its performance is poor. Therefore, in this subsection, a user - centered greedy algorithm based on CSI and computing resource availability is proposed: in each round of matching, give priority to allocating the AP with the best channel quality and sufficient computing resources for the user to access, ensuring the best throughput and the lowest task processing delay under the current channel state. Let denote the effective channel gain between the \(k\) - th user and the \(m\) - th AP point, be the matrix composed of the channel gains between all UEs and APs. The specific matching algorithm is shown in Algorithm 1.

[0047]

[0048]

[0049] Under the given UE-AP matching matrix D, user-side sub-message power allocation p, RIS phase shift matrix Φ, AP-side receive beamforming v, and sub-message decoding order ζ, the connection relationship between each user and its access point in the system is known at this time, and the data transmission rate between them is also determined. At this time, the optimization goal of the task offloading factor is to minimize the task processing delay of the entire system, ensure a balance between the local computing delay of the user and the edge computing delay of offloading subtasks to each AP point, so as to shorten the total task completion time as much as possible, that is: Substitute the calculation formula in step S3 into the above formula, and the relationship between the task offloading factors can be obtained: Combining constraint C6, C7 with the above formula, the optimal offloading factor can be solved The corresponding optimal can be obtained according to the proportional relationship. At this time, since the local computing delay, the delay of offloading and computing subtasks are equal, the total task processing delay of user k can be sorted out as:

[0050]

[0051] Under the condition of fixing the UE-AP matching matrix D, task offloading factor η, RIS phase shift matrix Φ, AP-side receive beamforming v, and decoding order ζ, the optimization problem P2 can be rewritten as:

[0052]

[0053] where C10 is obtained by substituting the sorted t in formula (2) k into it. Problem P3.1 is difficult to solve due to the non-convexity of constraints C6 and C10. Therefore, we need to convert the non-convex constraints into convex constraints. First, using the proportional constraint C7, C6 can be equivalent to:

[0054]

[0055] Further converted to:

[0056]

[0057] The converted form is equivalent to the difference between two linear functions, so it is a convex constraint.

[0058] Constraint C10 can be equivalent to:

[0059]

[0060] where, Obviously Γ kis a variable that is only affected by T. Let Γ min satisfies Then, the optimization problem P3.1 can be equivalently transformed into:

[0061]

[0062] Introduce slack variables and The constraint C10 can be equivalently transformed into the following three constraints:

[0063]

[0064] (8) and (9) are still non-convex. We first deal with (24) and rewrite it in the form of convex difference:

[0065]

[0066] where is the introduced auxiliary variable. We use the Successive Convex Approximation (SCA) method to deal with (10). The principle of the SCA algorithm is mainly to use the value of the previous iteration to approximate the non-convex function into an equivalent convex function form. Perform the first-order Taylor expansion on the left side of (10):

[0067]

[0068] The r in the above formula represents the number of iterations.

[0069] The constraint (9) can use the fairness ratio constraint C7 again and be equivalently transformed into:

[0070]

[0071] where

[0072] represents the set of all sub-messages whose decoding order at the m-th AP is greater than or equal to s k,j That is Furthermore, the left side of the above formula can be rewritten in the form of convex difference, that is:

[0073]

[0074] The left side of the above formula is the logarithm of a concave affine function, while the right side is non-convex. We use the SCA method to transform the non-convexity of the constraint into convexity. Let and p k,j (r) represent the values obtained in the r-th iteration respectively. Then, at the r-th iteration, can be approximated by the first-order Taylor expansion formula as:

[0075]

[0076] Thus, the constraint can be approximately written in a convex form:

[0077]

[0078] At this time, the optimization problem P3.2 is transformed into problem P3.3:

[0079]

[0080] Thus, we can use the standard CVX toolbox to solve it, and the specific process is shown in Algorithm 2.

[0081]

[0082]

[0083] With the UE-AP matching matrix D, task offloading factor η, user sub-message power allocation p, and decoding order ζ fixed, the active and passive beamforming optimization problem can be formulated as:

[0084]

[0085] Due to the unit modulus constraints C2, C3 and non-convex constraints C5, C9, the optimization problem P4 is difficult to solve. To address the constant modulus constraint, we adopt the idea of alternating optimization and propose an Alternating Riemannian Manifold (ARM) algorithm based on the Riemannian Conjugate Gradient Descent (RCGD) algorithm to obtain a local optimal solution. We introduce an auxiliary function to handle C5, C9. For convenience of representation, define: where are respectively defined as:

[0086] μ 1 and μ 2 are two penalty factors, which are non-negative values.

[0087] By projecting the solution space onto a smooth Riemannian complex manifold, the constraints C2 and C3 can be represented as two different manifolds: The Riemannian gradient is obtained by projecting the Euclidean gradient onto the tangent vector space. Let the tangent vector spaces and respectively represent the sets of all tangent vectors tangent to the manifold at the point θ and tangent to the manifold at v. The Riemannian gradient g of the function θ at the point θ is calculated by the orthogonal projection of with respect to θ onto the Euclidean gradient. Obviously The Euclidean gradients can be respectively expressed as:

[0088]

[0089] Wherein, denotes taking the derivatives with respect to θ and v respectively. For the sake of convenient representation, let Then the above two derivatives are respectively:

[0090]

[0091] Wherein, the symbol ⊙ represents the Hadamard product of matrices. denotes taking the derivatives with respect to θ and v respectively, and let According to the chain rule, we can obtain:

[0092]

[0093]

[0094] Using the projection operator P x (X) to represent the orthogonal projection of the point x from the Riemannian space to the tangent space, which is defined as: P x (X) = X - Re(X ⊙ x) ⊙ x. The Riemannian gradient of the function at the point θ is projected onto the tangent vector space and can be expressed as:

[0095]

[0096] Similarly, the Riemannian gradient of the function at the point v can be expressed as:

[0097]

[0098] Next, based on the conjugate gradient method, the local optimal points on the manifold are respectively obtained in the iteration. Let the search direction at the b-th iteration. To ensure superlinear convergence, the Polak-Ribiere conjugate gradient descent direction is adopted, and λ θ (b + 1), λ v (b + 1) are the Polak-Ribiere parameters and can be calculated by the following formula:

[0099]

[0100] Then the search direction update at the (b + 1)-th iteration is the Polak-Ribiere conjugate gradient descent direction:

[0101] d θ (b + 1)= - grad θ (b + 1)+λ θ (b + 1)P θ(b+1) (d θ (b)) (26)

[0102] d v (b + 1)= - grad v (b + 1)+λ v (b + 1)P v(b+1) (d v (b)) (27)

[0103] The retraction method is used to project the tangent vector back into the recovery circle space as the next iteration point. Define the retraction operator Ret (·) (·). Then, θ(b + 1) and v(b + 1) after the (b + 1)-th iteration update are respectively:

[0104]

[0105]

[0106] Among them, γ θ , γ v represent the search step sizes and can be obtained based on the Armijo search algorithm. The solution process of the active and passive beamforming ARM optimization algorithm is shown in Algorithm 3.

[0107]

[0108] An appropriate decoding order can effectively eliminate the interference generated by the sub-messages of other users and improve the overall performance of the system. Selecting the optimal decoding order is a difficult mixed integer non-linear programming problem. Therefore, we propose a decoding order based on the UE-AP channel gain and the sub-message segmentation ratio of users.

[0109] Assume that two adjacent sub-messages s i and s j received at the AP are sent with powers p i and p j respectively, p i , p j ∈(0, p max , and the channel gains are G i , G j respectively. The upper limit of the interference of the sub-messages with decoding orders less than s i and s j is κ. Then, |G i | 2 p i+|G j | 2 p j +n+σ 2 ≤κ, where n represents the interference generated by sub - messages whose decoding order is higher than s i and s j If s i is decoded first, the signal - to - noise ratio needs to satisfy: From the above formula, the lower bounds of their powers can be deduced as Combined with the power constraint conditions, the corresponding upper bounds of power can be determined as: Let represent the sum of the signal strengths of sub - messages s i and s j , then there is:

[0110]

[0111] And if s j is decoded first, similarly, it can be deduced that:

[0112]

[0113] Compare the first item in the curly brackets of formula (30) and formula (31). If then there is At this time, letting s i be decoded before s j can provide a greater signal gain; on the contrary, if then there is At this time, it is better to let s j be decoded first. Therefore, the optimal sorting method of the decoding order is to let the sub - message with a higher channel gain be decoded first.

[0114] To sum up, at the m - th AP, for each sub - message s k,j , define:

[0115]

[0116] Then the optimal order for the AP to decode these sub - messages is to sort them in decreasing order according to the value of .

[0117] Based on the above analysis, we propose an alternating iterative optimization algorithm based on the BCD algorithm, and the overall process is shown in Algorithm 4.

[0118]

[0119]

[0120] The beneficial effects achieved by the present invention are as follows: By combining the dynamic optimization of RIS passive beamforming and the high-efficient signal processing ability of RSMA, multi-user interference is reduced, and the task scheduling ability of edge computing is enhanced. Through a multi-level optimization strategy, dynamic allocation of communication and computing resources is achieved, the stability and robustness of task offloading are improved, and the task offloading delay is minimized.

[0121] The technical means disclosed in the solution of the present invention are not limited to the technical means disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features.

Claims

1. A method for offloading edge tasks in a non-cellular uplink communication system based on rate division multiple access assisted by intelligent reflective surface, characterized in that: The following steps are involved: Step S1, firstly establish a RIS-assisted non-cellular uplink RSMA-MEC system model, including a communication model and a task offloading model, which includes multiple distributed access points AP, a smart reflection surface RIS and multiple user equipment UE; Construct a UE-AP matching factor matrix and design a matching strategy based on channel state information (CSI) and AP computing resource availability, so that the UE can adaptively select the optimal AP for communication; Step S2, using rate division multiple access (RSMA) technology, splitting the uplink transmission task into multiple sub-messages at the user end and allocating different transmit powers, decoding using successive interference cancellation (SIC) technology at the AP end, and improving the achievable rate of the sub-messages by optimizing power allocation and decoding order; introducing RIS to dynamically optimize passive beamforming and adjust the channel environment to optimize the communication performance of edge task offloading; Step S3, establish a task offloading model, in which each UE needs to perform a computationally intensive task, which can be divided into multiple subtasks; some tasks are calculated locally, and some tasks are offloaded to the AP-side MEC server communicating with it through the uplink and perform the calculation task. The offloading factor η k =(η k,0 ,η k,1 ,…,η k,M ),η k,m ∈[0,1] represents the proportion of subtask data that user k calculates locally and offloads to m MEC servers; Step S4, establishing an optimization problem, jointly optimizing UE-AP matching, task offloading factor, RIS phase shift matrix, AP receive beamforming and decoding order to minimize task processing delay and improve system throughput and energy efficiency; Step S5, using the block coordinate descent (BCD) optimization framework to solve, alternately optimizing each variable so that the system iteratively converges to the optimal solution.

2. According to claim 1, a method for unloading edge tasks in a non-cellular uplink communication system based on rate division multiple access assisted by an intelligent reflective surface, characterized in that: In step S1, the non-cellular communication system is equipped with M equipped with N r The AP has a receiving antenna, and K single-antenna users communicate with them uplink. A RIS containing N reflective units is deployed between the AP and the user to assist in communication. The CPU, as the central control unit, communicates with all APs through the backhaul network to exchange network information. A matching factor matrix is ​​established between the UE and AP points. where d m,k =1 means that the kth user successfully matches with the mth AP and establishes a data transmission link; represents the set of AP points selected by the kth user, Represents the set of users served by the mth AP.

3. According to claim 1, a method for unloading edge tasks in a non-cellular uplink communication system based on rate division multiple access assisted by an intelligent reflective surface, characterized in that: In step S2, when establishing uplink communication, the RSMA technology is used to divide each user's message into J sub-message parts, and different powers are allocated to these sub-messages and linearly superimposed, and sent to the AP at the same time and frequency resources. k,j Indicates the assignment to sub-message s k,j The transmission power of user equipment k satisfies At the receiving end, each AP uses SIC technology to decode all users’ sub-messages from the received message. Each time the AP decodes a sub-message, it will use the SIC technology to eliminate it from the entire data stream and then continue to decode the remaining content; therefore, for sub-message s k,j , all sub-messages that are decoded earlier than it have been eliminated, while the remaining sub-messages that have not been decoded continue to exist as interference.

4. According to claim 1, a method for unloading edge tasks in a non-cellular uplink communication system based on rate division multiple access assisted by an intelligent reflective surface, characterized in that: In step S3, an independent MEC server is deployed on each AP point side, and each user has a divisible computing-intensive task. Through the uplink communication link, the task offloading factor is determined according to the optimization strategy, and part of the computing task is offloaded to the corresponding MEC server for processing; the entire computing cycle is divided into four time periods; The first period: The RSMA transmission power allocation factor and task offloading factor are determined during the period from 0 to T1; Second period: During the period from T1 to T2, the user processes part of the task locally, and divides the rest into multiple subtasks and offloads the data to the MEC server deployed by the matching AP via the uplink; The third period: T2-T3, each MEC server performs edge computing tasks; in the second and third periods, i.e., T1-T3, user devices synchronously process local computing tasks; The fourth period: During the period from T3 to T4, the task calculation results are downloaded to the user end.

5. According to claim 1, a method for offloading edge tasks in a non-cellular uplink communication system based on rate division multiple access assisted by an intelligent reflective surface, characterized in that: In the step S4, UE-AP matching, user-side sub-message power allocation and task offloading factor, RIS phase shift matrix, AP-side receive beamforming and decoding order are jointly optimized to minimize the task processing delay of the RIS-assisted non-cellular uplink RSMA-MEC system under power constraints, quality of service QoS constraints, constant modulus constraints of beamforming, and sub-message fairness constraints.

6. The method for offloading edge tasks in a non-cellular uplink communication system based on rate division multiple access assisted by an intelligent reflective surface according to claim 1, characterized in that: In step S5, alternating optimization is performed based on the framework of the block coordinate descent (BCD) algorithm, including the following steps: Step S5-1, optimize the UE-AP matching matrix, and propose a UE-AP matching algorithm based on CSI and computing resource availability. In each round of matching, the AP with the best channel quality and sufficient computing resources is preferentially assigned to the user for access, ensuring the best throughput and the lowest task processing delay under the current channel state; Step S5-2, optimizing the decoding order based on the channel gain and the sub-message segmentation ratio, according to By sorting the sub-messages in descending order of the value of , the purpose of maximizing the signal gain can be achieved; Step S5-3, optimizing the power allocation factor, using the continuous convex approximation SCA method to optimize the optimal allocation of user sub-message power to reduce computational complexity and improve efficiency; Step S5-4, optimize the RIS phase shift matrix and receive beamforming, adopt the idea of ​​alternating optimization, and design an alternating Riemann manifold ARM algorithm based on Riemann conjugate gradient descent for optimization to ensure that the variables meet the constant modulus constraint and reduce the task offloading delay. Step S5-5, repeat the above steps S5-1 to S5-4, iterating until converging to the global optimal solution.

Images