Heterogeneous mobile computing power network resource allocation method and architecture

Abstract

The invention provides a heterogeneous mobile computing power network resource allocation method and architecture, and the method comprises the steps: carrying out the distribution modeling of a macro base station, a micro base station and a terminal user through employing an independent homogeneous Poisson point process PPP; selecting a computing power layer for unloading according to the type of the task generated by the terminal user; in a computing power layer, a terminal user selects a base station with the nearest space distance to access, a computing task is packaged into data packets of a first data volume and a second data volume, and the data packets are transmitted to a base station side through an uplink; the base station side decodes the data packet after receiving the data packet, and submits the successfully decoded data packet to a server of the base station side to execute calculation operation; the computing power of the central cloud server is greater than that of the edge cloud server; the central cloud computing buffer area supports an infinitely long waiting queue, and the size of the edge cloud computing buffer area is the finite length of each task; the terminal users are distributed according to the PPP, and the calculation tasks successfully unloaded to the server form the PPP; and transmitting the calculation result to a terminal user by adopting a downlink.

Description

Technical Field

[0001] This application relates to the field of mobile computing technology, specifically to a method and architecture for allocating resources in a heterogeneous mobile computing network. Background Technology

[0002] In recent years, a large number of computationally intensive applications have emerged, such as autonomous driving, the Industrial Internet, and smart cities, driving the demand for mobile computing networks. Vehicle-to-everything (V2X) communication must complete the process from command issuance to execution feedback within milliseconds. Computing nodes within factory parks ensure data remains within the park, enabling localized, high-speed processing. Urban video surveillance analysis generates massive amounts of data, but real-time requirements are relatively low. Mobile computing nodes can perform preliminary filtering and compression of the data, transmitting valuable data back to the cloud to save bandwidth. Unlike traditional cloud computing, which uses a centralized architecture, mobile computing networks deploy servers close to the user on the network side, including base stations, vehicles, and monitoring terminals, providing ubiquitous and flexible access. Any task-requesting terminal can access the network anytime, anywhere via 5G (Sub-6GHz, millimeter wave), 6G (terahertz), Wi-Fi, and other wireless methods.

[0003] In mobile computing networks, users offload computing tasks to network-side computing resources via the uplink. Computation results are sent back via the downlink, making communication performance crucial for both task uploading and result return. As key technologies for 5G / 6G, millimeter wave, terahertz, and WiFi can effectively improve data transmission rates by unlocking new frequency bands and achieving greater available bandwidth. Heterogeneous computing resources are deployed on different types of base stations, including macro base stations and micro base stations, resulting in differences in computing power and cache queues, which significantly impacts computing performance. Modeling, analyzing, and evaluating network performance in large-scale mobile computing networks can meet diverse task requirements and guide practical mobile computing network deployment. Summary of the Invention

[0004] The purpose of this invention is to solve the aforementioned technical problems existing in existing mobile computing networks.

[0005] To achieve the above objectives, this invention provides a heterogeneous mobile computing power network resource allocation method, applied to a network architecture composed of central computing power from macro base stations and edge computing power from micro base stations. This method employs an independent homogeneous Poisson point process (PPP) to model the distribution of macro base stations, micro base stations, and end users. Based on the type of task generated by the end user, a computing power layer is selected for offloading. At the computing power layer, the end user selects the base station with the closest spatial distance to access the network, encapsulates the computing task into data packets of a first data volume and a second data volume, and transmits them to the base station via the uplink. Upon receiving the data packets, the base station decodes them, and successfully decoded data packets are submitted to the server on the base station side for computation. The computing power of servers in different computing power layers is measured by computing capacity, with the central cloud server having a greater computing power than the edge cloud server. The central cloud computing buffer supports an infinitely long waiting queue, while the size of the edge cloud computing buffer is... The task has a finite length; the end user distributes the computational tasks to the server according to the PPP distribution, and the computational tasks are successfully offloaded to the server to form a PPP; the computational results are transmitted to the end user using a downlink.

[0006] On the other hand, this invention provides a heterogeneous mobile computing network architecture, consisting of a central computing layer accessed by macro base stations and edge computing layers accessed by micro base stations. A central cloud server cluster is deployed on the macro base station side, and edge servers are deployed on the micro base station side. The computing layer is selected for offloading based on the type of task generated by the end user. In the computing layer, the end user selects the base station with the closest spatial distance to access the network, encapsulates the computing task into data packets of a first data volume and a second data volume, and transmits them to the base station side via the uplink. Upon receiving the data packets, the base station side decodes them, and successfully decoded data packets are submitted to the server on the base station side for computation. The computing power of servers in different computing layers is measured by computing capacity, with the central cloud server having a greater computing power than the edge cloud server. The central cloud computing buffer supports an infinitely long waiting queue, and the size of the edge cloud computing buffer is... The finite length of each task; the computing tasks successfully offloaded to the server by the end user according to the PPP distribution form a PPP.

[0007] This invention presents a heterogeneous mobile computing network architecture, constructing a two-layer model consisting of central computing power from macro base stations and edge computing power from micro base stations, exhibiting heterogeneous characteristics in computing, storage, and network resources. To meet the differentiated task requirements of terminals, an edge-cloud collaborative task offloading mechanism and matching task processing strategies are designed. The mobile computing network is modeled and analyzed based on stochastic geometry and queuing theory. Key performance indicators such as successful offloading rate, transmission rate, and latency are evaluated, providing effective support for optimized deployment of network planning, resource allocation, and task scheduling. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the various embodiments disclosed in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only a few embodiments disclosed in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of a method for offloading edge-cloud collaborative tasks in a mobile computing network, provided by an embodiment of the present invention.

[0010] Figure 2 A schematic diagram of edge-cloud collaborative offloading for diverse tasks;

[0011] Figure 3 A schematic diagram of the modeling and evaluation process for heterogeneous mobile computing network models. Detailed Implementation

[0012] The present specification will now be described in further detail with reference to the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. The described embodiments are only a part of the embodiments described herein, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without inventive effort are within the scope of protection of this application.

[0013] This invention provides a heterogeneous mobile computing network architecture. This architecture, through the collaboration of macro base stations and micro base stations, achieves a gradient distribution of central and edge computing resources in terms of coverage and service capabilities, meeting the needs of diverse terminal applications. The core of this heterogeneous mobile computing network architecture is the construction of two complementary computing layers with different capabilities in computing, storage, and networking: a central computing layer accessed by macro base stations and an edge computing layer accessed by micro base stations.

[0014] The central computing layer for macro base station access: As the core anchor point for wide-area network coverage, macro base stations leverage the strong penetration and wide coverage advantages of traditional Sub-6GHz spectrum to provide a stable, continuous, and highly reliable access network. A central cloud server cluster with large-scale computing capabilities is deployed on the macro base station side, equipped with powerful central processing units (CPUs) and high-performance graphics processing units (GPUs) capable of handling large-scale parallel computing tasks. The computing buffer is designed to support near-infinitely long task waiting queues, accommodating massive concurrent computing requests without easily causing congestion. Simultaneously, the central cloud server cluster is equipped with massive storage space to store various computing services, including complex path planning, a vast high-precision map database, and massive video surveillance, providing computing services for computationally intensive and highly complex tasks.

[0015] Edge computing layer accessed by micro base stations: Micro base stations are densely deployed in hotspot areas, utilizing new millimeter wave and WiFi technologies to provide fast and flexible network access. Lightweight, low-power edge servers are deployed on the micro base station side. Due to physical size and cost limitations, computing resource capacity is relatively small, and the task waiting queue length is limited, potentially leading to task overflow under high load scenarios. Storage space is also relatively small, only able to cache the most frequently used and critical computing services in the current scenario, such as environmental recognition, event alerts, and visual quality inspection, to support rapid response.

[0016] Based on heterogeneous computing hardware, the heterogeneous mobile computing network architecture proposes differentiated service processing strategies: Macro base stations, acting as central computing nodes, leverage powerful computing capabilities, high throughput bandwidth, and massive storage resources, making them suitable for handling computationally intensive and complex tasks. These include large-scale environmental data mining and analysis, urban macro-traffic planning, and the training and iteration of complex artificial intelligence models. Micro base stations, acting as edge computing nodes, focus on handling low-computation, low-complexity, and latency-sensitive real-time computing services, including real-time rendering of augmented reality, obstacle avoidance decisions for autonomous vehicles, and high-frequency sensor data filtering.

[0017] Based on a heterogeneous mobile computing network architecture, embodiments of this invention provide an edge-cloud collaborative task offloading mechanism, such as... Figure 1 and Figure 2 As shown.

[0018] Smartphones, autonomous vehicles, smart cameras, and other terminal devices generate diverse computing tasks, requiring low latency, high computing power, and large bandwidth. However, due to limitations in size, power consumption, and cost, these devices' limited computing resources and battery life cannot support these demands. Therefore, a dynamic edge-cloud collaborative task offloading mechanism is proposed. Terminal devices encapsulate computing task data into data packets and upload these packets to the network via 5G / 6G wireless transmission links such as Sub-6GHz, millimeter wave, and WiFi. The base station-side network controller schedules the task to a computing node that stores the corresponding computing services based on task attributes and network status. On the computing node, the server retrieves tasks from the queue buffer in the order they arrive and executes the computation operations sequentially. After completing the computation service, the results are sent back to the end user via a downlink. The complete cycle of edge-cloud collaborative task offloading includes: task generation, data upload, computation execution, and result feedback.

[0019] The modeling of heterogeneous mobile computing networks is carried out over the entire task unloading cycle, using a combination of stochastic geometry and queuing theory.

[0020] I. In the task generation step, this embodiment of the invention employs an independent homogeneous Poisson point process (PPP) to model the distribution of macro base stations, micro base stations, and terminal users. The distributions x, y, z of macro base stations, micro base stations, and terminal users successively follow the density... , , PPP. Based on differences in execution complexity and computational load, tasks are divided into two categories: high-complexity tasks and low-complexity tasks. The central computing power connected to macro base stations stores computational services for high-complexity tasks. The edge computing power connected to micro base stations stores computational services for low-complexity tasks. End-user task generation is modeled as per unit time slot. The probability distributions are as follows: the request probabilities for the two types of tasks are respectively... and Each requires , The number of CPU cores completes the operation, and .

[0021] Second, in the data upload step, this embodiment of the invention selects an appropriate computing power layer for unloading based on the type of task generated by the end user.

[0022] At the specific computing layer, end users select the nearest spatially distant base station for access and encapsulate computing tasks into data volumes. , Data packets are transmitted to the base station via the uplink. Upon receiving the data packets, the base station decodes them, and successfully decoded packets are submitted to the base station's server for computation. Successful reception and decoding of data packets by the base station is crucial for offloading during data upload. In interference-free or controlled-interference scenarios, the offloading success rate is determined by the ratio of the received signal to white noise in the link.

[0023] Specifically, the signal-to-noise ratio (SNR) is the received signal power. With noise power ratio The SNR determines channel capacity and bit error rate. A higher SNR increases the probability of the base station successfully decoding data packets, resulting in a higher offload rate. Assuming the end user operates at a fixed power... Transmitted signals experience path loss and decline The received signal power of the base station Modeling as Path loss is a function of transmission distance and is related to factors such as carrier frequency, propagation environment, and antenna configuration. Small-scale fading is caused by multipath propagation and is closely related to two key factors: multipath propagation and relative motion.

[0024] In one example, taking a macro base station using sub-6Hz and a micro base station using millimeter wave, the path loss of the macro base station is modeled using a power-law model. ,in It is the path loss constant, which is related to the carrier frequency and the antenna. This is the path loss exponent, which depends on the propagation environment and represents the rate at which signal power decreases with distance. Millimeter-wave signals have extremely short wavelengths, poor diffraction capabilities, and are easily blocked by obstacles, relying on line-of-sight propagation. The path loss model uses a line-of-sight path loss model. , This indicates the maximum line-of-sight link length.

[0025] Depending on the type of generated task, the end user selects the nearest base station to unload from the computing layer. The cumulative distribution function (CDF) of the service distance can be obtained by deriving the Poisson point process, which guarantees that there is at least one node within a circular region of radius r. ,in, Indicates the first The service distance variable of the layer, This represents the computing power layer; 1 is the macro base station layer, and 2 is the micro base station layer. It is the first The density of the layered Poisson point process. The service distance probability density function PDF is obtained by differentiating the CDF. .

[0026] Small-scale fading in sub-6 Hz Using a Rayleigh distribution with a unit mean, The CDF function is represented as Small-scale fading in millimeter-wave links It follows Nakagami fading with shape parameter N. It is a normalized Gamma random variable, and the CDF function can be represented using a regularized incomplete gamma function. , It is a regularized incomplete gamma function, where N represents the shape parameter.

[0027] Based on the SNR threshold required for packet decoding A performance metric for the probability of successful uninstallation is proposed. This indicates that if the uplink SNR is greater than the predefined threshold... If so, the computational task is successfully unloaded. This is achieved using the service distance distribution function. and small-scale fading CDF function The probability of successful uninstallation can be derived as follows:

[0028]

[0029] in, Indicates the transmit power of the end user. It is path loss. This indicates small-scale fading. Indicates noise power. Indicates the SNR threshold. Indicates the maximum service distance.

[0030] The successful uninstallation rate of sub-6hz is expressed as follows:

[0031] in, Indicates uplink, Indicates the SNR threshold. Indicates the transmit power of the end user. It is the path loss constant of the sub-6Hz link. It is the path loss index of the sub-6Hz link. Indicates noise power. It is the service distance of the macro base station layer. The distribution function, It is the density of the Poisson point process at the macro base station layer.

[0032] The successful offloading rate of millimeter waves is expressed as:

[0033] in, Indicates uplink, Indicates the SNR threshold. Indicates the transmit power of the end user. Indicates noise power. It is an incomplete gamma function in terms of regularization. Indicates shape parameters. This indicates the maximum line-of-sight link length of the millimeter-wave link. It is the path loss constant of the millimeter-wave link. It is the path loss index of the millimeter-wave link. It is the service distance of the micro base station layer. The distribution function, It is the density of the Poisson point process at the micro base station layer.

[0034] III. In the computation execution step, the computing power of servers at different computing power tiers is measured using computing capacity, expressed as CPU cores per second. The computing power of the central cloud server... Greater than the computing power of edge cloud servers , The central cloud computing buffer supports an unlimited-length wait queue. The edge cloud computing buffer has a size of [missing information]. The finite length of each task.

[0035] According to PPP distribution, the computing tasks successfully offloaded to the server by the end user form PPP, and the server task arrival rate of different computing power layers. Depends on the number of computational tasks successfully offloaded by all terminal users on the access base station side in each time slot:

[0036]

[0037] in, Indicates the computing power layer. It is generated by the end user. The request probability of each type of task. Indicates the first The probability of successful offloading of the computing layer. This represents the density of the Poisson point process distribution for end users. Indicates the first The density of the Poisson point process distribution of the base station in the computing layer.

[0038] The service rate of a computing server is measured by the number of computing tasks processed per second. , Indicates the first The computing capacity of the computing layer Indicates execution of the first The number of CPU cores required for each type of task. Measure the service time of each computational task. It follows an independent exponential distribution. It can be utilized using queuing theory. , The system models computing services.

[0039] IV. In the result feedback step, this embodiment of the invention uses a downlink to transmit the calculation results to the end user. The transmission power of the macro base station and micro base station is... , Since the uplink and downlink are symmetrical, a performance metric for successful return rate is proposed. This indicates that if the downlink SNR is greater than a predefined threshold... If the result is successfully received, the calculation is then transmitted back. This is achieved using the service distance distribution function and the small-scale fading CDF function. , This allows us to deduce the successful return rate.

[0040] The successful backhaul rate of sub-6Hz is expressed as:

[0041] The successful return rate of millimeter waves is expressed as:

[0042] The following evaluation focuses on heterogeneous mobile computing networks:

[0043] In heterogeneous mobile computing networks, end-to-end latency is a key performance indicator. Latency permeates every stage from wireless transmission and task scheduling to heterogeneous computing, directly determining the user experience and even functional safety of real-time services such as cloud gaming, industrial IoT, and remote driving. It reflects the overall efficiency of network and computing power collaborative scheduling. The end-to-end latency of a computing task represents the total time from when a user terminal initiates a task request to when it receives the final computation result, and it consists of uplink transmission latency, computation latency, and downlink transmission latency.

[0044] Uplink transmission latency: The time required for task data to be transmitted from the terminal to the computing node, which depends on factors such as data volume, transmission bandwidth, and network quality. In uplink transmission, the user's bandwidth... Based on the requirement of successful packet offloading, the average uplink transmission rate is obtained. Uplink transmission latency is obtained by the ratio of the task data packet size to the average transmission rate. .

[0045] Computation service latency includes queuing latency and computation processing latency. Queuing latency is the waiting time for a task at a computing node, which depends on the current load of the computing node. Computation processing latency is the time it takes for a computing node to execute a computation task, which depends on the capacity of heterogeneous computing resources. The central server and edge servers use... , The queue system is used for computational service modeling, with a service rate of Using the theory of average queue dwell time in queuing theory, the customer's dwell time in the system includes queuing time and processing time, which allows us to calculate service latency in scenarios with both infinite and finite cache queues:

[0046] ,

[0047] Downlink transmission latency: The time required for calculation results to be transmitted from the computing node back to the terminal. Generally, the amount of data in the result is small, and the latency is also short. In downlink transmission, the bandwidth allocated to the user by the base station... Two types of task result data packets , Average transmission rate Downlink transmission delay is obtained by the ratio of the resulting data packet size to the average downlink transmission rate: .

[0048] End-to-end latency is modeled as the sum of three latency segments: uplink transmission, compute service, and downlink transmission.

[0049] By studying diverse business scenarios, a heterogeneous mobile computing network model is constructed to simulate behaviors such as task generation, data transmission, and computation execution. Key indicators such as offload rate, latency, and speed are quantitatively analyzed, revealing the inherent laws and potential bottlenecks of multi-dimensional parameters and multiple indicators in complex networks. This supports the deployment optimization of node location and topology design in practical wide-area mobile computing networks. The advancement of new technologies such as task scheduling algorithms and network transmission schemes is compared and verified. Resource scheduling and task allocation are continuously evaluated and adjusted to achieve optimal synergy between network reliability, service quality, and economic benefits.

[0050] Figure 3 A schematic diagram illustrating the basic process of modeling and evaluating heterogeneous mobile computing network models. (For example...) Figure 3 As shown, the basic process includes parameter setting, behavior modeling, performance evaluation, and deployment optimization:

[0051] Parameter settings: Set task load parameters (data input size, computing resource requirements, tolerable latency, result output size), network parameters (number of nodes, topology, bandwidth, path loss, fading), and computing power parameters (computing frequency, storage capacity, cache queue), etc.

[0052] Behavioral modeling: Using probabilistic and statistical modeling methods, we simulate behaviors such as task generation, data transmission, task execution, and result feedback.

[0053] Performance evaluation: Key network performance metrics such as offload rate, transmission rate, and latency are evaluated using theories such as stochastic geometry and queuing theory.

[0054] Deployment optimization: Based on the requirements of key network performance, optimize the configuration of the number of nodes, task scheduling and resource allocation by using artificial intelligence, traditional gradient descent and other optimization methods.

[0055] The heterogeneous mobile computing network architecture provided in this invention constructs a two-layer model consisting of central computing power from macro base stations and edge computing power from micro base stations, exhibiting heterogeneous characteristics in computing, storage, and network resources. To meet the differentiated task requirements of terminals, an edge-cloud collaborative task offloading mechanism and matching task processing strategies are designed. The mobile computing network is modeled and analyzed based on stochastic geometry and queuing theory. Key performance indicators such as successful offloading rate, transmission rate, and latency are evaluated, providing effective support for the optimized deployment of network planning, resource allocation, and task scheduling.

[0056] The specific embodiments described above further illustrate the purpose, technical solutions, and beneficial effects of the multiple embodiments disclosed in this specification. It should be understood that the above description is only within the scope of this specification. Any modifications, equivalent substitutions, improvements, etc., made based on the technical solutions of the multiple embodiments disclosed in this specification should be included within the protection scope of the multiple embodiments disclosed in this specification.

Claims

1. A method for resource allocation in a heterogeneous mobile computing network, applied to a network architecture composed of macro base station central computing power and micro base station edge computing power, characterized in that: a homogeneous Poisson point process (PPP) is used to model the distribution of macro base stations, micro base stations and terminal users; according to the type of the task generated by the terminal user, the computing layer is selected for offloading; in the computing layer, the terminal user selects the base station with the nearest distance in space to access, encapsulates the computing task into data packets of the first data amount and the second data amount, and transmits them to the base station side through the uplink; after receiving the data packets, the base station side decodes them, and submits the successfully decoded data packets to the server on the base station side for computing operation; and the computing result is transmitted to the terminal user through the downlink. 6.A heterogeneous mobile computing network architecture composed of a central computing power layer accessed by a macro base station and an edge computing power layer accessed by a micro base station; a central cloud server cluster is deployed on the macro base station side, and an edge server is deployed on the micro base station side; characterized in that: according to the type of the task generated by the terminal user, the computing layer is selected for offloading; in the computing layer, the terminal user selects the base station with the nearest distance in space to access, encapsulates the computing task into data packets of the first data amount and the second data amount, and transmits them to the base station side through the uplink. ​ ​ ​ The server computing capacity of different computing power layers is measured by computing capacity, and the computing capacity of the center cloud server is greater than that of the edge cloud server; the center cloud computing buffer supports an infinite long waiting queue, and the size of the edge cloud computing buffer is The limited length of a task; and the terminal user is distributed according to the PPP, and the successfully unloaded computing task to the server forms a PPP; ​ 2. The method of claim 1, wherein, The distribution x, y, z of the macro base station, micro base station and terminal user obeys the PPP with density , , The tasks are divided into two types of first complexity task and second complexity task according to the difference of execution complexity and calculation amount. The central computing power connected with the macro base station stores the computing service of the first complexity task; the edge computing power connected with the micro base station stores the computing service of the second complexity task; and the task generation of the terminal user is modeled as a probability distribution of unit time slots , the request probabilities of the two types of tasks are and , respectively, and , CPU cores are required to complete the operation, and .

3. The method of claim 1, wherein, During data upload, the offloading success rate is determined by the ratio of the received signal to white noise in the link; the signal-to-noise ratio (SNR) is the received signal power. With noise power ratio The SNR determines the channel capacity and bit error rate; the higher the SNR, the greater the probability that the base station will successfully decode the data packet and the higher the offloading rate.

4. The method of claim 1, wherein, The server task arrival rate at different computing power layers depends on the number of computing tasks successfully offloaded by all terminal users on the access base station side in each time slot: the service rate of the computing power server is measured by the number of computing tasks processed per second; the service time of each computing task follows an independent exponential distribution; and queuing theory is used. , The system models computing services.

5. The method of claim 1, wherein, Macro and micro base station transmit power is , , since the uplink and downlink are symmetric, the resulting successful backhaul rate performance metric , indicates that the result is successfully backhauled if the downlink SNR is greater than a predefined threshold ; the successful backhaul rate is derived using the distribution function of the service distance and the small scale fading CDF function , . ​ ​ The base station side receives the data packet and decodes it, and the successfully decoded data packet is submitted to the server on the base station side to perform a calculation operation; the calculation capacity of the servers of different computing power layers is measured by calculation capacity, and the calculation capacity of the central cloud server is greater than that of the edge cloud server; the central cloud computing buffer supports an infinite long waiting queue, and the size of the edge cloud computing buffer is The limited length of one task; the terminal user is distributed according to the PPP, and the successfully unloaded calculation task to the server forms a PPP.

7. The system of claim 6, wherein, The distribution x, y, z of the macro base station, the micro base station and the terminal user obeys the PPP with the density , , According to the difference of execution complexity and calculation amount, the task is divided into two types of high complexity task and low complexity task. The central computing power connected by the macro base station stores the computing service of high complexity task; the edge computing power connected by the micro base station stores the computing service of low complexity task; the task generation of the terminal user is modeled as the probability distribution of unit time slot, the request probabilities of the two types of tasks are and , respectively, and , CPU cores are required to complete the operation, and .

8. The system of claim 6, wherein, During data upload, the offloading success rate is determined by the ratio of the received signal to white noise in the link; the signal-to-noise ratio (SNR) is the received signal power. With noise power ratio The SNR determines the channel capacity and bit error rate; the higher the SNR, the greater the probability that the base station will successfully decode the data packet and the higher the offloading rate.

9. The system of claim 6, wherein, The server task arrival rate at different computing power layers depends on the number of computing tasks successfully offloaded by all terminal users on the access base station side in each time slot: the service rate of the computing power server is measured by the number of computing tasks processed per second; the service time of each computing task follows an independent exponential distribution; and queuing theory is used. , The system models computing services.

10. The system of claim 6, wherein, Macro and micro base station transmit power is , , since the uplink and downlink are symmetric, the resulting successful backhaul rate performance metric , indicates that the result is successfully backhauled if the downlink SNR is greater than a predefined threshold ; using the distribution function of the service distance and the small scale fading CDF function , , the successful backhaul rate can be derived.

Images