Task Offloading Method and Apparatus based on Multiple Radio Access Technologies for URLLC and eMBB Services

Abstract

The present disclosure provides a task offloading method and apparatus capable of improving performance and reducing energy consumption by maximizing resource utilization while satisfying different QoS requirements for each service, comprising the steps of: analyzing a task to be processed and collecting information for task offloading; performing neural network operations using a reinforcement learning model on the task and the collected information to estimate an offloading ratio to be transmitted to a Multi-access Edge Computing (MEC) server among at least one task packet constituting the task; separating at least one task packet into a local processing packet and an offloading packet according to the offloading ratio; if the task is a task for an enhanced Mobile Broadband (eMBB) service, separating the offloading packet according to each of a plurality of Radio Access Technologies (RATs) configured with the MEC server and storing it in a plurality of RAT queues of a local queue; and offloading the offloading packets separated and stored in the plurality of RAT queues to the MEC server using the plurality of RATs.

Description

Technology Field

[0001] The present disclosure relates to a task offloading method and apparatus, and more specifically to a task offloading method and apparatus for multiple wireless access technology to support URLLC and eMBB services. Background Technology

[0002] With the advancement of science and technology, there is a growing need for Internet of Things (IoT) networks to provide wireless users with services ranging from real-time to broadband. To this end, technology is required to simultaneously provide Ultra Reliability Low Latency Communication (URLC) services, which are representative 5G services characterized by ultra-low latency and high reliability, and Enhanced Mobile Broadband (eMBB) services with high bandwidth.

[0003] URLLC is designed to provide high reliability and ultra-low latency for 5G applications such as extended reality (XR), factory automation, autonomous vehicles, and remote medical surgery. On the other hand, eMBB is designed to support broadband wireless and high-volume data processing services for 5G users, such as cloud game streaming and high-definition video streaming. As these diverse heterogeneous services become more varied, it is becoming increasingly difficult for 5G networks to meet different and stringent Quality of Service (QoS) requirements. Furthermore, since most User Equipment (UEs) receiving these services operate on batteries, minimizing energy consumption is critical. However, providing latency-sensitive and computationally intensive tasks while meeting QoS requirements places a significant burden on resource-constrained User Equipment.

[0004] One solution to address these issues is to save energy by having Multi-access Edge Computing (MEC) systems perform task processing instead of user terminals. Furthermore, to minimize user terminal energy consumption, various variables must be considered depending on network conditions.

[0005] In network situations with many tasks, resources may be insufficient with only existing 5G Radio Access Technology (RAT) and a single MEC server. Therefore, instead of a single RAT, it is necessary to select an available MEC server using multi-RAT such as 5G, LTE, and Wi-Fi to perform offloading. Accordingly, user terminals and multiple MEC servers must have a structure suitable for a multi-RAT environment and be able to perform offloading efficiently. The problem to be solved

[0006] The purpose of the present disclosure is to provide a task offloading method and apparatus that can improve performance by maximizing resource utilization while satisfying different QoS requirements for each service.

[0007] The object of the present disclosure is to provide a task offloading method and apparatus capable of reducing energy consumption. means of solving the problem

[0008] A task offloading method according to one embodiment of the present disclosure comprises: a step of analyzing a task to be processed and collecting information for task offloading; a step of performing neural network operations using a reinforcement learning model on the task and the collected information to estimate an offloading ratio to be transmitted to a Multi-access Edge Computing (MEC) server among at least one task packet constituting the task; a step of separating the at least one task packet into a local processing packet and an offloading packet according to the offloading ratio; a step of, if the task is a task for an enhanced Mobile Broadband (eMBB) service, separating the offloading packet according to each of a plurality of Radio Access Technologies (RATs) configured with the MEC server and storing the packet in a plurality of RAT queues of a local queue; and a step of offloading the offloading packet separated and stored in the plurality of RAT queues to the MEC server using the plurality of RATs.

[0009] The step of estimating the offloading ratio above can estimate a plurality of RAT offloading ratios representing the ratio of task packets to be offloaded to the MEC server using each of the plurality of RATs among the at least one task packet.

[0010] The step of estimating the offloading ratio can adjust the number of task packets stored in each of the multiple RAT queues by estimating the multiple RAT offloading ratios differently according to channel state information collected for each of the multiple RATs.

[0011] If the task is a task for an Ultra Reliability Low Latency Communication (URLC) service, the above task offloading method selects the RAT with the highest data transmission rate among the multiple RATs based on collected channel state information, and can immediately offload the offloading packet to the MEC server through the selected RAT without storing it in the local queue.

[0012] The above task offloading method can store the local processing packets, separated according to the offloading ratio, in the local queue and process the stored local processing packets sequentially.

[0013] The step of estimating the offloading ratio above can estimate the offloading ratio such that the energy consumption of the user terminal offloading the task packet to the MEC server is minimized.

[0014] According to another embodiment of the present disclosure, a task offloading device comprises: a memory; and a processor that executes at least a portion of an operation according to a program stored in the memory, wherein the processor performs the steps of: analyzing a task to be processed and collecting information for task offloading; performing neural network operations using a reinforcement learning model on the task and the collected information to estimate an offloading ratio to be transmitted to an MEC server among at least one task packet constituting the task; separating the at least one task packet into a local processing packet and an offloading packet according to the offloading ratio; if the task is a task for an eMBB service, separating the offloading packet according to each of a plurality of RATs configured with the MEC server and storing it in a plurality of RAT queues of a local queue; and offloading the offloading packets separated and stored in the plurality of RAT queues to the MEC server using the plurality of RATs. Effects of the invention

[0015] The task offloading method and apparatus of the present disclosure can improve performance and reduce energy consumption by maximizing resource utilization while satisfying different QoS requirements for each service. Brief explanation of the drawing

[0016] Figure 1 shows an example of a multiple wireless access technology MEC system. FIG. 2 shows a configuration of a task offloading device according to one embodiment, classified according to operation. FIG. 3 shows an example of a user terminal and MEC server queue model according to an embodiment of the present invention. FIG. 4 illustrates an offloading method of a user terminal according to an embodiment of the present invention. FIG. 5 is a diagram illustrating a computing environment including a computing device according to one embodiment. Specific details for implementing the invention

[0017] Hereinafter, specific embodiments according to embodiments of the present disclosure will be described with reference to the drawings. The following detailed description is provided to facilitate a comprehensive understanding of the methods, apparatuses, and / or systems described herein. However, this is merely illustrative and the present invention is not limited thereto.

[0018] In describing the embodiments of the present disclosure, detailed descriptions of known technology related to the present invention are omitted if it is determined that such detailed descriptions would unnecessarily obscure the essence of the embodiments. Furthermore, terms described below are defined with consideration of their functions in the present invention, and these may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification. Terms used in the detailed description are intended merely to describe specific embodiments and should not be limiting. Unless explicitly stated otherwise, expressions in the singular form include the meaning of the plural form. In this description, expressions such as “include” or “compose” are intended to refer to certain characteristics, numbers, steps, actions, elements, parts thereof, or combinations thereof, and should not be interpreted to exclude the existence or possibility of one or more other characteristics, numbers, steps, actions, elements, parts thereof, or combinations thereof other than those described. Additionally, terms such as “...part,” “...unit,” “module,” and “block” described in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware, software, or a combination of hardware and software.

[0019] Figure 1 shows an example of a multiple wireless access technology MEC system.

[0020] Referring to FIG. 1, the MEC system may include multiple base stations (BS) and multiple user terminals (User Equipment) (UE1, UE2). Here, it is assumed that each of the multiple base stations (BS) is equipped with an MEC server that performs multiple access edge computing (MEC) operations. That is, each of the multiple base stations (BS) can operate as an MEC server, and the MEC server can process tasks offloaded from the user terminals and transmit the processing results to the user terminals (UE1, UE2). Here, it is assumed that the MEC server processes tasks for different services, namely Ultra Reliability Low Latency Communication (URLC) services and enhanced Mobile Broadband (eMBB) services.

[0021] Meanwhile, among the multiple user terminals (UE1, UE2) in FIG. 1, the first user terminal (UE1) represents a user terminal using the URLLC service, and the second user terminal (UE2) represents a user terminal using the eMBB service. That is, the distinction between the first and second user terminals (UE1, UE2) does not mean terminal devices having different configurations, but rather terminal devices distinguished according to the service used, such as the URLLC service or the eMBB service.

[0022] In addition, as described above, multiple base stations (BS) and multiple user terminals (UE1, UE2) can communicate with each other using Multi-Radio Access Technology (multi-RAT), such as 5G, LTE, and Wi-Fi. That is, they can communicate by connecting according to communication methods using different frequency bands. Accordingly, each of the multiple user terminals (UE1, UE2) connects to the base station (BS) using various connection methods and offloads tasks related to the services used to the MEC server of the base station (BS), thereby enabling the MEC server to handle tasks that are difficult to process with the limited performance and power of each user terminal (UE1, UE2). In other words, to ensure that users can receive high-quality services despite constraints on the performance and energy consumption of the user terminals (UE1, UE2), the user terminals (UE1, UE2) allow the MEC server of the base station (BS) to process some of the tasks that need to be processed.

[0023] However, distinct services such as URLLC and eMBB services require different QoS requirements. Specifically, while URLLC services prioritize real-time processing of tasks, eMBB services require high bandwidth and computational power to process large volumes of data. Since the QoS requirements of each service differ in this way, existing offloading techniques that perform offloading without distinguishing between services have limitations in that they cannot satisfy the QoS demands of each service.

[0024] In particular, for URLLC packets in URLLC services where real-time performance is paramount, tasks are composed of packets with very small data sizes (e.g., 32 bytes) for fast, low-latency processing. In contrast, for eMBB services that require processing large amounts of data, tasks may be composed of multiple eMBB packets with very large sizes (e.g., tens of Mbytes to hundreds of Mbytes). Consequently, if both URLLC packets and eMBB packets are offloaded to the MEC server, processing of URLLC packets may be delayed, potentially failing to satisfy the required real-time performance.

[0025] Accordingly, each of the multiple user terminals (UE1, UE2) must appropriately adjust the ratio of tasks to be offloaded to the MEC server, i.e., the offloading ratio, depending on the type of service used and the state of the multi-RAT-based MEC system, so as to satisfy the QoS required by the service used.

[0026] FIG. 2 shows a configuration of a task offloading device classified according to operation according to one embodiment, and FIG. 3 shows an example of a user terminal and MEC server queue model according to one embodiment of the present invention.

[0027] As described in FIG. 1, in a multi-RAT-based MEC system, task offloading is performed on each of the multiple user terminals (UE1, UE2), and the task offloading device shown in FIG. 2 can be configured to be included in each of the multiple user terminals (UE1, UE2).

[0028] Referring to FIG. 2, a task offloading device according to one embodiment may include a task analysis module (11), an information collection module (12), an offloading setting module (13), a MEC selection module (16), a local queue (14), a task processing module (15), and a communication module (17).

[0029] The task analysis module (11) collects and analyzes information regarding tasks generated according to the services provided to the user terminals (UE1, UE2). Here, tasks may be generated by the user terminals (UE1, UE2) but are not limited thereto; as a simple example, the explanation assumes that the user terminals (UE1, UE2) provide either a URLLC service or an eMBB service to the user.

[0030] The task analysis module (11) checks whether the collected task is a task for the URLLC service or a task for the eMBB service, and checks the size and number of task packets constituting each task.

[0031] Here, a task packet refers to data divided so that each task can be processed in a distributed manner. Since it is assumed that the user terminal (UE1, UE2) provides either a URLLC service or an eMBB service, the task packet can also be divided into a URLLC packet and an eMBB packet. However, in the case of the URLLC service, since the task size is small and low-latency real-time processing is required, it is not divided into multiple packets to prevent overhead caused by packet splitting and reconfiguration. That is, a task for the URLLC service consists of a single URLLC packet. On the other hand, since the eMBB service must process tasks for large-capacity data, it can be divided into multiple eMBB packets, and the size of each packet can also be adjusted in various ways. Accordingly, the task analysis module (11) can analyze the generated task to determine the type of service to be provided by processing the task, as well as the size and number of packets.

[0032] And the analyzed tasks can be separated according to the offloading ratio set in the offloading setting module (13) and delivered to the local queue (14).

[0033] The information collection module (12) collects various information that must be checked to adjust the rate at which user terminals (UE1, UE2) transmit task packets to the MEC server of the base station (BS), i.e., the offloading rate. For example, the information collection module (12) collects and obtains QoS conditions required by each service, the average packet arrival rate at which task packets reach the local queue (14), the local CPU frequency indicating the local processor operation speed of the user terminals (UE1, UE2), the number of CPU cycles required to process each task packet, and Channel State Information (CSI) according to each RAT of multiple RATs for each of the multiple base stations (BS). Here, the QoS conditions required by each service may include maximum latency and stability, and the Channel State Information can be collected by the user terminals (UE1, UE2) performing prior communication with the base station (BS) and each RAT.

[0034] The offloading setting module (13) determines the offloading ratio of task packets to be transmitted to the MEC server based on the task information analyzed by the task analysis module (11) and the information collected by the information collection module (12). The offloading setting module (13) estimates and sets an offloading ratio that minimizes energy consumption of user terminals (UE1, UE2) while satisfying the QoS conditions required by each service. Here, the offloading setting module (13) estimates the offloading ratio by learning based on a reinforcement learning model among neural network models, thereby causing the task analysis module (11) to distinguish each task packet from one another according to the estimated offloading ratio and transmit them to the local queue (14).

[0035] Meanwhile, as described above, while the tasks of the URLLC service consist of a single URLLC packet of small size, the tasks of the eMBB service may consist of multiple eMBB packets of relatively large size. Therefore, while URLLC packets can be offloaded to the MEC server through a single RAT, multiple eMBB packets can be distributed through multiple RATs and offloaded to the MEC server. Accordingly, the offloading setting module (13) can calculate the data transmission rate of each RAT from the channel state information of each of the multiple RATs between the user terminals (UE1, UE2) and the base station (BS) equipped with the MEC server, and estimate and set the RAT offloading ratio, which represents the ratio of task packets to be offloaded through each RAT among the task packets to be offloaded, based on the calculated data transmission rate.

[0036] In addition, when user terminals (UE1, UE2) and a base station (BS) equipped with an MEC server communicate through multiple RATs, the data transmission speed that can be transmitted through each RAT varies depending on the type of RAT and the channel status for each RAT. Accordingly, the offloading setting module (13) can set the RAT offloading ratios to be offloaded through each RAT differently so that a large number of task packets can be transmitted to the MEC server as efficiently as possible in a short time using multiple RATs. Furthermore, when offloading URLLC packets or eMBB packets to the MEC server using a selected RAT, the offloading setting module (13) can also determine the transmission power for each RAT to minimize energy consumption of the user terminals (UE1, UE2).

[0037] The offloading setting module (13) can estimate the RAT offloading ratio and transmission power per RAT using a reinforcement learning model based on the QoS conditions according to the service collected by the information collection module (12), the average packet arrival rate, the local CPU frequency, the number of required CPU cycles, and the channel state information according to each RAT of multiple RATs for the base station (BS) equipped with the selected MEC server. Here, the MEC server to which the task packet will be offloaded can be pre-selected by the MEC selection module (16).

[0038] The detailed operation of the offloading setting module (13) estimating the offloading ratio, RAT packet ratio, and RAT transmission power will be described later.

[0039] The local queue (14) temporarily stores task packets that are separated and transmitted from the task analysis module (11) and transmits the stored task packets to the task processing module (15) or the communication module (17). The local queue (14) may include service queues separated according to each service. Here, it is assumed that user terminals (UE1, UE2) use URLLC services and eMBB services, so the local queue (14) may include a URLLC queue (21) that receives and stores URLLC packets and an eMBB queue (22) that receives and stores eMBB packets, as shown in FIG. 3. The URLLC queue (21) consists only of local URLLC queues, whereas the eMBB queue (22) may be further divided into a local eMBB queue (31) and a RAT queue group (32), and the RAT queue group (32) may have a number of RAT queues (41 to 4k) equal to the number of multiple RATs.

[0040] Here, the local URLLC queue and local eMBB queue (31) are not offloaded to the MEC server, but receive and store URLLC packets and eMBB packets configured to be processed locally at user terminals (UE1, UE2), and transmit the stored URLLC packets and eMBB packets to the task processing module (15).

[0041] Then, the RAT queue group (32) receives and stores eMBB packets configured to be offloaded to the MEC server. In particular, multiple RAT queues (41 to 4k) of the RAT queue group (32) each receive and store eMBB packets to be transmitted to the MEC server according to a designated RAT among the eMBB packets to be offloaded. Then, the stored eMBB packets are transmitted to the MEC server through the communication module (17).

[0042] At this time, multiple eMBB packets to be offloaded to the MEC server through each RAT can be distributed and stored in multiple RAT queues (41 to 4k) of the RAT queue group (32) according to the RAT offloading ratio. To this end, the offloading setting module (13) sets the RAT offloading ratio so that the ratio of eMBB packets stored in multiple RAT queues (41 to 4k) differs according to the channel status of each analyzed RAT.

[0043] When user terminals (UE1, UE2) and a base station (BS) equipped with an MEC server communicate through multiple RATs, the data transmission speed that can be transmitted through each RAT varies depending on the type of RAT and the channel status of each RAT. Accordingly, the offloading setting module (13) sets the RAT offloading ratio so that a large number of eMBB packets can be transmitted to the MEC server as efficiently as possible in a short time using multiple RATs, and accordingly, multiple eMBB packets can be distributed and stored in multiple RAT queues (41 to 4k) according to the RAT offloading ratio.

[0044] Meanwhile, the eMBB queue (22) is divided into a local eMBB queue (31) and a RAT queue group (32) to store both eMBB packets to be processed at user terminals (UE1, UE2) and eMBB packets to be transmitted to the MEC server, whereas the URLLC queue (21) consists only of a local URLLC queue and does not store URLLC packets to be transmitted to the EC server. That is, authorized URLLC packets are immediately transmitted to the communication module (17) to be transmitted to the MEC server. This is because the size of the URLLC packet is small, so the communication module (17) can immediately transmit the authorized URLLC packet, and low-latency, high-speed processing is very important in URLLC services. That is, to suppress the increase in delay time caused by the temporary storage of URLLC packets, the offloaded URLLC packets can be processed after being temporarily stored in a server queue (not shown) provided in the MEC server. Then, the offloaded URLLC packet is transmitted to the MEC server via the communication module (17) using the RAT with the highest transmission efficiency selected by the MEC selection module (16) among a plurality of RATs.

[0045] Although the latency must be reduced as much as possible in the eMBB service, the constraints on latency are not as strict as those in the URLLC service, and because the eMBB packet is large, it is difficult for the communication module (17) to transmit the eMBB packet as is. Therefore, the eMBB queue (22) includes a RAT queue group (32) to ensure that eMBB packets that need to be offloaded are reliably transmitted to the MEC server through the communication module (17).

[0046] And the local queue (14) is equipped with a local URLLC queue and a local eMBB queue (31) so that the task processing module (15) can reliably process the generated URLLC packets and eMBB packets. This is so that task packets that need to be processed at user terminals (UE1, UE2) without being offloaded can be processed at the task processing module (15) without being lost.

[0047] The task processing module (15) receives and processes URLLC packets or eMBB packets stored in the local URLLC queue and the local eMBB queue (31) of the local queue (14). The task processing module (15) can be implemented as a local processor equipped in the user terminal (UE1, UE2), such as a CPU.

[0048] Meanwhile, the MEC selection module (16) selects one MEC server among multiple MEC servers to offload task packets based on channel state information (CSI) collected by the information collection module (12). That is, the MEC selection module (16) selects the MEC server equipped in the base station (BS) with the best channel state. Additionally, the MEC selection module (16) can identify the RAT with the highest data transmission rate according to the data transmission rate of each of the multiple RATs between the user terminal (UE1, UE2) and the base station (BS) equipped with the selected MEC server, and select it as the single RAT to transmit URLLC packets. As described above, multiple eMBB packets are transmitted to the MEC server of the base station (BS) through multiple RATs, and since they are already distributed and stored in multiple RAT queues (41 to 4k) corresponding to each of the multiple RATs in the local queue (14), a RAT for transmitting eMBB packets is not separately selected.

[0049] Here, for ease of understanding, the information collection module (12) and the MEC selection module (16) are shown separately, but the MEC selection module (16) may be configured to be included in the information collection module (12).

[0050] The communication module (17) transmits URLLC packets or eMBB packets authorized from the local queue (14) to the MEC server selected by the MEC selection module (16). In this case, for URLLC packets, the communication module (17) transmits the authorized URLLC packets to the MEC server through a single RAT selected by the MEC selection module (16). However, for eMBB packets, eMBB packets stored in multiple RAT queues (41 to 4k) are transmitted to the MEC server through the RAT corresponding to each RAT queue (41 to 4k). When the communication module (17) offloads the URLLC packets or eMBB packets to the MEC server using a RAT, it can offload them with the transmission power for each RAT determined by the offloading setting module (13).

[0051] Meanwhile, the server queue of the MEC server may also be separately equipped with a queue for storing URLLC packets and a queue for storing eMBB packets; however, in one embodiment, it is assumed that the MEC server operates in a process sharing mode, and accordingly, the server queue may be configured as a common queue independent of the service. As the MEC server operates in a process sharing mode, it ensures that its resources are always evenly distributed and utilized regardless of the frequency with which URLLC and eMBB packets are received. In other words, by distributing resources without distinguishing between types of services, the MEC server ensures that its resources are always fully utilized.

[0052] In one embodiment, the offloading configuration module (13) is configured such that when a user terminal (n, n = {1, ��, N}) of a multi-RAT-based MEC system offloads to a base station (m, m = {1, ��, M}) using a RAT (k, k = {1, ��, K}), it satisfies the QoS required for each service (ξ = {u, e}, where ξ is a service indicator representing a URLLC service (u) and an eMBB service (e)) while the energy (E) consumed by the user terminal (n) is satisfied. n ξ The RAT offloading ratio (o) that minimizes ) n,m,k ξ ) and RAT transmission power (p n,m,k ξ The purpose is to estimate ).

[0053] Accordingly, the objective function of the operation performed in the offloading setting module (13), which is composed of a reinforcement learning model, can be summarized as Equation 1.

[0054]

[0055] Here, N n,m,k ξ represents the number of resource blocks in the frequency domain for each RAT(k), and x n,m is a relationship indicator having a binary value of 1 if the user terminal (n) selects the MEC server (m) as an offloading target, and 0 if it does not select it. And f n ξ represents the local CPU frequency for the local processor of the user terminal (n).

[0056] Here, the number of resource blocks (N n,m,k ξ ) and relationship indicators(x n,m ) and local CPU frequency(f n ξ ) can be specified in advance.

[0057] And in mathematical formula 1, the local processing rate (o) that is locally processed in the task processing module (15) of the user terminal (n)n,m,0 ξ ) and multiple RAT offloading rate through multiple RATs (o n,m,k ξ The sum for ) is 1( It must become.

[0058] In this case, for the URLLC service, since only one of two cases can be specified where a single URLLC packet of a URLLC task is locally processed in the task processing module (15) of the user terminal (n) or offloaded to the MEC server (m) using a single RAT (k), the local processing rate (o) for the URLLC packet n,m,0 u ) and RAT offloading ratio(o n,m,k u ) is a constraint to have a value of 0 or 1( )becomes.

[0059] Also, in the case of a URLLC service, the local processing time (T) for processing URLLC packets at the user terminal (n) n u,L ) and offloading processing time (T) processed by offloading to the MEC server (m) n u,M URLLC service timeout (T) where ) is all set as QoS n u,max Constraint to be less than or equal to ) (T n u,L , T n u,M ≤ T n u,max )becomes.

[0060] On the other hand, in the case of an eMBB service, multiple eMBB packets are not only processed locally in the task processing module (15), but can also be offloaded to the MEC server (m) using multiple RATs (k) and processed together, so the local processing rate (o) for eMBB packets n,m,0 e ) and each RAT offloading ratio(o n,m,k e) is constrained to have a value between 0 and 1 ( )becomes.

[0061] And in the case of eMBB service, the local processing time (T) for processing eMBB packets at the user terminal (n) n e,L Offloading processing time (T) processed by offloading to the MEC server (m) via ) and multiple RATs n e,M The longest time (T) n e = max{T n e,L , T n e,M}) eMBB service timeout specified by QoS (T n e,max Constraint to be less than or equal to ) (T n e ≤ T n e,max )becomes.

[0062] Meanwhile, the energy consumption per bit for a URLLC packet in Equation 1 can be calculated as in Equation 2.

[0063]

[0064] Here, E n u,L represents local energy consumption due to local processing of the user terminal (n), and T n u,T θ is the packet transmission delay time between the user terminal (n) and the MEC server (m), T S is a time slot representing a unit transmission time, and b n u represents the URLLC packet size.

[0065] In mathematical equation 2, local energy consumption (E n u,L ) can be calculated as in mathematical formula 3.

[0066]

[0067] Here, en u,L represents the energy consumption per CPU cycle (J / cycle) of the user terminal (n), and c n u is URLLC packet processing CPU cycles, represents the energy consumption coefficient based on the processing performance of the CPU.

[0068] And the energy consumption per bit for an eMBB packet can be calculated as in Equation 4.

[0069]

[0070] Here, E n e,L represents local energy consumption due to local processing of the user terminal (n), and T n,m,k e,T θ is the packet transmission delay time through RAT(k) between the user terminal (n) and the MEC server (m), T S is a time slot representing a unit transmission time, and b n e represents the eMBB packet size.

[0071] In mathematical equation 4, local energy consumption (E n e,L ) can be calculated as in mathematical formula 5.

[0072]

[0073] Here, e n e,L represents the energy consumption per CPU cycle (J / cycle) of the user terminal (n), and c n e is eMBB packet processing CPU cycles, represents the energy consumption coefficient based on the processing performance of the CPU.

[0074] Accordingly, the offloading setting module (13) satisfies the above constraints and minimizes the energy consumption of the user terminal (n) calculated according to mathematical formulas 2 and 4, and the RAT offloading ratio (o n,m,k ξ ) and RAT transmission power (p n,m,k ξ It can be obtained by estimating ).

[0075] In addition, the offloading setting module (13) calculates the data transmission rate of each of the multiple RATs from the collected channel status information, and additionally applies limitations on the calculated data transmission rate, the queue workload of the MEC server's server queue, and the probability of loss of task packets to the RAT offloading rate (o n,m,k ξ ) and RAT transmission power (p n,m,k ξ ) can be estimated.

[0076] Consequently, the multi-RAT-based task offloading device of one embodiment transmits URLLC packets to be offloaded to the MEC server immediately without temporarily storing them in the local queue (14), and transmits eMBB packets to be offloaded to the MEC server after temporarily storing them in multiple RAT queues (41 to 4k) provided corresponding to each of the multiple RATs, and then transmits them to the MEC server through the respective RATs. Therefore, URLLC packets requiring low latency are immediately offloaded to the MEC server and processed without delay, whereas large volume multiple eMBB packets are distributed through various paths via multiple RATs and transmitted to the MEC server for processing. Thus, it is possible to improve performance and reduce energy consumption of the user terminal by maximizing resource utilization while satisfying different QoS requirements for each service.

[0077] In the illustrated embodiments, each component may have different functions and capabilities in addition to those described above and may include additional components not described. Additionally, in one embodiment, each component may be implemented using one or more physically separated devices, or by one or more processors or a combination of one or more processors and software, and may not be clearly distinguished in specific operation as in the illustrated examples.

[0078] And the task offloading device illustrated in FIG. 1 may be implemented in a logic circuit by hardware, firmware, software, or a combination thereof, or may be implemented using a general-purpose or specific-purpose computer. The device may be implemented using a hardwired device, a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), etc. Additionally, the device may be implemented as a system on chip (SoC) including one or more processors and controllers.

[0079] In addition, the task offloading device may be installed in a computing device or server equipped with hardware elements in the form of software, hardware, or a combination thereof. A computing device or server may refer to various devices that include, in whole or in part, communication devices such as communication modems for communicating with various devices or wired / wireless communication networks, memory for storing data for executing programs, and microprocessors for executing programs to perform calculations and commands.

[0080] FIG. 4 illustrates an offloading method of a user terminal according to an embodiment of the present invention.

[0081] An offloading method of one embodiment is performed at a user terminal of a multi-RAT-based MEC system. Referring to FIG. 4, the offloading method analyzes the tasks that need to be processed first (71). At this time, it is determined whether the task is a task for a URLLC service or a task for an eMBB service, and the size and number of task packets constituting each task can be determined. Here, the task may be a task generated by a user terminal (UE1, UE2), but is not limited thereto.

[0082] Along with this, various information that user terminals (UE1, UE2) need to check to adjust the offloading rate to the MEC server is collected and analyzed (72). The information collected here may include the QoS conditions required by the service of the analyzed task, the average packet arrival rate of the local queue (14), the local CPU frequency, the number of CPU cycles required, and channel state information (CSI) for each RAT of multiple RATs.

[0083] Afterward, it is determined whether the analyzed task is a URLLC task for the URLLC service (73). If it is determined to be a URLLC task, an offloading rate for the URLLC task is determined (74). Here, the offloading rate for the URLLC task can be determined by estimating it based on the collected information so that the energy consumption of the user terminal (UE1) is minimized within a range where the reinforcement learning model satisfies the constraints according to the QoS required in the URLLC service.

[0084] In the case of a URLLC task, since it consists of a single packet of small size, the offloading rate for the URLLC task is divided into two cases: whether the user terminal (UE1) processes it locally or offloads it to the MEC server for processing, and accordingly, the offloading rate can also be determined as 1 or 0. Once the URLLC offloading rate is determined, it is determined whether to offload the URLLC packet to the MEC server according to the determined URLLC offloading rate (75). If it is determined to offload to the MEC server, the URLLC packet is not temporarily stored in the local queue (14) but is immediately offloaded to the MEC server (76). At this time, the URLLC packet can be offloaded to the MEC server by selecting the RAT with the highest data transmission rate from the channel state information (CSI) according to each collected RAT.

[0085] On the other hand, if it is determined that the URLLC packet is not offloaded to the MEC server, the user terminal (UE1) must process the URLLC packet. Accordingly, the URLLC packet is stored in the local URLLC queue of the local queue (14) (77). Then, the local processor equipped in the user terminal (UE1) receives the URLLC packet stored in the local URLLC queue and processes it locally (78). Here, the local processor can receive and process the URLLC packet stored in the local URLLC queue sequentially according to the order in which they were stored.

[0086] Meanwhile, if the identified task is determined to be an eMBB task rather than a URLLC task, the RAT offloading ratio for the eMBB task is determined (79). Here, the offloading ratio for the eMBB task can also be determined by estimating it based on collected information so that the energy consumption of the user terminal (UE2) is minimized within the range where the reinforcement learning model satisfies the constraints according to the QoS required in the eMBB service. However, while the offloading ratio for the URLLC task is determined to be 0 or 1, the eMBB task consists of multiple eMBB packets with a large data size, so local processing and offloading to the MEC server must be performed together. In addition, when offloading to the MEC server, multiple eMBB packets can be offloaded through multiple paths via each of the multiple RATs. Accordingly, the offloading ratio for eMBB tasks can be estimated by dividing the RAT offloading ratio, in which multiple eMBB packets are offloaded through each of the multiple RATs, and the ratio to be locally processed is determined together such that the sum of all RAT offloading ratios and the local processing ratio to be locally processed is 1.

[0087] When the RAT offloading ratio and local processing ratio for each RAT are determined, multiple eMBB packets are divided and stored in multiple RAT queues (41 to 4k) and a local eMBB queue (31) according to the determined ratio (80). Then, the eMBB packets stored in the local eMBB queue (31) are authorized and processed locally by a local processor equipped in the user terminal (UE1), while the eMBB packets stored in each of the multiple RAT queues (41 to 4k) are offloaded to the MEC server through the RAT designated for each RAT queue (41 to 4k) (81).

[0088] Although FIG. 4 describes each process as being executed sequentially, this is merely an illustrative description, and a person skilled in the art can apply various modifications and variations by changing the order described in FIG. 4, executing one or more processes in parallel, or adding other processes, within the scope of not departing from the essential characteristics of the embodiment of the present invention.

[0089] FIG. 5 is a diagram illustrating a computing environment including a computing device according to one embodiment.

[0090] In the illustrated embodiments, each component may have different functions and capabilities in addition to those described below, and may include additional components in addition to those described below. The illustrated computing environment (90) may include a computing device (91) to perform the task offloading method illustrated in FIG. 5. In one embodiment, the computing device (91) may be one or more components included in the task offloading device illustrated in FIG. 2.

[0091] A computing device (91) includes at least one processor (92), a computer-readable storage medium (93), and a communication bus (95). The processor (92) may enable the computing device (91) to operate according to the exemplary embodiment described above. For example, the processor (92) may execute one or more programs (94) stored in the computer-readable storage medium (93). The one or more programs (94) may include one or more computer-executable instructions, and the computer-executable instructions may be configured to enable the computing device (91) to perform operations according to the exemplary embodiment when executed by the processor (92).

[0092] The communication bus (95) interconnects various other components of the computing device (91), including the processor (92) and the computer-readable storage medium (93).

[0093] The computing device (91) may also include one or more input / output interfaces (96) and one or more communication interfaces (97) that provide an interface for one or more input / output devices (98). The input / output interfaces (96) and communication interfaces (97) are connected to a communication bus (95). The input / output devices (98) may be connected to other components of the computing device (91) through the input / output interfaces (96). An exemplary input / output device (98) may include an input device such as a pointing device (such as a mouse or trackpad), a keyboard, a touch input device (such as a touchpad or touchscreen), a voice or sound input device, various types of sensor devices and / or imaging devices, and / or an output device such as a display device, a printer, a speaker and / or a network card. An exemplary input / output device (98) may be included inside the computing device (91) as a component constituting the computing device (91), or it may be connected to the computing device (91) as a separate device distinct from the computing device (91).

[0094] Although the present invention has been described in detail above through representative embodiments, those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of the present invention should be determined by the technical spirit of the appended claims.

Claims

Claim 1 A step of analyzing a task to be processed and collecting information including channel state information of each of multiple RATs (Radio Access Technology) as information for offloading the task; a step of performing neural network operations using a reinforcement learning model on the task and the collected information to estimate the offloading ratio to be transmitted to a MEC (Multi-access Edge Computing) server among at least one task packet constituting the task; a step of dividing the at least one task packet into a local processing packet and an offloading packet according to the offloading ratio; and if the task is a task for an URLLC (Ultra Reliability Low Latency Communication) service, a step of selecting one RAT among the multiple RATs with the highest data transmission rate according to the channel state information, and offloading the offloading packet to the MEC server through the selected RAT without storing it in a local queue. A task offloading method comprising the steps of: determining that the task is composed of multiple eMBB packets if the task is an enhanced Mobile Broadband (eMBB) service; estimating the offloading ratio to each of the multiple RATs according to the channel state information for each of the multiple RATs using a reinforcement learning model; distributing and storing the offloading packets in a number corresponding to the offloading ratio to each of the multiple RATs in multiple RAT queues of the local queue provided to correspond to each of the multiple RATs; and offloading the offloading packets stored in each of the multiple RAT queues to the MEC server through the RAT corresponding to each RAT queue. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 In claim 1, the task offloading method stores the local processing packets separated according to the offloading ratio in the local queue and processes the stored local processing packets sequentially. Claim 6 In claim 1, the step of estimating the offloading ratio is a task offloading method that estimates the offloading ratio such that the energy consumption of the user terminal offloading the task packet to the MEC server is minimized. Claim 7 In claim 1, the task offloading method is a task offloading method performed by the processor of a device comprising a memory and a processor. Claim 8 Memory; A device comprising a processor that executes at least a portion of an operation according to a program stored in the memory, wherein the processor analyzes a task to be processed and collects information including channel state information of each of multiple RATs (Radio Access Technology) as information for offloading the task; performs a neural network operation using a reinforcement learning model on the task and the collected information to estimate the offloading ratio to be transmitted to the MEC server among at least one task packet constituting the task; separates the at least one task packet into a local processing packet and an offloading packet according to the offloading ratio; if the task is a task for a URLLC (Ultra Reliability Low Latency Communication) service, selects one RAT with the highest data transmission rate among the multiple RATs according to the channel state information, and offloads the offloading packet to the MEC server through the selected RAT without storing it in a local queue; if the task is a task for an eMBB (enhanced Mobile Broadband) service, determines that the task is composed of multiple eMBB packets, and the reinforcement learning model according to the channel state information for each of the multiple RATs A task offloading device that estimates the offloading ratio to each of the multiple RATs, distributes and stores the offloading packets in a number corresponding to the offloading ratio to each of the multiple RATs in a plurality of RAT queues of the local queue provided to correspond to each of the multiple RATs, and performs the step of offloading the offloading packets stored in each of the plurality of RAT queues to the MEC server through the RAT corresponding to each RAT queue. Claim 9 delete Claim 10 delete Claim 11 delete Claim 12 In claim 8, the task offloading device wherein the processor stores the local processing packets separated according to the offloading ratio in the local queue and sequentially processes the stored local processing packets. Claim 13 In claim 8, the processor is a task offloading device that estimates the offloading ratio so that the energy consumption of the user terminal offloading the task packet to the MEC server is minimized.

Images