Energy-efficiency gain indication for offloading management

By estimating the uncertainty of energy efficiency gain indications using historical data, the network optimizes offloading decisions, ensuring reliable and efficient energy management in wireless communication networks.

WO2025140794A1PCT designated stage expired Publication Date: 2025-07-03TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
PCT/EP2024/059462
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-12-27
Filing Date
2024-04-08
Publication Date
2025-07-03

AI Technical Summary

Technical Problem

In wireless communication networks, there is a need for reliable techniques to manage offloading actions based on energy efficiency gain indications to avoid false or fraudulent estimates that can degrade network performance.

Method used

A first node in the network estimates the uncertainty of energy efficiency gain indications from a second node by using historical offloading data to train an ML model, allowing it to assess the reliability of proposed offloading actions and control the execution of tasks accordingly.

Benefits of technology

This approach ensures that offloading decisions are made based on trustworthy energy efficiency assessments, optimizing overall network energy efficiency and preventing degradation due to false or fraudulent indications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure EP2024059462_03072025_PF_FP_ABST
    Figure EP2024059462_03072025_PF_FP_ABST
Patent Text Reader

Abstract

A first node (101) of the wireless communication network obtains an indication of energy efficiency gain associated with offloading at least one task from a second node (102) of the wireless communication network to the first node (101). Based on data related to earlier offloading of tasks to the first node (101), the first node (101) estimates an uncertainty of the indication of energy efficiency gain. Based on the indication of energy efficiency gain and the estimated uncertainty, the first node (101) controls execution of the offloading of a set of one or more tasks from the second node (102) to the first node (101).
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Energy-efficiency gain indication for offloading management

[0002] Technical Field

[0003] The present invention relates to methods for managing a wireless communication network and to corresponding devices, systems, and computer programs.

[0004] Background

[0005] In wireless communication networks, e.g., based on the 4G (4th Generation) LTE (Long Term Evolution) or 5G (5th Generation) NR technology as specified by 3GPP (3rd Generation Partnership Project), there is an ongoing need for optimizations related to performance or efficiency. For example, the NG-RAN (Next Generation Radio Access Network) of the NR technology allows for utilizing a split architecture, in which functionalities of RAN nodes, denoted as gNBs, are distributed over multiple physical nodes. The split architecture is for example described in 3GPP TS 38.401 V17.6.0 (2023-09).

[0006] 3GPP RAN3 Study Item (SI) “Study on enhancement for data collection for NR and EN-DC” studied general high-level principles, a functional framework, and potential use cases for Al- enabled RAN, see 3GPP TR 37.817 V17.0.0 (2022-04). Objectives defined in related Work Item (Wl) “Artificial Intelligence (Al) / Machine Learning (ML) for NG-RAN”, 3GPP document RP- 220635, 3GPP TSG RAN Meeting #95e (Electronic Meeting, March 17-23, 2022), include specifying data collection enhancements and signaling support within existing NG-RAN (Next Generation Radio Access Network interfaces and architecture, including non-split architecture and split architecture, for AI / ML-based network energy saving, load balancing and mobility optimization. In relation to the AI / ML-based network energy saving, it was for example agreed to introduce the metric of Energy Cost (EC) as an AI / ML metric to be shared over the Xn interface among NG-RAN nodes (see “Report of 3GPP TSG RAN3 meeting #119”, 3GPP document R3-231101 , 3GPP TSG RAN3 meeting #119-bis-e).

[0007] The EC metric is a value at gNB level and can be an inferred energy consumption related to an additional load or an actual energy consumption value from a neighboring node for either additional load or current load. The additional load may refer to an offloading action where a certain number of UEs (User Equipments) served by the currently observed NG-RAN node, e.g., the RAN node proposing an action, such as an offloading action, are handed over to a neighboring NG-RAN node so they can be served there. An inferred EC related to an additional load would be a predicted EC assuming the offloading action will happen, i.e., the certain number of UEs will be transferred from the first NG-RAN node to the neighboring NG-RAN node.

[0008] The objective of the exchange of EC values is to allow an NG-RAN node to assess the energy impact of proposed actions, e.g., offloading UEs, in all affected RAN nodes and not just locally, both as an estimate before taking an action and as a measurement after the action is complete. In this way, AI / ML algorithms can be trained and employed such that the overall energy consumption of the network, e.g., the energy consumption of the local NG-RAN node and at least some of its direct neighboring NG-RAN nodes, can be reduced.

[0009] The EC metric could for example be used in a negotiation type interaction between NG-RAN nodes: NG-RAN node 1 proposes an action, e.g., offloading of a number of UEs, to NG-RAN node 2, along with a predicted EC or an energy efficiency (EE) score for NG-RAN node 1. NG RAN node 2 then decides if the action should be implemented or not by giving feedback to NG-RAN node 1. NG-RAN node 2 may initiate the process by requesting NG-RAN node 1 to report its EC metric and resource status. In such interaction, it could however happen that NG- RAN node 2 bases its decision whether to implement the proposed action on a false indication of the EC metric by NG-RAN node 1. For example, NG-RAN node 1 could make a wrong estimation of the EC metric due to problematic software, due limited data being available to make a reliable estimation, or intentionally. The latter case might be an attempt to coerce the NG-RAN node 2 to handle more traffic by offloading tasks.

[0010] Accordingly, there is a need for techniques which allow for using reported information on predicted energy impact of offloading actions in a reliable manner.

[0011] According to an embodiment, a method of controlling operation of a wireless communication network is provided. According to the method, a first node of the wireless communication network obtains an indication of energy efficiency gain associated with offloading at least one task from a second node of the wireless communication network to the first node. Based on data related to earlier offloading of tasks to the first node, the first node estimates an uncertainty of the indication of energy efficiency gain. Based on the indication of energy efficiency gain and the estimated uncertainty, the first node controls execution of the offloading of a set of one or more tasks from the second node to the first node. According to a further embodiment, a first node for a wireless communication network is provided. The first node is configured to obtain an indication of energy efficiency gain associated with offloading at least one task from a second node of the wireless communication network to the first node. Further, the first node is configured to, based on data related to earlier offloading of tasks to the first node, estimate an uncertainty of the indication of energy efficiency gain. Further, the first node is configured to, based on the indication of energy efficiency gain and the estimated uncertainty, control execution of the offloading of a set of one or more tasks from the second node to the first node.

[0012] According to a further embodiment, a first node for a wireless communication network is provided. The first node comprises at least one processor and a memory. The memory contains instructions executable by said at least one processor, whereby the first node is operative to obtain an indication of energy efficiency gain associated with offloading at least one task from a second node of the wireless communication network to the first node. Further, the memory contains instructions executable by said at least one processor, whereby the first node is operative to, based on data related to earlier offloading of tasks to the first node, estimate an uncertainty of the indication of energy efficiency gain. Further, the memory contains instructions executable by said at least one processor, whereby the first node is operative to, based on the indication of energy efficiency gain and the estimated uncertainty, control execution of the offloading of a set of one or more tasks from the second node to the first node.

[0013] According to a further embodiment of the invention, a computer program or computer program product is provided, e.g., in the form of a non-transitory storage medium, which comprises program code to be executed by at least one processor of a first node for a wireless communication network. Execution of the program code causes the first node to obtain an indication of energy efficiency gain associated with offloading at least one task from a second node of the wireless communication network to the first node. Further, execution of the program code causes the first node to, based on data related to earlier offloading of tasks to the first node, estimate an uncertainty of the indication of energy efficiency gain. Further, execution of the program code causes the first node to, based on the indication of energy efficiency gain and the estimated uncertainty, control execution of the offloading of a set of one or more tasks from the second node to the first node.

[0014] Details of such embodiments and further embodiments will be apparent from the following detailed description of embodiments. Brief of the

[0015] Fig. 1 schematically illustrates a wireless communication network according to an embodiment of the present disclosure.

[0016] Fig. 2 schematically illustrates a 5G network architecture in which offloading may be controlled according to an embodiment of the present disclosure.

[0017] Fig. 3 schematically illustrates a distributed architecture of a radio access node in which offloading may be controlled according to an embodiment of the present disclosure.

[0018] Fig. 4A schematically illustrates a training phase of an ML model for controlling offloading in accordance with an embodiment of the present disclosure.

[0019] Fig. 4B schematically illustrates an inference phase for controlling offloading in accordance with an embodiment of the present disclosure.

[0020] Fig. 5 schematically illustrates an offloading procedure in accordance with an embodiment of the present disclosure.

[0021] Fig. 6 shows a flowchart for schematically illustrating a method according to an embodiment of the present disclosure.

[0022] Fig. 7 schematically illustrates structures of a network node according to an embodiment of the present disclosure.

[0023] Detailed Description

[0024] In the following, concepts in accordance with exemplary embodiments of the invention will be explained in more detail and with reference to the accompanying drawings. The illustrated embodiments relate to controlling operation of a wireless communication network, in particular with respect to offloading of tasks among nodes of the wireless communication network. The wireless communication network may be based on the 5G NR technology specified by 3GPP. However, other technologies could be used as well, e.g., the 4G LTE technology specified by 3GPP or a future 6G (6thGeneration) technology. In the illustrated concepts, the wireless communication network is assumed to include multiple nodes. As further detailed below, such nodes may be RAN (Radio Access Network) nodes, e.g., gNBs of the NR technology. In a split architecture, such nodes could also be physical nodes implementing a certain RAN node, e.g., gNB of the NR technology. Tasks may be offloaded between the nodes. For example, such task could correspond to or be associated with serving of a number of UEs. The offloading of the task(s) could then involve handover the UEs from one node to the other node. Execution of the offloading is decided by one of the nodes, based on an expected energy efficiency (EE) gain indicated by the other node. Such EE gain could for example be indicated in terms of the EC (Energy Cost) metric agreed in the context of 3GPP.

[0025] Specifically, a first node may receive, from a second node, a proposal or request to offload at least one task from the second node to the first node, e.g., to handover one or more UEs from the second node to the first node. The result would be a reduced energy consumption at the second node, but an increased energy consumption at the first node. Overall energy efficiency, considering both the first node and the second node, may however still increase, which may be reflected in the EE gain indicated by the second node. If the indicated EE gain is sufficiently high, the first node may decide to perform the offloading, thereby improving energy efficiency. The indicated EE gain may thus be regarded as an incentive to accept the offloading proposed or requested by the second node. When deciding on whether to perform the offloading, the first node further considers uncertainty of the EE gain indicated by the second node. To this end, the first node considers data related to earlier offloading of tasks to the first node, in the following also denoted as historic offloading data. Such earlier offloading may for example include earlier offloading of tasks from the second node to the first node, but offloading of tasks from other nodes to the first node could be considered in addition or as an alternative.

[0026] The historic offloading data may for example be used to train an ML (Machine Learning) model to estimate the uncertainty of the EE gain. The historic offloading data could for example indicate the actual EE gain for each of a plurality of earlier offloading scenarios and related resource status data of the first node, and the ML model may be trained to predict the EE gain from the offloading scenario and its current resource status. The uncertainty may then be derived based on the deviation of the predicted EE gain from the indicated EE gain. The ML model could for example be a classification model. Alternatively, a Monte Carlo dropout model or a neural network could be used.

[0027] Accordingly, in the illustrated concepts, historic offloading data, representing earlier offloading scenarios and their effect on EE, may be collected and be used to train an ML model which estimates EE gain of future offloading tasks, thereby also allowing to assess uncertainty of an EE gain indicated by another node. In this way, it can be avoided that offloading actions are performed based on a faulty or fraudulent EE gain indicated by another node.

[0028] Fig. 1 illustrates exemplary structures of the wireless communication network. In particular, Fig. 1 shows UEs 10 which are served by access nodes 100 of the wireless communication network. Here, it is noted that the wireless communication network may actually include a plurality of access nodes 100 that may serve one or more sectors within the coverage area of the wireless communication network. Each sector may in turn include one or more cells. Cells within the same sector and cells of neighboring sectors are typically operated on different frequencies, i.e., on different carriers, so that interference among cells can be avoided. It is however noted that for cells which are sufficiently separated in space, the same frequency could be reused. The access nodes 100 could for example correspond to eNBs of the LTE technology, gNBs of the NR technology, or to similar access nodes of some other technology, e.g., of a 6G technology. It is also noted that access nodes of different technologies could be deployed in parallel.

[0029] The access nodes 100 may be regarded as being part of an RAN of the wireless communication network. Further, Fig. 1 schematically illustrates a CN (Core Network) 110 of the wireless communication network. In Fig. 1 , the CN 110 is illustrated as including a GW (gateway) 120 and one or more control node(s) 240. The GW 120 may be responsible for handling user plane data traffic of the UEs 10, e.g., by forwarding user plane data traffic from a UE 10 to a network destination or by forwarding user plane data traffic from a network source to a UE 10. Here, the network destination may correspond to another UE 10, to an internal node of the wireless communication network, or to an external node which is connected to the wireless communication network. Similarly, the network source may correspond to another UE 10, to an internal node of the wireless communication network, or to an external node which is connected to the wireless communication network. The GW 120 may for example correspond to a UPF (User Plane Function) of the 5G Core (EGC) or to an SGW (Serving Gateway) or PGW (Packet Data Gateway) of the 4G EPC (Evolved Packet Core). The control node(s) 240 may for example be used for controlling the user data traffic, e.g., by providing control data to the access nodes 100, the GW 120, and / or to the UE 10.

[0030] As illustrated by solid double-headed arrows, the access nodes 100 may send DL wireless transmissions to at least some of the UEs 10, and some of the UEs 10 may send UL wireless transmissions to the access nodes 100. The DL transmissions and UL transmissions may be used to provide various kinds of services to the UEs 10, e.g., a voice service, a multimedia service, or some other data service. Such services may be hosted in the CN 110, e.g., by a corresponding network node. By way of example, Fig. 1 illustrates an application service platform 150 provided in the CN 110. Further, such services may be hosted externally, e.g., by an AF (application function) connected to the CN 110. By way of example, Fig. 1 illustrates one or more application servers 160 connected to the CN 110. The application server(s) 160 could for example connect through the Internet or some other wide area communication network to the CN 110. The application service platform 150 may be based on a server or a cloud computing system and be hosted by one or more host computers. Similarly, the application server(s) 160 may be based on a server or a cloud computing system and be hosted by one or more host computers. The application server(s) 160 may include or be associated with one or more AFs that enable interaction with the CN 110 to provide one or more services to the UEs 10, corresponding to one or more applications. These services or applications may generate the user data traffic conveyed by the DL transmissions and / or the UL transmissions between the access node 100 and the UE 10. Accordingly, the application server(s) 160 may include or correspond to the above- mentioned network destination and / or network source for the user data traffic. In the respective UE 10, such service may be based on an application (or shortly “app”) which is executed on the UE 10. Such application may be pre-installed or installed by the user. Such application may generate at least a part of the user plane data traffic between the UEs 10 and the access nodes 100.

[0031] In the illustrated concepts, energy efficiency of the wireless communication network could for example be improved by handover of one or more UEs 10 from one of the access nodes 100 to the other. As outlined above, the decision whether to perform such offloading of tasks from one access node 100 to the other may be taken at the access node 100 to which the tasks are being offloaded, in the following also denoted as “offloading target”. However, the decision is based on energy efficiency related information from the access node 100 from which the tasks are being offloaded, in the following also denoted as “offloading source”: The information specially includes an indication of EE gain associated with the offloading. In the following examples, the EE gain can be indicated in terms of the EC metric agreed by 3GPP, but other forms of indicating the EE gain could be used as well.

[0032] The access node 100 which is the offloading target may base the offloading decision on the EC value indicated by the access node 100 which is the offloading source, but further take into account the uncertainty which is estimated from the historic offloading data. For this estimation, the offloading target may consider the historic offloading data collected by itself, which may increase trustworthiness of the estimation. In other words, the historic offloading data considered by the offloading target may be a dataset based on earlier offloading to this offloading target, and the impact of the earlier offloading may be measured by the offloading target itself. The estimation thus does not rely on information provided by other nodes.

[0033] With the historic offloading data collected by the offloading target itself, the estimation of impact of a future offloading task may be handled as a classification problem, and existing ML algorithms may be applied to measure uncertainty of the classification. Based on the uncertainty, the offloading target may decide how to handle the offloading proposed or requested by the offloading source, e.g., whether to fully accept the offloading, whether to partially accept the offloading, or whether to (fully) reject the offloading.

[0034] Fig. 2 further illustrates a possible architecture of the wireless communication network, in accordance with the principles given in 3GPP TS 38.300 V17.6.0 (2023-09). In the example of Fig. 2, the architecture includes a 5GC (“5G Core”) 210, which may implement the CN 110 of Fig. 1 , and a NG-RAN 220. The NG-RAN 220 includes a number of gNBs 200. The gNBs 200 are connected by an interface denoted as “Xn-C”. The interface for connecting the gNBs 200 to the 5GC 210 is denoted as “NG”. In some cases, the NG-RAN 220 could also include a set of of ng-eNBs, which correspond to an enhanced version of the eNB of the LTE technology. The gNBs 200 (or ng-eNBs) may implement the access nodes 100 of Fig. 1 .

[0035] In accordance with the split architecture specified in 3GPP TS 38.401 V17.6.0, a gNB 200 may be split into a gNB-Cll (gNB Central Unit) 201 and at least one gNB-DU (gNB Distributed Unit) 203. The gNB-CU 201 and the gNB-DU(s) 203 are connected via an interface denoted as “F1”, with each gNB-DU 203 being connected to only one gNB-CU 201 . The NG and Xn-C interfaces for a gNB 200 consisting of a gNB-CU 201 and gNB-DUs 203 terminate in the gNB-CU 201. The gNB-CU 201 and connected gNB-DUs 203 are visible to other gNBs 200 and the 5GC 210 as a single gNB.

[0036] Fig. 3 schematically illustrates an architecture for CP (Control Plane) and UP (User Plane) separation of the gNB-CU. In this case, a gNB 300 may consist of a gNB-CU-CP 301 , multiple gNB-CU-Ups 302, and multiple gNB-DUs 303. The gNB-CU-CP 301 is connected to the gNB- DU 303 through an interface denoted as “F1-C”. The gNB-CU-UP 302 is connected to the gNB-DU 303 through an interface denoted as “F1-U”. The gNB-CU-UP 302 is connected to the gNB-CU-CP 301 through an interface denoted as “E1”. As illustrated, one gNB-DU 303 is connected to only one gNB-CU-CP 301 , and one gNB-CU-UP 302 is connected to only one gNB-CU-CP 301. In a similar manner, an ng-eNB could consist of an ng-eNB-CU and one or more ng-eNB-DU(s). An ng-eNB-CU and an ng-eNB-DU are connected via an interface denoted as “W1”.

[0037] In the following, a more specific example of implementing the illustrated concepts is described with reference to Figs. 4A and 4B. In this example, offloading of tasks is controlled for two nodes, denoted as “NG_RAN_Node_1” and “NG_RAN_Node_2”. Here, it is noted that these nodes could correspond to any type of RAN node or functional entities of a RAN node. For example, the nodes could correspond to two gNBs or to a a gNB-Cll and a gNB-Dll withing the same gNB. The tasks being offloaded are assumed to correspond to serving of UEs. An offloading action may thus also be termed as “offloading of UEs”. The process underlying the example of Figs. 4A and 4B is organized in two phases: a training phase and an interference phase.

[0038] The training phase is illustrated by Fig. 4A. The goal of the training phase is to build a model that estimates the class of energy efficiency for the given point in time and the number of UEs. In the following, such model is also denoted as EE model. The EE model is an AI / ML model trained based on data related to past offloading actions. Each node is expected to maintain such EE model, based on its own inputs. Alternatively or in addition, the EE model can be maintained in by some other node, e.g., a CN node, such as an GAM (Operations, Administration, and Maintenance) node, OSS (Operations Support System) node, or SMO (Service Management and Orchestration) node.

[0039] In the example of Fig. 4A, NG_RAN_Node_2 first sends an AI / ML assistance data update 401 to NG_RAN_Node_1 . The AI / ML assistance data update may include a report on current energy efficiency and resource status of NG_RAN_Node_2. Based on this report, the EE model of NG_RAN_Node_1 is used to infer an offloading action at block 402. In this example the offloading action corresponds to offloading of n UEs from NG_RAN_Node_1 to NG_RAN_Node_2, which is predicted to improve energy efficiency (in the illustrated example measured as EE score) of NG_RAN_Node_1. The EE model of NG_RAN_Node_1 predicts that the EE score of NG_RAN_Node_1 changes by a value of X, i.e. , an EE gain of X.

[0040] NG_RAN_Node_1 then proposes the offloading action to NG_RAN_Node_2, by sending an EE action indication 403 to NG_RAN_Node_2. The EE action indication includes the proposed action (namely offloading of n UEs) and the predicted EE gain of X. Based on this proposal, the EE model of NG_RAN_Node_2 is used to assess the proposed offloading action at block 404. The EE model of NG_RAN_Node_1 predicts that the EE score of NG_RAN_Node_2 changes by a value of Y, i.e., an EE gain of Y.

[0041] NG_RAN_Node_1 and NG_RAN_Node_2 then implement the offloading action, as indicated by block 405, i.e., perform a handover of n UEs from NG_RAN_Node_1 to NG_RAN_Node_2.

[0042] NG_RAN_Node_2 may further provide feedback 406 to NG_RAN_Node_1. The feedback 406 may for example indicate the actual energy efficiency at NG_RAN_Node_2 after performing the offloading action.

[0043] Based on the data related to the performed offloading action, NG_RAN_Node_1 then trains its EE model, as indicated by block 407. Here, the training target is to match the prediction by the EE model to the actually observed energy efficiency after the offloading action has been performed. Similarly, NG_RAN_Node_2 trains its EE model, as indicated by block 408. Also here, the training target is to match the prediction by the EE model to the actually observed energy efficiency after the offloading action has been performed.

[0044] As input to train the EE model, timestamped data from historical offloading actions are used. These historical offloading actions relate also to offloading actions where the node was the offloading target. For each offloading action, the timestamped data also include an indication of the actually achieved energy efficiency upon completion of the offloading action.

[0045] In the example of Fig. 4A, it is assumed that the nodes operate correctly and each can properly measure its own energy efficiency, e.g., in terms of EE score, EC, or EE index, after having completed the offloading action. Here it is noted also noted that the training is based on the energy efficiency predicted and measured at the node itself, and not on the predicted values or feedback from other nodes, which could include faulty or fraudulent values. But is noted that in some scenarios also feedback like provided by message 406 could be considered in the training.

[0046] The estimation of the energy efficiency is treated as a classification problem where the value representing the energy efficiency, e.g., EE score, EC, or EE index is classified using a number of discretized bins. For example when the value representing the energy efficiency can have a value from a range of 0 to 100, 10 bins could be provided bins, where x1 = [0..10] , x2 = [10, 20], ... x10 = [90..100], The EE model may be based on using the “softmax” function in the output layer of the classification model, which for every prediction also yields a measure of its uncertainty. When for example assuming the above 10 bins for classification, for class x1 , certainty u1 = 10, for class x3, uncertainty u2 = 10, ... , and for class x10, uncertainty u10 =80, where u1+... +u10 = 100. As mentioned above, such EE model is maintained for every node, either locally in that node, or in some other node, e.g., an OAM node, OSS node, or SMO node.

[0047] Once the EE model is trained, it can be used in the inference phase to assess offloading actions proposed by other nodes. In the example of Fig. 4B, it is assumed that NG_RAN_node_1 proposes an offloading action to NG_RAN_node 2, and NG_RAN_node 2 uses its EE model to assess uncertainty of predicted EE score indicated by NG_RAN_node_1 .

[0048] In the inference phase, as illustrated by Fig. 4B, NG_RAN_Node_2 sends an AI / ML assistance data update 401 to NG_RAN_Node_1. The AI / ML assistance data update 411 includes a report on current energy efficiency and resource status of NG_RAN_Node_2. Based on this report, the EE model of NG_RAN_Node_1 is used to infer an offloading action, as indicated by block 412. In this example, the offloading action corresponding to offloading of n UEs from NG_RAN_Node_1 to NG_RAN_Node_2, which is predicted to improve energy efficiency (in the illustrated example measured as EE score) of NG_RAN_Node_1. The EE model of NG_RAN_Node_1 predicts that the EE score changes by a value of X, i.e. , predicts an EE gain of X.

[0049] NG_RAN_Node_1 then proposes the offloading action to NG_RAN_Node_2, by sending an EE action indication 413 to NG_RAN_Node_2. The EE action 413 indication includes the proposed action (namely offloading of n UEs) and the predicted EE gain of X.

[0050] Based on this proposal, NG_RAN_Node_2 applies its own EE model to predict the EE efficiency of NG_RAN_Node_2, as indicated by block 414. The EE model of NG_RAN_Node_2 predicts that the EE score changes by a value of Y, with uncertainty U. In the illustrated example, X-Y is assumed to be positive, and based on the corresponding overall EE gain and the related uncertainty, NG_RAN_Node_2 thus accepts the proposed offloading action.

[0051] In some cases, NG_RAN_Node_2 could discount the number of the UEs based on predictions for different offloading scenarios differing in the number of UEs to be offloaded. For example, NG_RAN_Node_2 could rank the uncertainties of the different offloading scenarios and decide to perform the offloading action based on the offloading scenario with the lowest uncertainty of the prediction.

[0052] In accordance with the accepted offloading action, NG_RAN_Node_1 and NG_RAN_Node_2 then implement the offloading action and perform the handover of the n UEs, as indicated by block 415.

[0053] Further, NG_RAN_Node_2 may provide feedback 416 to NG_RAN_Node_1. The feedback 416 could for example indicate the actual energy efficiency at NG_RAN_Node_2 after performing the offloading action.

[0054] In the following, the terms “node 1” and “node 2” are used to more generically represent the nodes involved in the offloading. With respect to the example of Figs. 4A and 4B, “node 1” may correspond to NG_RAN_Node_1 , and “node 2” may correspond to NG_RAN_Node_2.

[0055] In some scenarios, once node 1 determines an offloading action and signals it in a first message to node 2, together with a prediction of EE gain, node 2 may estimate the uncertainty of the predicted EE gain at node 1 and derives an offloading action that would maximize EE gain at node 1 and node 2 together. Node 2 may then reply to node 1 with a second message indicating the derived offloading action is, e.g. to offload a given number of UEs or a percentage of the total traffic from one or more cells served by node 1 to one or more cells served by node 2, as well as the EE gains calculated for node 1 by node 2, for each of such offloading actions or for the overall set of offloading actions. The second message may also inform node 1 of the estimate on the EE gain uncertainty at node 1 , as obtained by node 2. The second message may inform node 1 about preferred offloading action(s) according to the estimation by node 2.

[0056] In some scenarios, node 1 could send a third message after the second message. The third message could include the actual EE gain of node 1 in response to performing the offloading as suggested in the second message. Based on the third message, node 2 can check the accuracy of its uncertainty evaluation for the predictions made by node 1 and therefore improve such accuracy in the future.

[0057] As can be seen, in the illustrated a node can measure the uncertainty of a received EE gain indication by using its own EE model and information, without revealing anything proprietary to other nodes nor revealing any internal implementation of its model nor the model that it is using to measure the uncertainty. Further, a node can mitigate a proposed offloading action by discounting it into smaller subactions. These subactions may then be assessed and individually and be performed in order of increasing uncertainty. In this way, the implemented offloading decision may converge towards the original proposal. But in some cases, only part of the original proposal could be implemented. Further, the nodes involved in negotiation of offloading actions based on the illustrated concepts may be enabled to take better-informed energy saving decisions based on vetting of the estimated EE gains at different nodes and for specific offloading actions carried out by more than one node. As a result, offloading actions may be optimized in view of overall energy efficiency.

[0058] In the illustrated concepts, it can also be considered that actions as predicted or performed by a node can affect UEs that can be grouped or classified according to various criteria. For example, the predicted EE gain of an offloading action can be distinguished based on the fact that some UEs of the affected UEs in a group of n UEs run a certain service or a certain set of services, or based on at least one QoS (Quality of Service) attribute, or based on at least one NSSAI (Network Slice Selection Assistance Information) to which one or more of the UEs is associated. Service categories which may be considered could for example include URLLC (Ultra Reliable Low Latency Communication), eMBB (enhanced Mobile Broadband), or MTC (Machine Type Communication). For example, the EE model of node 1 , during the training phase, or during the inference phase, could infer a first offloading action for a group of n UEs with services mapped to NSSAI- ’URLLC” and generating a predicted EC score change of “X1” and a second offloading action for a group of m UEs with services mapped to NSSAI-’eMBB” and generating a predicted EC score change of “X2”.

[0059] Similarly, during the inference phase, node 2 could determine whether to accept the offloading action proposed by node 1 using a first predicted EC score change obtained for UEs with services mapped to NSSAI-’URLLC” and a second predicted EC score changed obtained for UEs with services mapped to NSSAI- ’eMBB”.

[0060] In the EE action indication 413 of Fig. 4B, NG_RAN_Node_1 sent a value “X” to NG_RAN_Node_2, and this value was used by NG_RAN_Node_2 to decide whether to accept the proposed offloading action. In a variant, such indication of the EE gain by node 1 could also be omitted or disregarded, and node 2 could determine whether the offloading can be accepted (totally or partially) based on other criteria. For example, node 2 could consider only the EE gain predicted by its own EE model, e.g., the value “Y” in the example of Fig. 4B. The EE gain predicted by the EE model of node 2 could for example be compared to a first threshold and the proposal accepted if the predicted value is above the first threshold. Or the sum of the current energy efficiency, e.g., in terms of energy score, and the predicted EE gain could be compared to a second threshold and the proposal accepted if the predicted value is above the second threshold.

[0061] Further, certain types of services or categories of UE, e.g., UEs running voice services or UEs running emergency calls, may be exempted from the estimation of the predicted EE gain.

[0062] Further, certain types of services, or categories of llE, e.g., UEs running voice services or UEs running emergency calls, may be considered by always accepting the offloading for these types of services or categories of UEs.

[0063] In some scenarios, node 2 could derive the predicted EE gain based on a mapping, which can be configurable, of the number n of UEs which the node 1 proposed to offload to EE gain at node 1 .

[0064] Fig. 5 shows an example of an offloading procedure which is based on the illustrated concepts. The method involves a first node 101 and a second node 102. The first node 101 may correspond to node 2 of the above examples and the second node 102 may correspond to node 1 of the above examples.

[0065] In the example of Fig. 5, the second node sends an action indication 501 to the first node 101 . The action indication 501 may propose or request offloading of one or more tasks from the second node 102 to the first node 101. The action indication further includes an indication of EE gain associated with the proposed offloading. In the illustrated example, the indication of EE gain is based on the EC metric, i.e. , corresponds to an EC value.

[0066] At block 502, the first node 101 estimates the uncertainty of the EC metric indicated by the second node. This estimation may be based on an ML model maintained by the first node 101 , such as the above-mentioned EE model. The ML model maintained by the first node 101 may be trained on data related to earlier offloading of tasks to the first node 101 .

[0067] At block 503, the first node 101 decides to accept the proposed offloading. This decision is based on the uncertainty estimated at block 502.

[0068] The first node 101 then sends a response 504 indicating the acceptance of the offloading to the second node 102, and the offloading is then performed by the nodes 101 , 102, as indicated by block 505. Fig. 6 shows a flowchart for illustrating a method, which may be utilized for implementing the illustrated concepts. More specifically, the method may be used to implement the above- mentioned functionalities for controlling offloading actions. The method of Fig. 6 may be used for implementing the illustrated concepts in a first node of the wireless communication network, which may accept offloading of one or more tasks from a second node of the wireless communication network. The first node and the second node could for example correspond to RAN nodes, such as the above-mentioned access nodes 100 or gNBs 200, 300. Further, the first node and second node could also correspond to subentities of a RAN node, e.g., to gNB CUs or to gNB Dlls. The tasks which might be offloaded may for example correspond to serving of one or more user devices, such as the above-mentioned UEs 10. The serving of the user devices may for example involve mobility management of the user devices, radio resource management of the user devices, processing of radio signals to and / or from the user devices, connection management of the user device, processing of data packets to and / or from the user devices, baseband processing of signals to and / or from the users devices, handling of security of the user devices, QoS management for the user devices, and / or charging related tasks for the user devices.

[0069] If a processor-based implementation of the first node is used, at least some of the steps of the method of Fig. 6 may be performed and / or controlled by one or more processors of the first node. Such first node may also include a memory storing program code for implementing at least some of the below described functionalities or steps of the method of Fig. 6.

[0070] At step 610, the first node obtains an indication of EE gain associated with offloading at least one task from the second node to the first node. In some scenarios, the first node may receive the indication of EE gain from the second node, e.g., like with message 413 of Fig. 4A. Alternatively or in addition, the first node could estimate the indication of EE gain, e.g., using an ML model maintained by the first node.

[0071] At step 620, the first node estimates an uncertainty of the indication of EE gain obtained at step 610. The estimation is based on data related to earlier offloading of tasks to the first node. Such earlier offloading may also include offloading of one or more tasks from the second node to the first node, but is not limited to offloading of tasks from the second node to the first node, i.e., could more generally include offloading of tasks from one or more other nodes of the wireless communication network to the first node.

[0072] In some scenarios, the first node may estimate the uncertainty based on an ML model which is trained based on the data related to earlier offloading of tasks from one or more other nodes of the wireless communication network to the first node, e.g., as explained in connection with Fig. 4A.

[0073] In some scenarios, the data related to the earlier offloading of tasks may include feedback from each of the one or more other nodes, the feedback indicating energy efficiency of the other node after offloading of at least one task to the first node, such as the feedback indicated by message 406 of Fig. 4A.

[0074] The ML model may be based on classification of the energy efficiency gain into a plurality of classes and the uncertainty may then be based on distribution of energy efficiency gains over the classes. Alternatively or in addition, the ML model could be based on a Monte Carlo dropout model or on a neural network.

[0075] At step 630, the first node controls execution of the offloading of a set of one or more tasks from the second node to the first node. This control is based on the indication of EE gain and the estimated uncertainty.

[0076] In some scenarios, step 630 may involve that, based on the estimated uncertainty value, the first node decides between accepting the offloading for the complete set of one or more tasks, accepting the offloading for only a subset of the set of one or more tasks, and rejecting the offloading for the complete set of one or more tasks. When accepting the offloading for only a subset of the set of one or more tasks, the first node could select the subset based on the uncertainty value.

[0077] In some scenarios, the controlling of execution of the offloading of the set of one or more tasks from the second node to the first node of step 630 may further be based on one or more types of service associated with the one or more tasks. The one or more types of service could for example include: a type of service corresponding to a URLLC service category, a type of service corresponding to an eMBB service category, and / or a type of service corresponding to an MTC service category. Further, the controlling of execution of the offloading of the set of one or more tasks from the second node to the first node of step 630 could be based on one or more categories of user devices associated with the one or more tasks, e.g., UE categories like MTC UE, eMBB UE, or URLLC UE.

[0078] After offloading at least a subset of the set of one or more tasks from the second node to the first node, the first node may provide feedback to the second node. The feedback may indicate energy efficiency of the first node after the offloading. Message 416 is an example of such feedback.

[0079] Fig. 7 illustrates a processor-based implementation of a node 700 for a wireless communication network, which may be used for implementing the above-described concepts. Structures like illustrated in Fig. 7 could for example be used to implement the above- mentioned first node.

[0080] As illustrated, the node 700 may include one or more interfaces 710. The interface(s) 710 may for example be used for communicating with other nodes of the wireless communication network.

[0081] Further, the node 700 may include one or more processors 750 coupled to the interface(s) 710 and a memory 760 coupled to the processor(s) 750. By way of example, the interface(s) 710, the processor(s) 750, and the memory 760 could be coupled by one or more internal bus systems of the node 700. The memory 760 may include a read-only memory (ROM), e.g., a flash ROM, a random-access memory (RAM), e.g., a dynamic RAM (DRAM) or static RAM (SRAM), a mass storage, e.g., a hard disk or solid state disk, or the like. As illustrated, the memory 760 may include software 770 and / or firmware 780. The memory 760 may include suitably configured program code to be executed by the processor(s) 750 so as to implement the above-described functionalities for controlling offloading of tasks among nodes, e.g., in accordance with the method of Fig. 6.

[0082] It is to be understood that the structures as illustrated in Fig. 7 are merely schematic and that the node 700 may actually include further components which, for the sake of clarity, have not been illustrated, e.g., further interfaces or further processors. Also, it is to be understood that the memory 760 may include further program code for implementing known functionalities of RAN nodes of a wireless communication network. According to some embodiments, also a computer program may be provided for implementing functionalities of the node 700, e.g., in the form of a physical medium storing the program code and / or other data to be stored in the memory 760 or by making the program code available for download or by streaming. Further, it is noted that in some scenarios multiple nodes 700 with structures as illustrated in Fig. 7 could be used in combination, e.g., as a cloud system, to implement the above-described functionalities for controlling offloading of tasks among nodes.

[0083] As can be seen, the concepts as described above may be used for efficiently controlling offloading of tasks based on an indication of EE gain. In particular, it can be avoided that a false or fraudulent indication of EE gain by some other node causes unreasonable acceptance of offloading tasks, which might lead to degradation of overall performance of the wireless communication network.

[0084] It is to be understood that the examples and embodiments as explained above are merely illustrative and susceptible to various modifications. For example, the illustrated concepts may be applied in connection with various kinds of wireless communication technologies. Further, the tasks being offloaded could be of various nature and could for example also include management tasks which are not directly related to serving specific user devices. Further, while the above examples referred to an indication of EE gain, which could be based on the EC metric agreed by 3GPP, other forms of similar indications related to power or energy could be used as well, for example “energy consumption”, “energy consumption score”, “energy efficiency index”, “energy consumption class”, “energy efficiency class”, “energy saving score”, “energy gain score”, “power consumption”, “power saving”, “gain in power saving”, “power score”, “power index”, or the like.

[0085] Moreover, it is to be understood that the above concepts may be implemented by using correspondingly designed software to be executed by one or more processors of an existing device or apparatus, or by using dedicated device hardware. Further, it should be noted that the illustrated apparatuses or devices may each be implemented as a single device or as a system of multiple interacting devices or modules.

Claims

Claims1. A method of managing a wireless communication network, the method comprising: a first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) of the wireless communication network obtaining an indication of energy efficiency gain associated with offloading at least one task from a second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) of the wireless communication network to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700); based on data related to earlier offloading of tasks to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700), the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) estimating an uncertainty of the indication of energy efficiency gain; and based on the indication of energy efficiency gain and the estimated uncertainty, the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) controlling execution of the offloading of a set of one or more tasks from the second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700).

2. The method according to claim 1 , wherein the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) estimates the uncertainty based on a machine learning, ML, model which is trained based on the data related to earlier offloading of tasks to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700).

3. The method according to claim 2, wherein the data related to the earlier offloading of tasks comprise feedback (406) from each of the one or more other nodes (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700), the feedback indicating energy efficiency of the other node after offloading of at least one task to the first node.

4. The method according to claim 2 or 3, wherein the ML model is based on classification of the energy efficiency gain into a plurality of classes and the uncertainty is based on distribution of energy efficiency gains over the classes.

5. The method according to claim 2 or 3, wherein the ML model is based on a Monte Carlo dropout model.

6. The method according to any of claims 2 to 5, wherein the ML model is based on a neural network.

7. The method according to any of claims 1 to 6, comprising: the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) receiving the indication of energy efficiency gain from the second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700).

8. The method according to any of claims 1 to 5, comprising: the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) estimating the indication of energy efficiency gain.

9. The method according to any of the preceding claims, wherein controlling execution of the offloading of the set of one or more tasks from the second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) comprises, based on the estimated uncertainty value, deciding between accepting the offloading for the complete set of one or more tasks, accepting the offloading for only a subset of the set of one or more tasks, and rejecting the offloading for the complete set of one or more tasks.

10. The method according to claim 9, comprising: when accepting the offloading for only a subset of the set of one or more tasks, the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) selecting the subset based on the uncertainty value.

11. The method according to any of the preceding claims, comprising: wherein controlling execution of the offloading of the set of one or more tasks from the second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) is further based on one or more types of service associated with the one or more tasks.

12. The method according to claim 11 , wherein the one or more types of service comprise a type of service corresponding to a Ultra Reliable Low Latency Communication, URLLC, service category, a type of service corresponding to an enhanced Mobile Broadband, eMBB, service category, and / or a type of service corresponding to a Machine Type Communication, MTC, service category.

13. The method according to any of the preceding claims, comprising:wherein controlling execution of the offloading of the set of one or more tasks from the second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) is further based on one or more categories of user devices associated with the one or more tasks.

14. The method according to any of the preceding claims, comprising: after offloading at least a subset of the set of one or more tasks from the second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700), the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) providing feedback (416) to the second node, the feedback (416) indicating energy efficiency of the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) after the offloading.

15. The method according to any of the preceding claims, wherein the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) corresponds to a first Radio Access Network, RAN, node and the second node (100; 101 , 102; 200, 201 , 203;300, 301 , 302, 303; 700) corresponds to a second RAN node.

16. The method according to any of the preceding claims, wherein the one or more tasks comprise serving of user devices (10).

17. A first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) for a wireless communication network, the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) being configured to: obtain an indication of energy efficiency gain associated with offloading at least one task from a second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) of the wireless communication network to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700); based on data related to earlier offloading of tasks to the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700), estimate an uncertainty of the indication of energy efficiency gain; and based on the indication of energy efficiency gain and the estimated uncertainty, control execution of the offloading of a set of one or more tasks from the second node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) to the first node (100; 101 , 102; 200, 201 , 203; 300,301 , 302, 303; 700).

18. The first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) according to claim 17, wherein the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) is configured to perform a method according to any one of claims 2 to 16.

19. The first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) according to claim 17 or 18, comprising: at least one processor (750), and a memory (760) containing program code executable by the at least one processor (750), whereby execution of the program code by the at least one processor (750) causes the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) to perform a method according to any one of claims 1 to 16.

20. A computer program or computer program product comprising program code to be executed by at least one processor (750) of a first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) of a communication network, whereby execution of the program code causes the first node (100; 101 , 102; 200, 201 , 203; 300, 301 , 302, 303; 700) to perform a method according to any one of claims 1 to 16.

Citation Information

Patent Citations

  • Methods for inter-node coordination for ran energy saving

    WO2022229260A1

  • Using ai-based models for network energy savings

    WO2023069534A1