Use of machine learning models in communication procedures of a wireless network

Abstract

Present disclosure provides a user equipment, UE, configured for using a machine learning, ML, model at the UE for inference related to a communication procedure of a wireless network; and for allowing switching between using the ML model at the UE and using an ML model at a network entity of the wireless network for the inference during the communication procedure The communication procedure may be a handover procedure.

Description

USE OF MACHINE LEARNING MODELS IN COMMUNICATION PROCEDURES OF A WIRELESS NETWORKTECHNICAL FIELD

[0001] Various example embodiments relate to use of machine learning models in communication procedures of a wireless network.BACKGROUND

[0002] This section illustrates useful background information without admission of any technique described herein representative of the state of the art.

[0003] The use of artificial intelligence (Al) I machine learning (ML) techniques in wireless networks is currently being studied. Present disclosure provides some solutions related to use of machine learning models in communication procedures of a wireless network.SUMMARY

[0004] The scope of protection sought for various embodiments of present disclosure is set out by the independent claims. The embodiments and features, if any, described in this specification that do not fall under the scope of the independent claims are to be interpreted as examples useful for understanding various example embodiments.

[0005] According to a first example aspect of the present disclosure, there is provided a user equipment, UE, comprising: means for using a machine learning, ML, model at the UE for inference related to a communication procedure of a wireless network; and means for allowing switching between using the ML model at the UE and using an ML model at a network entity of the wireless network for the inference during the communication procedure.

[0006] In some example embodiments of the first aspect, the UE further comprises means for determining occurrence of a predefined trigger.

[0007] In some example embodiments of the first aspect, the UE further comprises means for signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

[0008] In some example embodiments of the first aspect, the UE further comprises means for activating or deactivating use of the ML model at the UE responsive to determining occurrence of the predefined trigger.

[0009] In some example embodiments of the first aspect, the UE further comprisesmeans for temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0010] In some example embodiments of the first aspect, the UE further comprises means for collecting monitoring data during use of the ML model at the UE, and means for sharing the collected monitoring data with the network entity.

[0011] In some example embodiments of the first aspect, the UE further comprises means for sharing with the network entity prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0012] According to a second example aspect of the present disclosure, there is provided a method, comprising: using a machine learning, ML, model at the UE for inference related to a communication procedure of a wireless network; and allowing switching between using the ML model at the UE and using an ML model at a network entity of the wireless network for the inference during the communication procedure.

[0013] In some example embodiments of the second aspect, the method further comprises determining occurrence of a predefined trigger.

[0014] In some example embodiments of the second aspect, the method further comprises signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

[0015] In some example embodiments of the second aspect, the method further comprises activating or deactivating use of the ML model at the UE responsive to determining occurrence of the predefined trigger.

[0016] In some example embodiments of the first aspect, the method further comprises temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0017] In some example embodiments of the second aspect, the method further comprises collecting monitoring data during use of the ML model at the UE, and sharing the collected monitoring data with the network entity.

[0018] In some example embodiments of the second aspect, the method further comprises sharing with the network entity prediction state information in connection with switching between using the ML model at the UE and using the ML model at the networkentity.

[0019] According to a third example aspect of the present disclosure, there is provided a computer program comprising instructions for performing at least the following: using a machine learning, ML, model at the UE for inference related to a communication procedure of a wireless network; and allowing switching between using the ML model at the UE and using an ML model at a network entity of the wireless network for the inference during the communication procedure.

[0020] In some example embodiments of the third aspect, the computer program further comprises instructions for performing determining occurrence of a predefined trigger.

[0021] In some example embodiments of the third aspect, the computer program further comprises instructions for performing signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

[0022] In some example embodiments of the third aspect, the computer program further comprises instructions for performing activating or deactivating use of the ML model at the UE responsive to determining occurrence of the predefined trigger.

[0023] In some example embodiments of the third aspect, the computer program further comprises instructions for performing temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0024] In some example embodiments of the third aspect, the computer program further comprises instructions for performing collecting monitoring data during use of the ML model at the UE, and sharing the collected monitoring data with the network entity.

[0025] In some example embodiments of the third aspect, the computer program further comprises instructions for performing sharing with the network entity prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0026] According to a fourth example aspect of the present disclosure, there is provided a user equipment, UE, comprising: at least one processor; and at least one memory including executable instructions that, when executed by the at least one processor, cause the UE to use a machine learning, ML, model at the UE for inference related to a communication procedure of a wireless network; and to allow switching between using the ML model at the UE and using an MLmodel at a network entity of the wireless network for the inference during the communication procedure.

[0027] In some example embodiments of the fourth aspect, the instructions further cause the UE to determine occurrence of a predefined trigger.

[0028] In some example embodiments of the fourth aspect, the instructions further cause the UE to signal an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

[0029] In some example embodiments of the fourth aspect, the instructions further cause the UE to activate or deactivate use of the ML model at the UE responsive to determining occurrence of the predefined trigger.

[0030] In some example embodiments of the fourth aspect, the instructions further cause the UE to temporarily allow simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0031] In some example embodiments of the fourth aspect, the instructions further cause the UE to collect monitoring data during use of the ML model at the UE, and to share the collected monitoring data with the network entity.

[0032] In some example embodiments of the fourth aspect, the instructions further cause the UE to share with the network entity prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0033] In some example embodiments of the first to fourth aspects, the switching comprises at least switching from using the ML model at the UE to using the ML model at the network entity or vice versa.

[0034] In some example embodiments of the first to fourth aspects, the communication procedure is a handover procedure.

[0035] In some example embodiments of the first to fourth aspects, the network entity is a source network entity or a target network entity of the handover procedure.

[0036] In some example embodiments of the first to fourth aspects, the switching is performed so that the ML model at the UE is used before and after the handover procedure and the ML model at the network entity is used during the handover procedure.

[0037] In some example embodiments of the first to fourth aspects, the communication procedure is a beam management procedure or is related to a carrier aggregation reconfiguration.

[0038] In some example embodiments of the first to fourth aspects, the ML model isconfigured to be used for a channel state information, CSI, prediction or for positioning.

[0039] In some example embodiments of the first to fourth aspects, predefined trigger is determined based on one or more of the following: an event, a measurement, ML model reliability measure, energy consumption value.

[0040] According to a fifth example aspect of the present disclosure, there is provided a network entity comprising means for using a machine learning, ML, model at the network entity for inference related to a communication procedure of a wireless network, wherein the communication procedure is related to a UE; and means for allowing switching between using the ML model at the network entity and using an ML model at the UE for the inference during the communication procedure.

[0041] In some example embodiments of the fifth aspect, the network entity further comprises means for determining occurrence of a predefined trigger.

[0042] In some example embodiments of fifth aspect, the network entity further comprises means for signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

[0043] In some example embodiments of fifth aspect, the network entity further comprises means for activating or deactivating use of the ML model at the network entity responsive to determining occurrence of the predefined trigger.

[0044] In some example embodiments of the fifth aspect, the network entity further comprises means for temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0045] In some example embodiments of the fifth aspect, the network entity further comprises means for collecting monitoring data during use of the ML model at the network entity, and means for sharing the collected monitoring data with the UE.

[0046] In some example embodiments of the fifth aspect, the network entity further comprises means for sharing with the UE prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0047] According to a sixth example aspect of the present disclosure, there is provided a method, comprising: using a machine learning, ML, model at the network entity for inference related to a communication procedure of a wireless network, wherein thecommunication procedure is related to a UE; and allowing switching between using the ML model at the network entity and using an ML model at the UE for the inference during the communication procedure.

[0048] In some example embodiments of the sixth aspect, the method further comprises determining occurrence of a predefined trigger.

[0049] In some example embodiments of the sixth aspect, the method further comprises signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

[0050] In some example embodiments of the sixth aspect, the method further comprises activating or deactivating use of the ML model at the network entity responsive to determining occurrence of the predefined trigger.

[0051] In some example embodiments of the sixth aspect, the method further comprises temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0052] In some example embodiments of the sixth aspect, the method further comprises collecting monitoring data during use of the ML model at the network entity, and sharing the collected monitoring data with the UE.

[0053] In some example embodiments of the sixth aspect, the method further comprises sharing with the UE prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0054] According to a seventh example aspect of the present disclosure, there is provided a computer program comprising instructions for performing at least the following: using a machine learning, ML, model at the network entity for inference related to a communication procedure of a wireless network, wherein the communication procedure is related to a UE; and allowing switching between using the ML model at the network entity and using an ML model at the UE for the inference during the communication procedure.

[0055] In some example embodiments of the seventh aspect, the computer program further comprises instructions for performing determining occurrence of a predefined trigger.

[0056] In some example embodiments of the seventh aspect, the computer program further comprises instructions for performing signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

[0057] In some example embodiments of the seventh aspect, the computer programfurther comprises instructions for performing activating or deactivating use of the ML model at the network entity responsive to determining occurrence of the predefined trigger.

[0058] In some example embodiments of the seventh aspect, the computer program further comprises instructions for performing temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0059] In some example embodiments of the seventh aspect, the computer program further comprises instructions for performing collecting monitoring data during use of the ML model at the network entity, and sharing the collected monitoring data with the UE.

[0060] In some example embodiments of the seventh aspect, the computer program further comprises instructions for performing sharing with the UE prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0061] According to an eighth example aspect of the present disclosure, there is provided a network entity, comprising: at least one processor; and at least one memory including executable instructions that, when executed by the at least one processor, cause the network entity to use a machine learning, ML, model at the network entity for inference related to a communication procedure of a wireless network, wherein the communication procedure is related to a UE; and to allow switching between using the ML model at the network entity and using an ML model at the UE for the inference during the communication procedure.

[0062] In some example embodiments of the eighth aspect, the instructions further cause the network entity to determine occurrence of a predefined trigger.

[0063] In some example embodiments of the eighth aspect, the instructions further cause the network entity to signal an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

[0064] In some example embodiments of the eighth aspect, the instructions further cause the network entity to activate or deactivate use of the ML model at the network entity responsive to determining occurrence of the predefined trigger.

[0065] In some example embodiments of the eighth aspect, the instructions further cause the network entity to temporarily allow simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0066] In some example embodiments of the eighth aspect, the instructions further cause the network entity to collect monitoring data during use of the ML model at the network entity, and to share the collected monitoring data with the UE.

[0067] In some example embodiments of the eighth aspect, the instructions further cause the network entity to share with the UE prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0068] In some example embodiments of the fifth to eighth aspects, the switching comprises at least switching from using the ML model at the UE to using the ML model at the network entity or vice versa.

[0069] In some example embodiments of the fifth to eighth aspects, the communication procedure is a handover procedure.

[0070] In some example embodiments of the fifth to eighth aspects, the network entity is a source network entity or a target network entity of the handover procedure.

[0071] In some example embodiments of the first to fourth aspects, the switching is performed so that the ML model at the UE is used before and after the handover procedure and the ML model at the network entity is used during the handover procedure.

[0072] In some example embodiments of the fifth to eighth aspects, the communication procedure is a beam management procedure or is related to a carrier aggregation reconfiguration.

[0073] In some example embodiments of the fifth to eighth aspects, the ML model is configured to be used for a channel state information, CSI, prediction or for positioning.

[0074] In some example embodiments of the fifth to eighth aspects, predefined trigger is determined based on one or more of the following: an event, a measurement, ML model reliability measure, energy consumption value.

[0075] The computer program disclosed herein may be stored in a non-transitory computer readable memory medium. The term non-transitory as used herein is a limitation of the medium itself (i.e. tangible, not a signal) as opposed to a limitation on data storage persistency (e.g. RAM vs. ROM).

[0076] Any foregoing memory medium may comprise a digital data storage such as a data disc or diskette, optical storage, magnetic storage, holographic storage, opto-magnetic storage, phase-change memory, resistive random access memory, magnetic random access memory, solid-electrolyte memory, ferroelectric random access memory, organic memory or polymer memory. The memory medium may be formed into a device without other substantial functions than storing memory or it may be formed as part of a device with other functions, including but not limited to a memory of a computer, a chip set, and a sub assembly of anelectronic device.

[0077] The various means of various example aspects may comprise a processor; and a memory including executable instructions that, when executed by the processor, cause the performance of the means.

[0078] Different non-binding example aspects and embodiments of the present invention have been illustrated in the foregoing. The embodiments in the foregoing are used merely to explain selected aspects or steps that may be utilized in implementations of the embodiment of the present disclosure. Some embodiments may be presented only with reference to certain example aspects. It should be appreciated that corresponding embodiments may apply to other example aspects as well.BRIEF DESCRIPTION OF THE DRAWINGS

[0079] For a more complete understanding of example embodiments of the present invention, reference is now made to the following descriptions taken in connection with the accompanying drawings in which:

[0080] Fig. 1 illustrates an example scenario;

[0081] Fig. 2 shows an example signaling diagram of Conditional Handover procedure;

[0082] Figs. 3 and 4 are flow diagrams of methods according to some example embodiments;

[0083] Fig. 5 depicts an example embodiment;

[0084] Fig 6 shows a signaling diagram according to an example embodiment;

[0085] Figs. 7A-7B show signaling diagrams of a handover embodiment;

[0086] Figs. 8A-8C show signaling diagrams of some UE mobility embodiments; and

[0087] Figs. 9 and 10 show block diagrams of apparatuses according to an example embodiment.DETAILED DESCRIPTON OF THE DRAWINGS

[0088] An example embodiment of the present disclosure and its potential advantages are understood by referring to Figs. 1 through 10 of the drawings. In this document, like reference signs denote like parts or steps.

[0089] Solutions of present disclosure may be applied in 5G or 6G network technologies, for example. It is however to be noted that various example embodiments of present disclosure may be equally applied to other mobile networks and radio communication technologies (e.g. future technologies or earlier technologies). An example of applicable network technologies is 5G-Advanced, which corresponds to 3GPP Release 18 and beyond.

[0090] Various embodiments are discussed in the context of communications and communication procedures involving a user equipment (UE) and a network entity. The UE is to be interpreted broadly to cover any type of communication device that may perform communications over a wireless communications network. The UE may be e.g. a mobile phone, a smart phone, a laptop, a computer, an loT device. The network entity is to be interpreted broadly to cover any network entity involved in air-interface procedures of a wireless network. The network entity may be a base station (gNB in 5G) or part of a base station, for example.

[0091] The radio access network (RAN) of a wireless network may be split into a radio unit (RU), a distributed unit (DU), and a central unit (CU). The RU is where the radio frequency signals are transmitted, received, amplified, and digitized. The RU is located near or integrated into, the antenna. The DU and CU are the computation parts of the base station sending the digitalized radio signal into the network. The split of the functionalities between the DU and CU may vary. The CU may provide support for the higher layers of the protocol stack while the DU provides support for the lower layers of the protocol stack. Some example embodiments of present disclosure define certain functionalities in a DU or CU. The network entity of various embodiments of present disclosure may be for example a source DU or source node, a target DU or a target node, a CU. It is to be noted that even though some specific terms are used in this disclosure, some other names may be used for the respective entities, too, e.g. in future technologies.

[0092] Leveraging artificial Intelligence (Al) / Machine Learning (ML) based algorithms in air-interface of wireless networks is currently under investigation in the 3GPP standardization. There are various procedures in which AI / ML approaches may be used, such as handover and beam management, for example. AI / ML-based algorithms may provide for example enhanced performance and / or reduced complexity / overhead. The enhanced performance may refer to e.g. improved throughput, robustness, accuracy, or reliability, etc.

[0093] Fig. 1 illustrates an example scenario in which various embodiments of present disclosure may be applied. Fig. 1 shows a UE 101 and a network entity 111 as well as a ML model 102 at the UE 101 and a ML model 112 at the network entity 111. The UE 101 and the network entity 111 are capable of using the respective ML model 102 and ML model 112 for predictions or inference related to a communication procedure 1.3 in which the UE 101 and the network entity 111 interact with each other.

[0094] Inference herein refers to the phase of using an ML model for providing an output. Prior to using the ML model for inference (or prior to the ML model inference phase) the ML model has been trained using a suitable training data set and validated using suitable validation data set. Inference may be referred to as obtaining a prediction, a classification ora score using the ML model. Dual inference with ML models refers to a scenario where there is no dependency between the ML model at the UE and the ML model at the network entity and both ML models may be used for the same purpose independently.

[0095] Solutions of present disclosure provide flexible handling of inference pertaining to ML assisted functionality in wireless networks. In cases where there is a network side ML model and a UE side ML model for a given use case, control of the prediction / inference is allowed to flexibly move between the UE and the network.

[0096] In some embodiments of present disclosure, it is assumed that dual inference involving the ML model at the UE and the ML model at the network entity allows an accurate enough final outcome of an ML assisted functionality while using either the ML model at the UE or using the ML model at the network entity. There is no need for one or the other to provide better results. Instead, depending on the phase of a communication procedure either the UE or the network side may provide better or more reliable results.

[0097] It is to be noted that details of the ML model or ML algorithm in the context of present disclosure may vary and various different ML algorithms may be used in embodiments of present disclosure. Examples of ML models types are: supervised learning based models, unsupervised learning based models, semi-supervised learning based model or reinforcement learning based models. Other examples include combination of two or more of the above ML model types.

[0098] Various embodiments of present disclosure may be used in connection with a handover procedure. Fig. 2 shows an example signaling diagram of Conditional Handover procedure. The shown signaling diagram may be adapted to include one or more features of embodiments of present disclosure. Fig. 2 shows a UE 101 , a source node 202, a target node 203, other nodes 204 (other potential target nodes), S-GW / UPF 205 and MME / AMF 206.

[0099] In steps 2.1-2.9, a configured event triggers the UE 101 to send a measurement report 2.1 . Based on this report, the source node 202 can prepare one or more target nodes 203, 204 for the handover (2.3 and 2.4 CHO Request + 2.7 and 2.8 CHO Request Acknowledge) and then sends an RRC Reconfiguration (2.9 CHO command) to the UE 101. This is according to baseline handover procedure of 3GPP NR Rel. 15 specifications.

[0100] For a baseline handover of NR Rel. 15, the UE 101 will immediately access the target node 203 to complete the handover. Instead, in CHO, the UE 101 will only access the target node 204 once an additional CHO execution condition expires 2.10-2.12 (i.e., the HO preparation and execution phases are decoupled). The condition is configured by the source node 202 in CHO Command 2.9.

[0101] Once the UE 101 completes the handover execution 2.13-2.15 to the target node 203 (e.g. UE 101 has sent 2.15 RRC Reconfiguration Complete), the target node 203sends to the source node 202 a handover success indication 2.16. When receiving this indication from the target node 203, the source node 202 stops its TX / RX to / from the UE 101 and starts data forwarding to the target node 203 in 2.17-2.19. Moreover, the source node 202 may release the CHO preparations 2.20 in other nodes 204 (which are no longer needed).

[0102] The advantage of the CHO is that the HO command can be sent early, when the UE is still safe in the source node / cell, without risking the access in the target node / cell and the stability of its radio link. In this way, conditional handover provides mobility robustness.

[0103] Various embodiments of present disclosure may be used in connection with CHO and in connection with normal HO as well. A more detailed example will be discussed in detail in connection with Figs. 7A and 7B.

[0104] Various embodiments of present disclosure may be used in connection with L1 / L2 triggered mobility (LTM) procedure, too. The LTM procedure can be summarized under three main steps. During a preparation phase, the UE sends measurement reports containing the measurements of the serving and target cells. Based on the received measurement reports, the CU configures the LTM candidate cells, which are then forwarded to the UE by using an RRC Reconfiguration message. During an execution phase, the UE reports measurements to the serving DU. The serving DU makes a decision on whether to trigger the cell switch (e.g, by checking that L1-RSRP of target beam measurement > L1-RSRP of serving beam measurement + Offset for a time period (i.e. Time-to-Trigger (TTT) period)). In this case, the DU triggers the HO by sending a MAO CE command to the UE and the UE can then try to connect to the target DU. During a completion phase, the UE context is released from the source DU and a path switch is performed to a new serving DU.

[0105] For example ML model based beam management can be used in connection with LTM procedure.

[0106] Legacy beam management procedures (P1 , P2, P3) require time-consuming operation of sweeping all the Tx and Rx beams by configuring the UE with a large number of SSB / CSI-RSs measurements. ML model may used for spatial beam prediction and / or time beam prediction.

[0107] In a beam management example, the ML model input may include L1-RSRP measurements of subset of narrow beams or L1-RSRP measurements of wide beams. The ML model output may provide best narrow beam ID or best narrow beam RSRP and / or QoS value for beam selection, for example.

[0108] More detailed examples of using various embodiments of present disclosure in connection with LTM procedure will be discussed in detail in connection with Figs. 8A and 8B and 8C.

[0109] In another example, the ML model may be used for CSI prediction. A UE sideML model may be used for this purpose but additionally also network side ML model may provide CSI predictions to certain degree of accuracy.

[0110] In another example, the ML model may be used for positioning. In such case the output of the ML model may be position / location information of the UE or the output of the ML model may provide new measurements and / or enhancement of existing measurements.

[0111] Various embodiments of present disclosure aim at time and spatial continuity and stability of ML model inference during UE mobility / handover procedure or other relevant communication procedure while maintaining sufficient performance of the ML model based functionality. Let’s consider for example beam prediction in time and spatial domain during a handover procedure, wherein a UE ML model is capable of predicting top K beams for a prediction window Twin when the UE is being served by a given cell. In such case, the UE is required to perform target cell (and beam) measurements. During the handover procedure the UE moves from a source cell X to a target cell Y and the beam prediction “reliability” may not remain constant during this movement if the UE ML model that the UE is applying does not work well in the cell overlap region between the source cell X and the target cell Y. Consequently, when the UE moves towards the target cell Y, the source cell X beam prediction may weaken to the point that beam switching based on the ML model may no longer be reliable. This may cause loss of UE throughput in the source cell X. When the UE enters into target cell Y coverage, the target cell Y beam prediction may not be reliable enough due to weak signal and selecting target beam for random access (RA) may no longer be reliable. This may cause that chances of RACH failure are high. Hence, if the ML model used for beam prediction cannot perform target cell beam specific predictions reliably, it may cause handover failure during RACH procedure at handover execution. Solutions of various embodiments of present disclosure may help in avoiding such problems.

[0112] In case of the ML model enabled CSI prediction, the reliability of the prediction strongly depends on the number of past CSI estimates used as input and on the target prediction window. When the UE is required to perform several neighbor / target cell measurements, there might be cases when there is no sufficient time left for the UE BB processing to collect enough samples from the source cell. Hence, the CSI prediction for the source cell may degrade. Additionally or alternatively, due to the increased radio interference from the closest neighbor cells, the input CSI may become more impacted by noise. Hence, the accuracy of the predicted CSI may degrade. Solutions of various embodiments of present disclosure may help in avoiding such problems.

[0113] In case of the ML model enabled positioning, wherein the UE performs the ML model inference or provides assistance to the network to perform the ML model inference, the reliability of the inferred position depends on the timing and also on the number of samplesof the input data, such as PRS-TRP measurements. When the UE starts a handover procedure which involves neighbor / target cells measurements, there might cases where there is not enough compute power (resource) left for the UE to keep providing the input measurements required for the ML model-enabled positioning or assistance information. Additionally or alternatively, the received PRS signal level may decrease for all measured TRPs (source and neighbor cells). Hence, the derived channel impulse response (CIR) or power delay profile (PDP) may be impacted by noise and the accuracy of the positioning predictions / inference may decrease. Solutions of various embodiments of present disclosure may help in avoiding such problems.

[0114] Figs. 3 and 4 are flow diagrams of methods according to some example embodiments. The methods may be implemented in one or more computer programs comprising instructions for performing the steps of the respective method. One or more of the steps of the flow diagrams may be optional and the order of the steps may vary from the shown order.

[0115] Fig. 3 shows a process to be performed in a UE. The process comprises the following steps:

[0116] 411 : Using a ML model at the UE for inference related to a communication procedure of a wireless network. The communication procedure may be for example a handover procedure or a beam management procedure or the communication procedure may be related to a carrier aggregation reconfiguration. The ML model may be configured to be used for a CSI prediction or for positioning.

[0117] 412: Allowing switching between using the ML model at the UE and using an ML model at a network entity of the wireless network for the inference during the communication procedure. The switching may refer to switching from using the ML model at the UE to using the ML model at the network entity or vice versa or both.

[0118] In an embodiment, wherein the communication procedure is handover, the switching is performed so that the ML model at the UE is used before and after the handover and the ML model at the network entity is used during the handover.

[0119] 413: Determining occurrence of a predefined trigger. The predefined trigger triggers the switching of the ML model inference responsibility from the UE to the network entity or vice versa. The predefined trigger may be one or more of the following: an event, a measurement result, ML model reliability measure, energy consumption value. More specific examples of the predefined trigger include for example a mobility event based on radio measurements (e.g., any of the Ax events), an RRC reconfiguration event, a non-radio measurement result (e.g., determining change in location or tracking area), ML model output reliability measure (e.g., accuracy degradation below a configured threshold), energyconsumption value (e.g., increased UE energy consumption due to extensive radio measurements).

[0120] 414: Signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger. In this option, the UE indicates to the network entity that the ML model inference responsibility is to be switched between the UE and the network (either from the UE to the network entity or vice versa).

[0121] 415: Activating or deactivating use of the ML model at the network entity responsive to determining occurrence of the predefined trigger. In this option there is not necessarily a need to communicate with the UE about switching between using the ML model at the UE and using the ML model at the network entity. Instead, the UE and the network entity may independently determine that the responsibility is switched.

[0122] 416: Temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity. In this way, an overlap period is provided, which may mitigate potential impact of ping pong effect of switching back and forth between the UE and the network. For example in the LTM procedure it is undesirable to have the UE switching back-and-forth between beams.

[0123] 417: Receiving monitoring data during use of the ML model at the network entity.

[0124] 418: Collecting monitoring data during use of the ML model at the UE.

[0125] 419: Sharing the collected monitoring data with the network entity.

[0126] 420: Sharing with the network entity prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0127] Fig. 4 shows a process to be performed in a network entity, such as a gNB, source node, source DU, target node, target DU, CU. The process comprises the following steps:

[0128] 441 : Using a ML model at the network entity for inference related to a communication procedure of a wireless network, wherein the communication procedure is related to a UE. The communication procedure may be for example a handover procedure or a beam management procedure or the communication procedure may be related to a carrier aggregation reconfiguration. The ML model may be configured to be used for a CSI prediction or for positioning.

[0129] 442: Allowing switching between using the ML model at the network entity and using an ML model at the UE for the inference during the communication procedure. Theswitching may refer to switching from using the ML model at the UE to using the ML model at the network entity or vice versa or both.

[0130] In an embodiment, wherein the communication procedure is handover, the switching is performed so that the ML model at the UE is used before and after the handover and the ML model at the network entity is used during the handover.

[0131] 443: Determining occurrence of a predefined trigger. The predefined trigger triggers the switching of the ML model inference responsibility from the UE to the network entity or vice versa. The predefined trigger may be one or more of the following: an event, a measurement result, ML model reliability measure, energy consumption value. More specific examples of the predefined trigger include for example a mobility event based on radio measurements (e.g., any of the Ax events), an RRC reconfiguration event, a non-radio measurement result (e.g., determining change in location or tracking area), ML model output reliability measure (e.g., accuracy degradation below a configured threshold), energy consumption value (e.g., increased UE energy consumption due to extensive radio measurements).

[0132] 444: Signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger. In this option, the network entity indicates to the UE that the ML model inference responsibility is to be switched between the UE and the network (either from the UE to the network entity or vice versa).

[0133] 445: Activating or deactivating use of the ML model at the network entity responsive to determining occurrence of the predefined trigger. In this option there is not necessarily a need to communicate with the UE about switching between using the ML model at the UE and using the ML model at the network entity. Instead, the UE and the network entity may independently determine that the responsibility is switched.

[0134] 446: Temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity. In this way, an overlap period is provided, which may mitigate potential impact of ping pong effect of switching back and forth between the UE and the network. For example, in the LTM procedure it is undesirable to have the UE switching back-and-forth between beams.

[0135] 447: Receiving, from the UE, monitoring data collected during use of the ML model at the UE.

[0136] 448: Collecting monitoring data during use of the ML model at the network entity.

[0137] 449: Sharing the collected monitoring data with the UE.

[0138] 450: Sharing with the UE prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

[0139] Fig. 5 depicts an example embodiment. Fig 5 shows a source gNB 502 and a target gNB 503 and a UE 101 that is handed over from the source gNB 502 to the target gNB 503. ML model inference responsibility is switched between the UE 101 and the source gNB 502 and the target gNB 503 during the handover procedure as follows:

[0140] 511 : ML model inference is initially at the UE 101. This is performed using configuration(s) provided by the source gNB 502.

[0141] 512: Handover procedure is initiated, and the ML model inference is switched to the source gNB 502.

[0142] 513: The ML model inference is switched to the target gNB 503.

[0143] Alternatively, a combined or shared network side ML model inference may be used for the source and target gNB (e.g. at gNB-CU, LMF, NWDAF).

[0144] During handover procedure, the UE 101 may provide different type of reports (data) for the network side predictions. The associated monitoring data collection at the UE may be deactivated during the handover procedure after the ML model inference is switched to the network side.

[0145] 514: There may be a prediction time overlap period after the handover procedure. Both the UE and the network side perform prediction during such time overlap period. Similar time period may be used before or in the beginning of the handover procedure.

[0146] 515: After the handover procedure, ML model inference at the UE 101 is reenabled using the configuration(s) provided by the target gNB 503.

[0147] In an embodiment, inference data collection and / or monitoring data collection is performed at the UE before and after the handover procedure and at the network during the handover procedure.

[0148] Switching the ML model inference responsibility may be stateless or stateful. Using prediction state information (such as input time series, latent space, model weights & biases, etc.) exchanged between the UE and the network side the prediction can be made stateful, which may provide increased accuracy. In a stateless switch, the prediction state information before the switch from the UE to the network side and / or from the network side to the UE, and / or from cell to cell, is not exchanged, and prediction is initialized after the switch with completely new model and / or set of input data. In a stateful switch, the prediction state information before the switch from the UE to the network side and / or from the network side to the UE, and / or from cell to cell, is passed between the UE and the network side, and prediction continues after the switch with same model and / or common input data.

[0149] Fig. 6 shows a signaling diagram according to an example embodiment. Fig. 6 shows a UE 101, a source node 202, a target node 203, and other nodes 204 (other potential target nodes). The signaling diagram includes the following steps / messages / phases:

[0150] 6.1: The UE 101 conveys inference status to the source node 202.

[0151] 6.2: The source node 202 switches inference to the network side.

[0152] 6.3: Inference responsibility is shared between the networks side entities (the source node 202, the target node 203 and optionally with the other nodes 204).

[0153] 6.4: The source node 202 conveys to the UE an indication of switching the inference responsibility.

[0154] Figs 7A-7B show signaling diagrams of a handover embodiment. Figs. 7A- 7B show a UE 101 , a source node 202, a target node 203, and other nodes 204 (other potential target nodes). The signaling diagram of Fig. 7A includes the following steps / messages / phases:

[0155] 7.1: The UE 101 indicates in an uplink measurement report a requirement for the network to support flexible inference. Flexible inference support is required in a dualinference ML model scenario where the UE and the network are able to independently support a given ML functionality (e.g., beam prediction in time or spatial domain) using ML model at the UE and at the network side. The UE may set the indication based on an estimation within the UE regarding a source / target cell pair where the UE may not be fully capable of supporting inference at the UE side during the HO procedure. One reason to set such an indication may come from a restriction of the ML functionality or ML model that may not support inference in a coverage area of two or more cells or may support reliable inference only when the source or target cell radio quality exceeds a given threshold (e.g., RSRP better than -75 dBm). In another example for a beam prediction scenario, due to higher interference in the coverage area between two cells, the ML functionality may not be able to predict more than one best beam, or the prediction window may be too short in the future. In some cases, the beams that are predicted may predict a lot of unnecessary beam switches and this may be undesirable from the network point of view of desired switching resources (e.g., signalling, random access). This necessitates switching inference from the UE to the network.

[0156] 7.2: The source node 202 decides to perform a HO to the target node 203 and potentially other target nodes 204.

[0157] 7.3, 7.4: A (C)HO request is set to the target node 203 and the other nodes 204. The source node 202 sets the indication to the targets with a configuration request for flexible inference. The configuration may comprise of a source node proposal of the conditions (e.g., RSRP thresholds for signal quality, inference KPI threshold of 90% below which UE cannot continue inference at its side, etc.,) in which the inference will be switched back to thenetwork, a set of monitoring resources configured for the UE to determine the conditions. The request may also comprise an indication to the target node to also make the inference determination i.e. , both the source and target node perform the inference and share it among themselves.

[0158] 7.5, 7.6 : The target nodes 203, 204 process the (C)HO request from the source node 202. The target nodes 203-204 make the necessary resource reservations for allowing the inference to be performed at the respective target nodes 203, 204.

[0159] 7.7, 7 9: The target nodes 203, 204 indicate back to the source how they plan to share the inference results to the source node 202 during the HO procedure.

[0160] 7.8, 7.10: The source node 202 collects the configurations for the different target cells 203, 204 and finalizes the message to be sent towards the UE 101 in step 7.11. The network is now able to initiate inference at the source node 202 and the target nodes 203, 204 and communication between them to share the inference results.

[0161] 7.11 : The RRC reconfiguration message is sent to the UE 101 to configure the conditions for the UE to disable inference at the UE.

[0162] 7.12: The UE 101 completes the RRC reconfiguration and sends indication to the source node 202.

[0163] 7.13: The source node 202 initiates the inference switch from its side thus initiating the dual-inference mode of operation (inference performed both at the UE and the source node).

[0164] 7.14: Inference is switched from the UE 101 to the network. Further details of this phase are given in Fig. 7B.

[0165] 7.19: It is optionally possible that the UE does not stop inference during HO but e.g. switches to another HO-specific ML model (i.e. a ML model specifically designed to be used during HO procedures). Once the UE successfully completes the HO to the target node 203, the UE may provide information to the new serving node (earlier target node) about its inference status.

[0166] 7.20: The network decides to switch the inference back to the UE. This may be potentially or optionally performed after a time overlap period (to avoid inference responsibility ping-pongs, etc).

[0167] 7.21 , 7.22: The network / target node initiates the switch of the inference responsibility from the network to the UE. The UE acknowledges the inference switch.

[0168] The signaling diagram of Fig. 7B provides further details of step 7.14 of Fig. 7 A and includes the following steps / messages / phases:

[0169] 7.34, 7.38: The UE evaluates (C)HO conditions and when these are fulfilled it executes the configured HO procedures.

[0170] 7.35, 7.36, 7.37: In one or more possible instants of time or conditions, during the time interval between steps 7.34 and 7.38, the UE 101 may trigger an UL indication wherein it includes the status of inference at the UE and adds the output of the inference in the report towards the network. The report may comprise of the status of the prediction made by the UE and if it is suspended due to the conditions met due to configuration received at 7.11 of Fig. 7A. The report may also comprise of ground truth measurements to assist the network. Based on the report 7.35, the network (the source node 202) may decide to switch inference to the network and use the results from the network inference. Correspondingly, the network (the source node 202) indicates this decision to the UE 101 in the DL indication messages from the network to UE direction.

[0171] The ground truth measurements above and later in this disclosure refer to data / information which are used as target (reference) ML output when performing supervised ML model training. In case of beam prediction, the ground truth measurements are e.g. the actual measured beams and their signal strength (RSRP) values. In case positioning use case, the ground truth measurements are e.g. the actual location information (X-Y-Z- coordinates).

[0172] Figs. 8A-8C show signaling diagrams of some UE mobility embodiments. Figs. 8A-8C show a UE 101 , a source DU 802, a target DU 803, and CU 804. The signaling diagram of Fig. 8A includes the following steps / messages / phases:

[0173] 8.1 , 8.2 : The UE 101 indicates in an uplink measurement report a requirement for the network to support flexible inference. Flexible inference support is required in a dual-inference ML model scenario where the UE and the network are able to independently support a given ML functionality (e.g., beam prediction in time or spatial domain) using ML model at the UE and at the network side. The UE 101 may set the indication based on an estimation at the UE regarding a source / target cell pair where the UE may not be fully capable of supporting-inference at the UE side during the HO procedure. One reason to set such an indication may come from a restriction of the ML functionality or ML model that may not support inference in a coverage area of two or more cells or may support reliable inference only when the source or target cell radio quality exceeds a given threshold (e.g., RSRP better than -75 dBm). In another example for beam prediction scenario, due to higher interference in the coverage area between two cells, the ML functionality may not be able to predict more than one best beam, or the prediction window may be too short in the future. In some cases, the beams that are predicted may predict a lot of unnecessary beam switches and this may be undesirable from the network point of view of desired switching resources (e.g., signalling, random access). This necessitates switching inference from the UE to the network.

[0174] 8.3: The CU 804 decides to configure LTM to the target node and potentiallyother target nodes for which it formats an LTM request.

[0175] 8.4, 8 5: The CU 804 then sets the indication to the target DUs 803 (including the current source DU 802 if it is configured to be an LTM candidate cell) with a configuration request for flexible inference. The configuration may comprise of a CU proposal of the conditions (e.g., RSRP thresholds for signal quality, inference KPI threshold of 90% below which UE cannot continue inference at its side, etc.,) in which the inference will be switched back to the network, a set of monitoring resources configured for the UE to determine the conditions. The request may also comprise an indication to the target DU 803 to also make the inference determination i.e., both the source and target DUs perform the inference and share it among themselves.

[0176] 8.6, 8 7: The target DUs process the LTM request from the CU and make the necessary resource reservations for allowing the inference to be performed at the respective target DUs.

[0177] 8.8, 8 9 : The target DUs indicate back to the CU how they plan to share the inference results to the serving DU during the HO procedure.

[0178] 8.10: The CU 804 collects the configurations for the different target DUs and finalizes the message to be sent towards the UE in steps 8.11 , 8.12. The network is now able to initiate inference at the source and target DUs, and communication between them to share the inference results.

[0179] 8.11 , 8.12: The RRC reconfiguration message is sent to the UE to configure the conditions for the UE to disable inference at the UE.

[0180] 8.13: The UE completes the RRC reconfiguration and sends indication to the source DU 802.

[0181] 8.14: The source DU initiates the inference switch from its side thus initiating the dual-inference mode of operation (inference performed both at the UE and source DU).

[0182] 8.16: Inference is switched from the UE 101 to the network. Further details of this phase are given in Figs. 8B and 8C.

[0183] 8.25, 8.26, 8.27: LTM completion procedure according to 3GPP specifications takes place.

[0184] The signaling diagram of Fig. 8B provides further details of step 8.16 of Fig. 8A and includes the following steps / messages / phases:

[0185] 8.37, 8.38, 8.39: In one or more possible instants of time or conditions, during the time interval between steps 8.37 and 8.40, the UE may trigger an UL indication wherein it includes the status of inference at the UE and adds the output of the inference in the report towards the network. The report may comprise of the status of the prediction made by the UE and if it is suspended due to the conditions met due to configuration received at 8.12 of Fig.8A. The report may also comprise of ground truth measurements to assist the network. Based on the report 8.37, the source DU 802 may decide to switch inference to the network and use the results from the network inference. Correspondingly, the network indicates this decision to the UE via a DL indication message (e.g., it can be a MAC CE message) from the network to UE direction.

[0186] 8.40: The UE can send measurement reports to the source DU (as indicated in step 8.37). The source DU can then instruct the UE to switch to another configured LTM target DU. The UE disconnects from the source, and begins the random-access procedure with the target DU.

[0187] 8.41 : It is optionally possible that the UE does not stop inference during HO but e.g. switches to another HO-specific ML model (i.e. a ML model specifically designed to be used during HO procedures). Once the UE successfully completes the HO to the target DU 803, the UE may provide information to the new serving DU (earlier target DU) about its inference status.

[0188] 8.42: The network decides to switch the inference back to the UE. This may be potentially or optionally performed after a time overlap period (to avoid inference responsibility ping-pongs, etc).

[0189] 8.43, 8.44: The network / target DU initiates the switch of the inference responsibility from the network to the UE. The UE acknowledges the inference switch.

[0190] The signaling diagram of Fig. 8C provides another alternative of step 8.16 of Fig. 8A. In Fig 8B, the inference is performed at the DU, which can exchange MAC-level messages with the UE. Fig. 8C shows another alternative wherein the inference is performed at the CU. Fig. 8C includes the following steps / messages / phases.

[0191] 8.57-8.62: In one or more possible instants of time or conditions, during the time interval between step 8.57 and 8.63, the UE may trigger an UL indication wherein it includes the status of inference at the UE and adds the output of the inference in the report towards the network. The report may comprise of the status of the prediction made by the UE and if it is suspended due to the conditions met due to configuration received at 8.12 of Fig. 8A. The report may also comprise of ground truth measurements to assist the network. Based on the report 8.37, the CU 804 may decide (step 8.60) to switch inference to the network and use the results from the network inference. Correspondingly, the network indicates this decision to the UE via a DL indication message (e.g., it can be a RRC Reconfiguration message, steps 8.61 , 8.62) from the network to UE direction.

[0192] 8.63: The UE can send measurement reports to the source DU (as indicated in step 8.57). The source DU can then instruct the UE to switch to another configured LTM target DU. The UE disconnects from the source, and begins the random-access procedurewith the target DU.

[0193] 8.64, 8.65: It is optionally possible that the UE does not stop inference during HO but e.g. switches to another HO-specific ML model (i.e. a ML model specifically designed to be used during HO procedures). Once the UE successfully completes the HO to the target DU 803, the UE may provide information to the CU about its inference status.

[0194] 8.66: The CU decides to switch the inference back to the UE. This may be potentially or optionally performed after a time overlap period (to avoid inference responsibility ping-pongs, etc).

[0195] 8.67-8.70: The CU initiates the switch of the inference responsibility from the network to the UE. The UE acknowledges the inference switch.

[0196] Figs. 9-10 show block diagrams of apparatuses according to example embodiments.

[0197] Fig. 9 shows a block diagram of an apparatus 900 according to an example embodiment. The apparatus 900 may operate as user equipment, such as the UE 101 of Figs. 1-8C. The apparatus 900 generally comprises a memory 940 including a computer program code 950. The apparatus 900 further comprises a processor 920 for controlling the operation of the apparatus 900 using the computer program code 950, and a communication unit 910 for communicating with other network entities and other devices, and a user interface unit 930 for interacting with a user.

[0198] The communication unit 910 comprises, for example, one or more of: a local area network (LAN) port; a wireless local area network (WLAN) unit; Bluetooth unit; cellular data communication unit; or satellite data communication unit. The communication interface 910 may support one or more different communication technologies. The communication interface 910 may support Ethernet communications and / or IP based communications. The apparatus 900 may also or alternatively comprise more than one of the communication interfaces 910. The processor 920 comprises, for example, any one or more of: a master control unit (MCU); a microprocessor; a digital signal processor (DSP); an application specific integrated circuit (ASIC); a field programmable gate array; and a microcontroller. Various parts may be implemented using more than one corresponding or different elements, such as memories and storages may be multiplied for capacity and / or redundancy purposes. Similarly, processing and / or communications may be implemented with multiple parallel or elements for capacity and / or redundancy purposes.

[0199] The computer program code 950 may control the apparatus 900 to provide one or more example embodiments of present disclosure, such as suitable parts of Figs. 1- 8C.

[0200] Fig. 10 shows a block diagram of an apparatus 1000 according to an exampleembodiment. The apparatus 1000 may operate as a network entity, such as a base station or gNB, source node, source DU, target node, target DU, or CU of Figs. 1-8C. The apparatus 1000 generally comprises a memory 1040 including a computer program code 1050. The apparatus 1000 further comprises a processor 1020 for controlling the operation of the apparatus 1000 using the computer program code 1050, and a communication unit 1010 for communicating with other network entities and other devices. Further, the apparatus 1000 may comprise a user interface unit 1030.

[0201] The communication unit 1010 comprises, for example, one or more of: a wireless local area network (WLAN) unit; Bluetooth unit; cellular data communication unit; or satellite data communication unit. The communication interface 1010 may support one or more different communication technologies. The communication interface 1010 may support IP based communications. The apparatus 1000 may also or alternatively comprise more than one of the communication interfaces 1010. The processor 1020 comprises, for example, any one or more of: a master control unit (MCU); a microprocessor; a digital signal processor (DSP); an application specific integrated circuit (ASIC); a field programmable gate array; and a microcontroller. The user interface unit 1030 may comprise a circuitry for receiving input from a user of the apparatus 1000, e.g., via a keyboard; graphical user interface of a display; speech recognition circuitry; or an accessory device; such as a headset; and for providing output to the user via, e.g., a graphical user interface or a loudspeaker. Various parts may be implemented using more than one corresponding or different elements, such as memories and storages may be multiplied for capacity and / or redundancy purposes. Similarly, processing and / or communications may be implemented with multiple parallel or elements for capacity and / or redundancy purposes.

[0202] The computer program code 1050 may control the apparatus 1000 to provide one or more example embodiments of present disclosure, such as suitable parts of Figs. 1- 8C.

[0203] As used in this application, the term “circuitry” may refer to one or more or all of the following:(a) hardware-only circuit implementations (such as implementations in only analog and / or digital circuitry) and;(b) combinations of hardware circuits and software, such as (as applicable):(i) a combination of analog and / or digital hardware circuit(s) with software / firmware; and(ii) any portions of hardware processor(s) with software (including digital signal processor(s)), software, and memory(ies) that work together to cause an apparatus, such as a mobile phone or server, to perform various functions); and(c) hardware circuit(s) and or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (e.g., firmware) for operation, but the software may not be present when it is not needed for operation.

[0204] This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device.

[0205] Without in any way limiting the scope, interpretation, or application of the claims appearing below, a technical effect of one or more of the example embodiments disclosed herein is a flexible switch between ML models at the UEs and at the network side.

[0206] Embodiments of the present disclosure may be implemented in software, hardware, application logic or a combination of software, hardware, and application logic. The software, application logic and / or hardware may reside e.g. on the first apparatus and / or the second apparatus of the foregoing disclosure. In an example embodiment, the application logic, software, or an instruction set is maintained on any one of various conventional computer-readable media. In the context of this document, a “computer-readable medium” may be any non-transitory media or means that can contain, store, communicate, propagate, or transport the instructions for use by or in connection with an instruction execution system, apparatus, or device, such as a computer, with one example of a computer described and depicted in Fig. 9 or Fig. 10. A computer-readable medium may comprise a computer-readable storage medium that may be any media or means that can contain or store the instructions for use by or in connection with an instruction execution system, apparatus, or device, such as a computer.

[0207] If desired, the different functions discussed herein may be performed in a different order and / or concurrently with each other. Furthermore, if desired, one or more of the before-described functions may be optional or may be combined.

[0208] Although various aspects of present disclosure are set out in the independent claims, other aspects may comprise other combinations of features from the described embodiments and / or the dependent claims with the features of the independent claims, and not solely the combinations explicitly set out in the claims.

[0209] It is also noted herein that while the foregoing describes example embodiments, these descriptions should not be viewed in a limiting sense. Rather, there areseveral variations and modifications which may be made without departing from the scope of the present disclosure as defined in the appended claims.LIST OF ABBREVIATIONSAAI Artificial IntelligenceBS Basestation, synonym for 5G gNBBM Beam managementCSI Channel State InformationCU Central unit (5G CU-gNB)DNN Deep neural network (5G DU-gNB)DU Distributed unit (5G DU-gNB) gNB 5th generation Node B (5G gNB)KPI Key performance indicatorLTM Lower layer triggered mobilityML Machine learningNN Neural networkRL Reinforcement learningRSRP Reference signal received powerUL UplinkRS Reference SignalNW NetworkCN Core NetworkSL Side LinkProSe Proximity Service

[0210] The following table lists definitions of some terminology related to the solution of present disclosure.

Claims

CLAIMS1. A User Equipment, UE, comprising means for using a machine learning, ML, model at the UE for inference related to a communication procedure of a wireless network; and means for allowing switching between using the ML model at the UE and using an ML model at a network entity of the wireless network for the inference during the communication procedure.

2. The UE of claim 1 , wherein said switching comprises at least switching from using the ML model at the UE to using the ML model at the network entity or vice versa.

3. The UE of any preceding claim, further comprising means for determining occurrence of a predefined trigger; and means for signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

4. The UE of claim 1 or 2, further comprising means for determining occurrence of a predefined trigger; and means for activating or deactivating use of the ML model at the UE responsive to determining occurrence of the predefined trigger.

5. The UE of any preceding claim, further comprising means for temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

6. The UE of any preceding claim, further comprising means for collecting monitoring data during use of the ML model at the UE; and means for sharing the collected monitoring data with the network entity.

7. The UE of any preceding claim, further comprising means for sharing with the network entity prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

8. The UE of any preceding claim, wherein the communication procedure is a handover procedure.

9. The UE of claim 8, wherein the switching is performed so that the ML model at the UE is used before and after the handover procedure and the ML model at the network entity is used during the handover procedure.

10. The UE of any one of claims 1-7, wherein the communication procedure is a beam management procedure or is related to a carrier aggregation reconfiguration.11 . The UE of any preceding claim, wherein the ML model is configured to be used for a channel state information, CSI, prediction or for positioning.

12. A network entity comprising means for using a machine learning, ML, model at the network entity for inference related to a communication procedure of a wireless network, wherein the communication procedure is related to a UE; and means for allowing switching between using the ML model at the network entity and using an ML model at the UE for the inference during the communication procedure.

13. The network entity of claim 12, further comprising means for determining occurrence of a predefined trigger; and means for signaling an indication of switching between using the ML model at the UE and using the ML model at the network entity responsive to determining the occurrence of the predefined trigger.

14. The network entity of any one of claims 12-13, further comprising means for determining occurrence of a predefined trigger; and means for activating or deactivating use of the ML model at the network entity responsive to determining occurrence of the predefined trigger.

15. The network entity of any one of claims 12-14, further comprising means for temporarily allowing simultaneous use of the ML model at the UE and the ML model at the network entity in connection with switching between using the ML model at the UE and using the ML model at the network entity.

16. The network entity of any one of claims 12-15, further comprising means for sharing with the UE prediction state information in connection with switching between using the ML model at the UE and using the ML model at the network entity.

17. The network entity of any one of claims 12-14, wherein the communication procedure is a handover procedure, and the network entity is a source network entity or a target network entity of the handover procedure.

18. The network entity of claim 17, wherein the switching is performed so that the ML model at the UE is used before and after the handover procedure and the L model at the network entity is used during the handover procedure.

19. The network entity of any one of claims 12-16, wherein the communication procedure is a beam management procedure or is related to a carrier aggregation reconfiguration.

20. The network entity of any one of claims 12-19, wherein the ML model is configured to be used for a channel state information, CSI, prediction or for positioning.21 . The network entity of any one of claims 12-20, wherein said predefined trigger is determined based on one or more of the following: an event, a measurement, ML model reliability measure, energy consumption value.

Images