Uplink, downlink and joint uplink-downlink radio link failure prediction

Abstract

According to an aspect, there is provided an apparatus for performing the following. The apparatus transmits, to an access node, radio link failure, RLF, prediction capabilities of the apparatus. The RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm. The ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics. The one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

Description

UPLINK, DOWNLINK AND JOINT UPLINK-DOWNLINK RADIO LINK FAILURE PREDICTIONTECHNICAL FIELD

[0001] Various example embodiments relate to wireless communications.BACKGROUND

[0002] In cellular networks such as Fifth Generation New Radio (5G NR) networks, a radio link failure may occur in uplink and / or downlink direction for various reasons. Said reasons may include, for example, signal interference, physical obstructions appearing in the propagation path, multipath fading, network congestion, environmental factors (e.g., weather), power control problems at a terminal device or an access node. RLFs may have a significant detrimental effect on the operation of the cellular network. Moreover, overcoming an RLF once it has already occurred may be time-consuming and costly in terms of service interruptions and / or signaling overhead. Thus, it is important to take proactive actions to prevent RLFs from occurring as much as possible.SUMMARY

[0003] According to an aspect, there is provided the subject matter of the independent claims. Embodiments are defined in the dependent claims.

[0004] According to a first aspect, there is provided an apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: receiving, from an access node, a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm for predicting future RLFs in uplink, UL, and downlink, DL, directions based at least on values of one or more radio link quality metrics, wherein the RLF prediction mode is one of: an UL-DL mode for RLF prediction in the UL and DL directions, an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction;performing RLF prediction using the received RLF prediction mode of the ML algorithm; and transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction.

[0005] According to a second aspect, there is provided an apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: transmitting, to a terminal device, a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm for predicting future RLFs in uplink, UL, and downlink, DL, directions based on values of one or more radio link quality metrics, wherein the RLF prediction mode is one of: an UL-DL mode for RLF prediction in the UL and DL directions, an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction; and receiving, from the terminal device, a prediction report comprising at least some of results of an RLF prediction carried out by the terminal device using the transmitted RLF prediction mode of the ML algorithm.

[0006] According to a third aspect, there is provided a method comprising: receiving, from an access node, a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm for predicting future RLFs in uplink, UL, and downlink, DL, directions based at least on values of one or more radio link quality metrics, wherein the RLF prediction mode is one of: an UL-DL mode for RLF prediction in the UL and DL directions, an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction; performing RLF prediction using the received RLF prediction mode of the ML algorithm; and transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction.

[0007] According to a fourth aspect, there is provided a method comprising: transmitting, to a terminal device, a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm for predicting future RLFs in uplink, UL, and downlink, DL, directions based on values of one or more radio link quality metrics, wherein the RLFprediction mode is one of: an UL-DL mode for RLF prediction in the UL and DL directions, an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction; and receiving, from the terminal device, a prediction report comprising at least some of results of an RLF prediction carried out by the terminal device using the transmitted RLF prediction mode of the ML algorithm.

[0008] According to a fifth aspect, there is provided a non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: receiving, from an access node, a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm for predicting future RLFs in uplink, UL, and downlink, DL, directions based at least on values of one or more radio link quality metrics, wherein the RLF prediction mode is one of: an UL-DL mode for RLF prediction in the UL and DL directions, an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction; performing RLF prediction using the received RLF prediction mode of the ML algorithm; and transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction.

[0009] According to a sixth aspect, there is provided a non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: transmitting, to a terminal device, a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm for predicting future RLFs in uplink, UL, and downlink, DL, directions based on values of one or more radio link quality metrics, wherein the RLF prediction mode is one of: an UL-DL mode for RLF prediction in the UL and DL directions, an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction; and receiving, from the terminal device, a prediction report comprising at least some of results of an RLF prediction carried out by the terminal device using the transmitted RLF prediction mode of the ML algorithm.

[0010] According to a seventh aspect, there is provided an apparatus comprising:at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: transmitting, to an access node, radio link failure, RLF, prediction capabilities of the apparatus, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

[0011] According to an eighth aspect, there is provided an apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: receiving, from a terminal device, radio link failure, RLF, prediction capabilities of the terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

[0012] According to a ninth aspect, there is provided a method comprising: transmitting, to an access node, radio link failure, RLF, prediction capabilities of a terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

[0013] According to a tenth aspect, there is provided a method comprising: receiving, from a terminal device, radio link failure, RLF, prediction capabilities of the terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

[0014] According to an eleventh aspect, there is provided a non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: transmitting, to an access node, radio link failure, RLF, prediction capabilities of the apparatus, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

[0015] According to a twelfth aspect, there is provided a non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: receiving, from a terminal device, radio link failure, RLF, prediction capabilities of the terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

[0016] According to a thirteenth aspect, there is provided an apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: performing RLF prediction using a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction, or an UL-DL mode for RLF prediction in UL and DL directions; determining that a reporting trigger condition is satisfied based on the results of the RLF prediction; based on the determining, transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction; and in response to receiving, from the access node, a request for triggering a primary cell handover for the apparatus, triggering the primary cell handover according to the request.

[0017] According to a fourteenth aspect, there is provided an apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: performing RLF prediction using a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is an UL mode for RLF prediction in an UL direction, a DL mode for RLF prediction in a DL direction or an UL-DL mode for RLF prediction in UL and DL directions, the apparatus being configured to use a primary UL configuration of the apparatus; and determining, based on the results of the RLF prediction, that an RLF will likely occur for the UL direction; and based on the determining, either- switching from the primary UL configuration to a supplementary uplink, SUL, configuration to minimize a chance of the RLF occurring, or- transmitting, to the access node, a request requesting a switch from using the primary UL configuration to using a SUL configuration to minimize a chance of the RLF occurring, and, in response to receiving, from the access node, a confirmation for the switch to the SUL configuration, switching from the primary UL configuration to the SUL configuration.

[0018] According to a fifteenth aspect, there is provided an apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: receiving, from a terminal device, a prediction report comprising at least some of results of RLF prediction performed, at the terminal device, using an RLF prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction or an UL-DL mode for RLF prediction in UL and DL directions; determining, based on the prediction report, that an RLF will likely occur; and based on the determining, transmitting, to the terminal device, a request for triggering a primary cell handover for the terminal device.

[0019] According to a sixteenth aspect, there is provided an apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: receiving, from a terminal device, a prediction report comprising at least some of results of radio link failure, RLF, prediction performed, at the terminal device, using an RLF prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction or an UL-DL mode for RLF prediction in UL and DL directions; receiving, from the terminal device, a request requesting a switch from using a primary UL configuration of the terminal device to using a supplementary uplink, SUL, configuration to minimize a chance of an RLF occurring;determining, based on the prediction report and the request, that the RLF will likely occur for the UL direction; and transmitting, to the terminal device, a confirmation for the switch to the SUL configuration.

[0020] According to a seventeenth aspect, there is provided a method comprising: performing RLF prediction using a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction, or an UL- DL mode for RLF prediction in UL and DL directions; determining that a reporting trigger condition is satisfied based on the results of the RLF prediction; based on the determining, transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction; and in response to receiving, from the access node, a request for triggering a primary cell handover for a terminal device, triggering the primary cell handover according to the request.

[0021] According to an eighteenth aspect, there is provided a method comprising: performing RLF prediction using a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is an UL mode for RLF prediction in an UL direction, a DL mode for RLF prediction in a DL direction or an UL-DL mode for RLF prediction in UL and DL directions; and determining, based on the results of the RLF prediction, that an RLF will likely occur for the UL direction; and based on the determining, either- switching from a primary UL configuration to a supplementary uplink, SUL, configuration to minimize a chance of the RLF occurring, or- transmitting, to the access node, a request requesting a switch from using the primary UL configuration to using a SUL configuration to minimize a chance of the RLFoccurring, and, in response to receiving, from the access node, a confirmation for the switch to the SUL configuration, switching from the primary UL configuration to the SUL configuration.

[0022] According to a nineteenth aspect, there is provided a method comprising: receiving, from a terminal device, a prediction report comprising at least some of results of RLF prediction performed, at the terminal device, using an RLF prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction or an UL-DL mode for RLF prediction in UL and DL directions; determining, based on the prediction report, that an RLF will likely occur; and based on the determining, transmitting, to the terminal device, a request for triggering a primary cell handover for the terminal device.

[0023] According to a twentieth aspect, there is provided a method comprising: receiving, from a terminal device, a prediction report comprising at least some of results of radio link failure, RLF, prediction performed, at the terminal device, using an RLF prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction or an UL-DL mode for RLF prediction in UL and DL directions; receiving, from the terminal device, a request requesting a switch from using a primary UL configuration of the terminal device to using a supplementary uplink, SUL, configuration to minimize a chance of an RLF occurring; determining, based on the prediction report and the request, that the RLF will likely occur for the UL direction; and transmitting, to the terminal device, a confirmation for the switch to the SUL configuration.

[0024] According to a twenty-first aspect, there is provided a non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following:performing RLF prediction using a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction, or an UL-DL mode for RLF prediction in UL and DL directions; determining that a reporting trigger condition is satisfied based on the results of the RLF prediction; based on the determining, transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction; and in response to receiving, from the access node, a request for triggering a primary cell handover for the apparatus, triggering the primary cell handover according to the request.

[0025] According to a twenty-second aspect, there is provided a non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: performing RLF prediction using a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is an UL mode for RLF prediction in an UL direction, a DL mode for RLF prediction in a DL direction or an UL-DL mode for RLF prediction in UL and DL directions, the apparatus being configured to use a primary UL configuration of the apparatus; and determining, based on the results of the RLF prediction, that an RLF will likely occur for the UL direction; and based on the determining, either- switching from the primary UL configuration to a supplementary uplink, SUL, configuration to minimize a chance of the RLF occurring, or- transmitting, to the access node, a request requesting a switch from using the primary UL configuration to using a SUL configuration to minimize a chance of the RLF occurring, and, in response to receiving, from the access node, a confirmation for the switch to the SUL configuration, switching from the primary UL configuration to the SUL configuration.

[0026] According to a twenty-third aspect, there is provided a non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: receiving, from a terminal device, a prediction report comprising at least some of results of RLF prediction performed, at the terminal device, using an RLF prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction or an UL-DL mode for RLF prediction in UL and DL directions; determining, based on the prediction report, that an RLF will likely occur; and based on the determining, transmitting, to the terminal device, a request for triggering a primary cell handover for the terminal device.

[0027] According to a twenty-fourth, there is provided a non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: receiving, from a terminal device, a prediction report comprising at least some of results of radio link failure, RLF, prediction performed, at the terminal device, using an RLF prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction or an UL-DL mode for RLF prediction in UL and DL directions; receiving, from the terminal device, a request requesting a switch from using a primary UL configuration of the terminal device to using a supplementary uplink, SUL, configuration to minimize a chance of an RLF occurring; determining, based on the prediction report and the request, that the RLF will likely occur for the UL direction; and transmitting, to the terminal device, a confirmation for the switch to the SUL configuration.

[0028] One or more examples of implementations are set forth in more detail in the accompanying drawings and the description below. Other features will be apparent from the description and drawings, and from the claims.BRIEF DESCRIPTION OF THE DRAWINGS

[0029] FIG. 1 illustrates a system to which some embodiments may be applied;

[0030] FIG. 2 illustrates signaling between a terminal device and an access node according to some embodiments;

[0031] FIG. 3 illustrates an exemplary machine-learning algorithm which may be used in some embodiments for performing RLF prediction;

[0032] FIGs. 4A & 4B illustrate processes for joint UL-DL RLF prediction and reporting according to some embodiments;

[0033] FIGs. 5 & 6 illustrate signaling between a terminal device and an access node according to some embodiments;

[0034] FIG. 7 illustrates an exemplary machine-learning algorithm which may be used in some embodiments for measurement-based beam prediction;

[0035] FIGs. 8 and 9 illustrate signaling between a terminal device and an access node according to some embodiments; and

[0036] FIG. 10 illustrates an apparatus according to some embodiments.DETAILED DESCRIPTION OF SOME EMBODIMENTS

[0037] The following embodiments are only presented as examples. Although the specification may refer to “an”, “one”, or “some” embodiment(s) and / or example(s) in several locations of the text, this does not necessarily mean that each reference is made to the same embodiment(s) or example(s), or that a particular feature only applies to a single embodiment and / or example. Single features of different embodiments and / or examples may also be combined to provide other embodiments and / or examples.

[0038] As used herein, “at least one of the following: ” and “at least one of ” and similar wording, where the list of two or more elements are joined by “and” or “or”, mean at least any one of the elements, or at least any two or more of the elements, or at least all the elements.

[0039] In Figures to be discussed below, dashed lines are used for indicating optional features.

[0040] In the following, different exemplifying embodiments will be described using, as an example of an access architecture to which the embodiments may be applied, a radio access architecture based on long term evolution advanced (LTE Advanced, LTE-A) or new radio (NR, 5G), without restricting the embodiments to such an architecture, however. It is obvious for a person skilled in the art that the embodiments may also be applied to other kinds of communications networks having suitable means by adjusting parameters and procedures appropriately. Some examples of other options for suitable systems are the universal mobile telecommunications system (UMTS) radio access network (UTRAN or E-UTRAN), long term evolution (LTE, the same as E-UTRA), wireless local area network (WLAN or WiFi), worldwide interoperability for microwave access (WiMAX), Bluetooth®, personal communications services (PCS), ZigBee®, wideband code division multiple access (WCDMA), systems using ultra-wideband (UWB) technology, sensor networks, mobile ad- hoc networks (MANETs), Internet Protocol multimedia subsystems (IMS), rebel SIM (R-SIM) for code division multiple access (CDMA) technologies such as lx and lx evolution data optimized (IxEV-DO), global system for mobile communications (GSM), open radio access network (O-RAN) or any combination thereof.

[0041] FIG. 1 depicts examples of simplified system architectures only showing some elements and functional entities, all being logical units, whose implementation may differ from what is shown. The connections shown in FIG. 1 are logical connections; the actual physical connections may be different. It is apparent to a person skilled in the art that the system typically comprises also other functions and structures than those shown in FIG. 1.

[0042] The embodiments are not, however, restricted to the system given as an example but a person skilled in the art may apply the solution to other communication systems provided with necessary properties.

[0043] The example of FIG. 1 shows a part of an exemplifying radio access network.

[0044] A communications system typically comprises more than one (e / g)NodeB 104 in which case the (e / g)NodeBs may also be configured to communicate with one another over links, wired or wireless, designed for the purpose. These links may be used for signaling purposes. The (e / g)NodeB is a computing device configured to control the radio resources of communication system it is coupled to. The NodeB may also be referred to as a base station, an access point or any other type of interfacing device including a relay station capable of operating in a wireless environment. The (e / g)NodeB includes or is coupled to transceivers. From the transceivers of the (e / g)NodeB, a connection is provided to an antenna unit that establishes bi-directional radio links to user devices. The antenna unit may comprise a plurality of antennas or antenna elements. The (e / g)NodeB is further connected to core network 110 (CN or next generation core NGC). Depending on the system, the counterpart on the CN side can be a serving gateway (S-GW, routing and forwarding user data packets), packet data network gateway (P-GW), for providing connectivity of user devices (UEs) to external packet data networks, or mobile management entity (MME), etc.

[0045] The user device 100, 102 (also called UE, user equipment, user terminal, terminal device, etc.) illustrates one type of an apparatus to which resources on the air interface are allocated and assigned, and thus any feature described herein with a user device may be implemented with a corresponding apparatus, such as a relay node. An example of such a relay node is a layer 3 relay (self-backhauling relay) towards the base station. The user equipment may comprise a mobile equipment and at least one universal integrated circuit card (UICC).

[0046] The user device 100, 102 typically refers to a portable computing device that includes wireless mobile communication devices operating with or without a subscriber identity (or identification) module (SIM) or UICC, including, but not limited to, the following types of devices: a mobile station (mobile phone), smartphone, personal digital assistant (PDA), handset, device using a wireless modem (alarm or measurement device, etc.), laptop and / or touch screen computer, tablet, game console, notebook, and multimedia device. Here, the SIM may be a physical SIM which may be removable by a user or an embedded SIM (eSIM) embedded directly into the user device 100, 102 (and thus not being removable by a user). It should be appreciated that a user device may also be a nearly exclusive uplink only device, of which an example is a camera or video camera loading images or video clips to a network. A user device may also be a device having capability to operate in Internet of Things (loT) network which is a scenario in which objects are provided with the ability to transfer data over a network without requiring human-to-human or human-to-computer interaction. Thus,the user devices may not enable direct user interaction or may enable only limited user interaction (e.g., during setup). The user device (or in some embodiments a layer 3 relay node) is configured to perform one or more of user equipment functionalities. The user device may also be called a terminal device, a subscriber unit, mobile station, remote terminal, access terminal, user terminal or user equipment (UE) just to mention but a few names or apparatuses. Each user device 100, 102 may comprise one or more antennas.

[0047] Various techniques described herein may also be applied to a cyber-physical system (CPS) (a system of collaborating computational elements controlling physical entities). CPS may enable the implementation and exploitation of massive amounts of interconnected ICT devices (sensors, actuators, processors microcontrollers, etc.) embedded in physical objects at different locations. Mobile cyber physical systems, in which the physical system in question has inherent mobility, are a subcategory of cyber-physical systems. Examples of mobile physical systems include mobile robotics and electronics transported by humans or animals.

[0048] Additionally, although the apparatuses have been depicted as single entities, different units, processors and / or memory units (not all shown in FIG. 1) may be implemented.

[0049] 5G enables using MIMO antennas, many more base stations or nodes than the LTE (a so-called small cell concept), including macro sites operating in co-operation with smaller stations and employing a variety of radio technologies depending on service needs, use cases and / or spectrum available. 5G mobile communications supports a wide range of use cases and related applications including video streaming, augmented reality, different ways of data sharing and various forms of machine type applications, including vehicular safety, different sensors and real-time control. 5G is expected to have multiple radio interfaces, namely below 6GHz, cmWave and mmWave, and also being integrable with existing legacy radio access technologies, such as the LTE. Integration with the LTE may be implemented, at least in the early phase, as a system, where macro coverage is provided by the LTE and 5G radio interface access comes from small cells by aggregation to the LTE. In other words, 5G is planned to support both inter-RAT operability (such as LTE-5G) and inter-RI operability (inter-radio interface operability, such as below 6GHz - cmWave, below 6GHz - cmWave - mmWave). One of the concepts considered to be used in 5G networks is network slicing in which multiple independent and dedicated virtual sub-networks (network instances) may be created within thesame infrastructure to run services that have different requirements on latency, reliability, throughput and mobility.

[0050] The current architecture in LTE networks is fully distributed in the radio and fully centralized in the core network. The low latency applications and services in 5G require to bring the content close to the radio which leads to local break out and multi-access edge computing (MEC). 5G enables analytics and knowledge generation to occur at the source of the data. This approach requires leveraging resources that may not be continuously connected to a network such as laptops, smartphones, tablets and sensors. MEC provides a distributed computing environment for application and service hosting. It also has the ability to store and process content in close proximity to cellular subscribers for faster response time. Edge computing covers a wide range of technologies such as wireless sensor networks, mobile data acquisition, mobile signature analysis, cooperative distributed peer-to-peer ad hoc networking and processing also classifiable as local cloud / fog computing and grid / mesh computing, dew computing, mobile edge computing, cloudlet, distributed data storage and retrieval, autonomic self-healing networks, remote cloud services, augmented and virtual reality, data caching, Internet of Things (massive connectivity and / or latency critical), critical communications (autonomous vehicles, traffic safety, real-time analytics, time-critical control, healthcare applications).

[0051] The communication system is also able to communicate with other networks, such as a public switched telephone network or the Internet 112, or utilize services provided by them. The communication network may also be able to support the usage of cloud services, for example at least part of core network operations may be carried out as a cloud service (this is depicted in FIG. 1 by “cloud” 114). The communication system may also comprise a central control entity, or a like, providing facilities for networks of different operators to cooperate for example in spectrum sharing.

[0052] Edge cloud may be brought into the RAN by utilizing network function virtualization (NVF) and software defined networking (SDN). Using edge cloud may mean access node operations to be carried out, at least partly, in a server, host or node operationally coupled to a remote radio head or unit (RU) 116, 118 or base station comprising radio parts. It is also possible that node operations will be distributed among a plurality of servers, nodes or hosts. Application of cloudRAN architecture enables RAN real time functions being carried out at the RAN side (in a distributed unit, DU 104) and non-real time functions being carriedout in a centralized manner (in a central or centralized unit, CU 108). Thus, in summary, the RAN may comprise, in some embodiments, at least one distributed access node comprising a central unit 108, one or more distributed units 104 communicatively connected to the central unit 108 and one or more (remote) radio heads or units 116, 118, each of which is communicatively connected to at least one of the one or more distributed units 104.

[0053] In some embodiments, element 104 is a non-distributed access node. In such embodiments, RUs 116, 118 are omitted.

[0054] It should also be understood that the distribution of labor between core network operations and base station operations may differ from that of the LTE or even be non-existent. Some other technology advancements probably to be used are Big Data and all-IP, which may change the way networks are being constructed and managed. 5G (or new radio, NR) networks are being designed to support multiple hierarchies, where MEC servers can be placed between the core and the base station or nodeB (gNB). It should be appreciated that MEC can be applied in 4G networks as well.

[0055] 5G may also utilize satellite communication to enhance or complement the coverage of 5G service, for example by providing backhauling. Possible use cases are providing service continuity for machine-to-machine (M2M) or Internet of Things (loT) devices or for passengers on board of vehicles, or ensuring service availability for critical communications, and future rail-way / maritime / aeronautical communications. Satellite communication may utilize geostationary earth orbit (GEO) satellite systems, but also low earth orbit (LEO) satellite systems, in particular mega-constellations (systems in which hundreds of (nano)satellites are deployed). Each satellite 106 in the mega-constellation may cover several satellite-enabled network entities that create on-ground cells. The on-ground cells may be created through an on-ground relay node 104 or by a gNB located on-ground or in a satellite.

[0056] It is obvious for a person skilled in the art that the depicted system is only an example of a part of a radio access system and in practice, the system may comprise a plurality of (e / g)NodeBs, the user device may have an access to a plurality of radio cells and the system may comprise also other apparatuses, such as physical layer relay nodes or other network elements, etc. At least one of the (e / g)NodeBs or may be a Home(e / g)nodeB. Additionally, in a geographical area of a radio communication system a plurality of different kinds of radio cells as well as a plurality of radio cells may be provided. Radio cells may be macro cells (or umbrella cells) which are large cells, usually having a diameter of up to tens of kilometers, orsmaller cells such as micro-, femto- or picocells. The (e / g)NodeBs of FIG. 1 may provide any kind of these cells. A cellular radio system may be implemented as a multilayer network including several kinds of cells. Typically, in multilayer networks, one access node provides one kind of a cell or cells, and thus a plurality of (e / g)NodeBs are required to provide such a network structure.

[0057] For fulfilling the need for improving the deployment and performance of communication systems, the concept of “plug-and-play” (e / g)NodeBs has been introduced. Typically, a network which is able to use “plug-and-play” (e / g)NodeBs, includes, in addition to Home (e / g)NodeBs (H(e / g)nodeBs), a home node B gateway, or HNB-GW (not shown in FIG. 1). A HNB Gateway (HNB-GW), which is typically installed within an operator’s network may aggregate traffic from a large number of HNBs back to a core network.

[0058] 6G architecture is targeted to enable easy integration of everything, such as a network of networks, joint communication and sensing, non-terrestrial networks and terrestrial communication. 6G systems are envisioned to encompass machine learning algorithms as well as local and distributed computing capabilities, where virtualized network functions can be distributed over core and edge computing resources. Far edge computing, where computing resources are pushed to the very edge of the network, will be part of the distributed computing environment, for example in “zero-delay” scenarios. Some 5G systems may also employ such capabilities. More generally, the actual (radio) communication system is envisaged to be comprised of one or more computer programs executed within a programmable infrastructure, such as general-purpose computing entities (servers, processors, and like).

[0059] 6G networks are expected to adopt flexible decentralized and / or distributed computing systems and architecture and ubiquitous computing, with local spectrum licensing, spectrum sharing, infrastructure sharing, and intelligent automated management underpinned by mobile edge computing, artificial intelligence, short-packet communication, distributed ledgers and blockchain technologies. Key features of 6G will include intelligent connected management and control functions, programmability, integrated sensing and communication, reduction of energy footprint, trustworthy infrastructure, scalability and affordability. In addition to these, 6G is also targeting new use cases covering the integration of localization and sensing capabilities into system definition to unifying user experience across physical and digital worlds.

[0060] Embodiments to be discussed below are directed to prediction of radio link failures (RLFs) in wireless cellular networks such as the one illustrated in FIG. 1. In such networks, a radio link failure may occur in uplink and / or downlink direction for various reasons such as signal interference, physical obstructions appearing in the propagation path, multipath fading, network congestion, environmental factors (e.g., weather), power control problems at a terminal device or an access node. In the case of downlink (DE) communications, a radio link failure may be declared by a terminal device, for example, based on monitoring of out-of-sync indications associated with a particular DL reference signal (e.g., synchronization signal block, SSB, or demodulation reference signal, DMRS), signal quality (e.g., reference signal received power, RSRP, or reference signal received quality, RSRQ, or signal-to-interference -plus-noise Ratio, SINR). Namely, in response to reception of a pre-defined number of consecutive lower layer out-of-sync indications at the terminal device (as determined by an N310 counter for counting consecutive out-of-sync indications), the terminal device may start a T310 timer. In response to an expiration of the T310 timer, the terminal device may detect a radio link failure. In the case of uplink (UE) communications, a radio link failure may be declared by a terminal device, for example, in response to reaching a pre-defined number of UL radio link control (RLC) retransmissions or a pre-defined number of preamble transmissions (i.e., a pre-defined number of random access attempts). In the case of UL communications, a radio link failure may also be declared by the access node, for example, in response to the received signal power or quality falling below a reference quality metric that may be defined by the access node implementation.

[0061] RLFs may have a significant detrimental effect on the operation of the wireless cellular network. Moreover, overcoming an RLF once it has already occurred may be timeconsuming and costly in terms of service interruption and / or signaling overhead. For example, in the case of the DL RLF example mentioned above, a terminal device experiences poor radio link quality (i.e., reduced throughput) both during the counting of the N310 counter and during the running of the T310 timer. After the detection of the RLF, the terminal device still needs to perform radio resource control (RRC) re-establishment which results in additional service interruption and signaling overhead.

[0062] To overcome the aforementioned issues, machine-learning (ML) or artificial intelligence (Al) based solutions at the terminal device may be employed for predicting RLFs before they happen and taking appropriate proactive actions to either avoid the RLF altogether (e.g., by triggering a handover earlier) or to at least minimize the impact of the RLF (e.g., byperforming re-establishment earlier and / or faster). The ML / Al solutions employed may be, for example, supervised learning solutions (i.e., ML algorithms which are trained on a labeled dataset). Examples of supervised learning solutions include, e.g., artificial neural network based solutions (e.g., multilayer perceptron), support-vector machines (SVMs), linear regression, logistic regression, decision trees and k-nearest neighbors (k-NN) methods.

[0063] The embodiments to be discussed below relate in particular to differentiation between UL-based and DL-based ML RLF prediction. Namely, the ML-based RLF prediction at the terminal device may benefit from supplementing and combining aspects of both DL and UL based RLF predictions for improving the effectiveness of responding to different radio link failure scenarios. Providing information on terminal device capabilities relating to UL and DL based RLF prediction to the network in advance can ensure that the network is able to provide tailored configurations and trigger appropriate actions and, thus, also enable the terminal device to respond effectively and efficiently to any upcoming RLFs. Different aspects covered by the embodiments comprise, for example, terminal device capability reporting, signaling aspects for the network configuration and terminal device behavior and DL and UL or joint UL-DL RLF prediction triggering and reporting.

[0064] FIG. 2 illustrates signaling between a terminal device and an access node (or equally a network node or entity) for performing RLF detection according to some embodiments. The terminal device may be any of the terminal devices 100, 102 of FIG. 1. The access node may be a non-distributed access node such as the access node 104 of FIG. 1 or a distributed access node or a particular unit thereof (e.g., the DU 104 or the CU 108 of FIG. 1).

[0065] Referring to FIG. 2, it may be assumed initially that the terminal device is maintaining, in at least one memory comprised in or connected to the terminal device, a machine-learning (ML) algorithm for predicting future RLFs in UL and / or DL directions based on at least on values of one or more radio link quality metrics (or values of one or more UL radio link quality metrics and / or one or more DL radio link quality metrics, respectively). In other words, the one or more radio link quality metrics correspond to features of the ML algorithm while a label (i.e., the output) of the ML algorithm is an RLF prediction result (e.g., having a binary true / false value indicating positive / negative RLF detection). The RLF prediction may relate to one or more (future) time windows (which may optionally be defined separately for UL and DL). In other words, the output of the ML algorithm may indicatewhether an UL and / or DL RLF is likely (according to some pre-defined standard for likelihood) to occur within the one or more (future) time windows.

[0066] The one or more radio link quality metrics, corresponding to features of the ML algorithm, may comprise one or more UL radio link quality metrics and / or one or more DL radio link quality metrics. The one or more UL radio link quality metrics and the one or more DL radio link quality metrics may be fully different sets of metrics or be sets sharing at least some of the same metrics. In some embodiments, the one or more UL radio link quality metrics and the one or more DL radio link quality metrics may be used, respectively, for UL and DL RLF prediction though, in other embodiments, at least some of the one or more DL radio link quality metrics may have an effect on the UL RLF prediction and / or at least some of the one or more UL radio link quality metrics may have an effect on the DL RLF prediction. The one or more UL radio link quality metrics may comprise, for example, at least one of a number of attempted RLC retransmissions, a number of random access attempts (or a number of associated preamble transmissions), an RLC acknowledgment latency, power headroom, a hybrid automatic repeat request (HARQ) statistics metric (e.g., the number of HARQ positive acknowledgments, ACK, and / or the number of negative HARQ acknowledgements, NACK) or an UL power control parameter (e.g., a physical uplink shared channel, PUSCH, transmit power). The one or more DL radio link quality metrics may comprise, for example, at least one of a DL block error rate (BLER), a DL power control parameter (e.g., a transmit power control, TPC, command), a reference signal received power (RSRP), a reference signal received quality (RSRQ), a DL beam measurement metric (e.g., signal strength or quality, such as RSRP or RSRQ, per beam), a DL cell quality measurement metric (e.g., signal strength or quality, such as RSRP or RSRQ, per cell), or a number of consecutive out-of-sync indications associated with a DL reference signal (e.g., as indicated by the N310 counter).

[0067] As mentioned above, features (i.e., inputs) of the ML algorithm comprise at least the one or more radio link quality metrics (comprising, e.g., one or more UL radio link quality metrics and / or one or more DL radio link quality metrics). In some embodiments, the features of the ML algorithm may further comprise one or more kinematic metrics of the terminal device. The one or more kinematic metrics may comprise a (current) location of the terminal device and / or a (current) speed (or velocity) of the terminal device. In some cases, the one or more kinematic metrics may further comprise an altitude of the terminal device and / or an acceleration of the terminal device. In some embodiments, the one or more kinematic metricsof the terminal device may be used, by the ML algorithm, for one of the UL or DL RLF detection but not for the other.

[0068] In some embodiments, features (i.e., inputs) of the ML algorithm may further comprise an indication for using a line of sight (LOS) propagation channel or a non-line of sight (NLOS) propagation channel.

[0069] The ML algorithm may be assumed to support a plurality of different RLF prediction modes. Said plurality of RLF prediction modes may comprise at least two (or three) of an UL mode for RLF prediction in an UL direction, a DL mode for RLF prediction in a DL direction or an UL-DL mode for RLF prediction in both UL and DL directions. A single RLF prediction mode may be active at any given time. Each RLF prediction mode may be associated with a different set of features (i.e., inputs) comprising, each, one or more radio link quality metrics and, optionally, one or more kinematic metrics (e.g., a location and / or speed or velocity of the terminal device).

[0070] In some embodiments, the ML algorithm may comprise separate UL and DL ML models. In other words, the ML algorithm may comprise an UL ML model configured to predict the future RLFs in the UL direction based on values of said one or more UL radio link quality metrics. In some embodiments, the prediction of the future RLFs using the UL ML model may be, additionally or alternatively, based on said one or more DL radio link quality metrics as poor DL radio link quality may, in most cases, be assumed to be also indicative of poor UL radio link quality due to radio channel reciprocity (e.g., in time division duplex, TDD). Additionally, the prediction of the future RLFs using the UL ML model may, in some embodiments, be based on the one or more kinematic metrics of the terminal device comprising, e.g., location and / or speed (or velocity) of the terminal device. Additionally or alternatively, the ML algorithm may comprise a DL ML model configured to predict the future RLFs in the DL direction based on values of said one or more DL radio link quality metrics. In some embodiments, the prediction of the future RLFs using the DL ML model may be, additionally or alternatively, based on one or more UL radio link quality metrics (as poor UL radio link quality may, in most cases, be assumed to be also indicative of poor DL radio link quality due to reciprocity). Additionally, the prediction of the future RLFs using the DL ML model may, in some embodiments, be based on the one or more kinematic metrics comprising, e.g., the location and / or speed (or velocity) of the terminal device. The UL and DL modes ofthe ML algorithm may employ, respectively, the UL and DL ML models of the ML algorithm while the UL-ML mode may employ both of the UL and DL models of the ML algorithm.

[0071] Instead of employing two separate ML models for UL and DL RLF detection, in some embodiments, the ML algorithm of the terminal device may comprise a (joint) UL-DL ML model configured to predict the future RLFs in the UL direction based on values of one or more UL radio link quality metrics and / or one or more DL radio link quality metrics and to predict the future RLFs in the DL direction based on values of the one or more UL radio link quality metrics and / or the one or more DL radio link quality metrics. In such a case, the same UL-DL ML model may be employed irrespective of the RLF prediction mode used.

[0072] In some embodiments, the ML algorithm (and, thus, any of the UL ML model, the DL ML model or the UL-DL ML model defined therein) may be based on supervised learning. For example, the ML algorithm may be based on one or more recurrent neural networks (RNNs). The one or more RNNs may comprise, e.g., one or more long short-term memories (LSTMs) and / or one or more gated recurrent units (GRUs). RNNs enable efficient processing memory and / or historical information especially when it comes to radio signals where analysis of several time steps of radio signals is typically required in order to confirm a prediction. RNNs are efficient type of ML model especially for processing sequential information (e.g., time series). Due to the depth of memory elements, they can evaluate past information and temporal dependencies. RNNs are widely used in time series analysis and other sequential data problems, where it is required to predict future values or to recognize patterns in sequential data. In general, the ML algorithm may be based on at least one of: one or more RNNs, an artificial neural network based solution (e.g., multilayer perceptron), one or more support-vector machines (SVMs), linear regression, logistic regression, one or more decision trees or a k-nearest neighbors (k-NN) method. One exemplary implementation of the ML algorithm using a GRU implementation is discussed below in detail in connection with FIG. 3.

[0073] The ML algorithm may have been (generated and) trained previously by the terminal device or by some other device or entity (e.g., by another terminal device, an access node or a core network entity). The ML algorithm may have been trained based on history information. Said history information may comprise: a first plurality of sets of values of one or more UL radio link quality metrics and / or one or more DL radio link quality metrics corresponding to a first plurality of differentpast time instances and information on whether or not each of the first plurality of sets of values lead to an UL RLF within a first pre-defined time window, and / or• a second plurality of sets of values of the one or more DL radio link quality metrics and / or the one or more UL radio link quality metrics corresponding to a second plurality of different past time instances and information on whether or not each of the second plurality of sets of values lead to a DL RLF within a second pre-defined time window.The one or more UL and DL radio link quality metrics may be defined here as described above. In some embodiments, the first plurality of sets may comprise at least values of the one or more UL radio link quality and / or the second plurality of sets may comprise at least values of the one or more DL radio link quality. In some embodiments, the first and / or second plurality of sets may further comprise values of the one or more kinematic metrics such as a location and / or a speed (or velocity) of the terminal device.

[0074] As a first step of FIG. 2., the access node transmits, in message 201, an RLF prediction mode of the ML algorithm of the terminal device. As mentioned above, the ML algorithm of the terminal device is configured to predict future RLFs in UL and / or DL directions based at least on said values of one or more radio link quality metrics. Here, the RLF prediction mode (currently in use) is one of: the UL mode for RLF prediction in the UL direction, the DL mode for the RLF prediction in the DL direction or the UL-DL mode for the RLF prediction in both UL and DL directions. Message 201 serves to enable the network to make a selection regarding which RLF prediction mode the terminal device should employ, i.e., which type of RLF predictions the terminal device should carry out. The message 201 may be an RRC reconfiguration message comprising the RLF prediction mode.

[0075] In some embodiments, message 201 may comprise further configuration information relating to the RLF prediction and / or reporting thereof. Said further configuration information may specify, for example, under which conditions should the RLF prediction and / or reporting be triggered and / or how should the values of the features of the ML algorithm be obtained (e.g., directly via measurements or via measurement-based predictions) and / or how should the ML algorithm be configured (e.g., which values should the hyperparameters of ML algorithm have). These functionalities will be described in further detail in connection with FIG. 5.

[0076] The terminal device receives, in block 202, from the access node, the RLF prediction mode of the ML algorithm. Then, the terminal device performs, in block 203, RLFprediction using the received RLF prediction mode of the ML algorithm. Namely, the terminal device may obtain (current or most recent) values of the one or more radio link quality metrics (and optionally one or more kinematic metrics of the terminal device) and apply them to the ML algorithm as inputs. The obtaining of the (current) values of the one or more radio link quality metrics may comprise carrying out (radio) measurements of the one or more radio link quality metrics and / or retrieving, from a memory, results of most recently performed measurements. The measurements may relate to one or more different beams associated with one or more different cells. In some embodiments, measurement-based predictions (e.g., ML- based predictions based on measurement data) may be employed, in addition to direct measurements, as will be discussed in more detail in connection with FIG. 7. Similarly, the one or more kinematic metrics of the terminal device (e.g., a location and / or speed or velocity of the terminal device) may be measured by the terminal device, for example, using global positioning system (GPS) receiver, a Wi-Fi based positioning scheme or one or more sensors (e.g., a motion sensor, a speed sensor and / or an accelerometer).

[0077] In some embodiments, initiating or triggering of the RLF prediction of block 203 using the received RLF prediction mode of the ML algorithm may be carried out according to an UL (prediction) trigger condition and / or a DL (prediction) trigger condition (depending on the RLF prediction mode). In other words, the RLF prediction of block 203 may be carried out in response to an UL trigger condition and / or a DL trigger condition being satisfied. If the RLF prediction mode is the UL mode or the DL mode, the UL trigger condition or the DL trigger condition is evaluated, respectively. If the RLF prediction mode is the UL-DL mode, both UL & DL trigger conditions may be evaluated so that, depending on the embodiment, only one or both of the UL & DL trigger conditions may need to be satisfied in order for the RLF prediction using the UL-ML mode to be triggered. In general, the UL or DL trigger condition may be triggered when it is detected that UL or DL radio link quality has dropped below a pre-defined limit.

[0078] The UL trigger condition may be based on at least one of: a number of attempted RLC retransmissions, a number of random access attempts or a power headroom. The UL trigger condition may comprise a separate criterion (e.g., a threshold) for any of the listed quantities (i.e., the satisfying of the UL trigger condition may require satisfying multiple criteria relating to respective multiple different quantities). In some embodiments, a plurality of independent UL trigger conditions may be configured (e.g., one for the number of attempted RLC retransmissions, one for the number of random access attempts and / or one for the powerheadroom), where satisfying of any one of the plurality of independent UL trigger conditions may trigger the UL RLF prediction. Each of the plurality of independent UL trigger conditions may be based on at least one of: a number of attempted RLC retransmissions, a number of random access attempts or a power headroom. Thus, the UL trigger condition may be satisfied, e.g., if the number of attempted RLC retransmissions exceeds an associated threshold, a number of random access attempts exceeds an associated threshold and / or a power headroom falls below an associated threshold.

[0079] The DL trigger condition may be based on at least one of: a DL serving cell power, a DL target cell power or a number of consecutive out-of-sync indications associated with a DL reference signal (e.g., SSB or DMRS). The DL trigger condition may comprise a separate criterion (e.g., a threshold) for any of the listed quantities (i.e., the satisfying of the DL trigger condition may require satisfying multiple criteria relating to respective multiple different quantities. In some embodiments, a plurality of independent DL trigger conditions may be configured (e.g., one for the DL serving cell power, one for the DL target cell power and / or one for the number of consecutive out-of-sync indications associated with a DL reference signal), where satisfying of any one of the plurality of independent DL trigger conditions may trigger the DL RLF prediction. Each of the plurality of independent UL trigger conditions may be based on at least one of: a DL serving cell power, a DL target cell power or a number of consecutive out-of-sync indications associated with a DL reference signal. Thus, the DL trigger condition may be satisfied, e.g., if the DL serving cell power falls below an associated threshold, the DL target cell power exceeds an associated threshold, the DL target cell power exceeds the DL serving cell power, and / or the number of consecutive out-of-sync indications associated with a DL reference signal exceeds an associated threshold.

[0080] In some embodiments, the UL trigger condition and / or the DL trigger condition may have been transmitted from the access node to the terminal device simultaneous to or before the transmitting of the RLF prediction mode to be used by the terminal device (i.e., in message 201 or before message 201).

[0081] The terminal device transmits, in message 204, to the access node, a prediction report comprising at least some of results of the RLF prediction.

[0082] Said at least some of the results of the RLF prediction included in the prediction report (i.e., message 204) may comprise one or more probabilities for an UL RLF within respective one or more first pre-defined future time windows and / or one or more probabilitiesfor a DL RLF within respective one or more second pre-defined future time windows (depending on the RLF prediction mode used).

[0083] Additionally or alternatively, said at least some of the results of the RLF prediction included in the prediction report (i.e., message 204) may comprise one or more indications (e.g., one or more binary flags) that an UL RLF or no UL RLF will likely occur within respective one or more first pre-defined future time windows and / or one or more indications that an DL RLF or no DL RLF will likely occur within respective one or more second pre-defined future time windows.

[0084] Additionally or alternatively, said at least some of the results of the RLF prediction included in the prediction report (i.e., message 204) may comprise one or more probabilities that an UL radio link quality falls below a first pre-defined radio link quality threshold within respective one or more first pre-defined future time window and / or one or more probabilities that a DL radio link quality falls below a second pre-defined radio link quality threshold within respective one or more second pre-defined future time window. Here, the UL radio link quality may be quantified using (or correspond to), e.g., any of the one or more UL radio link quality metrics (and possibly the one or more DL radio link quality metrics) and / or any of the parameters listed in connection with UL (and / or DL) trigger conditions. Moreover, the DL radio link quality may be quantified using (or correspond to), e.g., any of the one or more DL radio link quality metrics (and possibly the one or more UL radio link quality metrics), and / or any of the parameters listed in connection with DL (and / or UL) trigger conditions

[0085] In some embodiments, the prediction report (i.e., message 204) may further comprise a confidence level and / or an accuracy for each or at least some of included probabilities or for each or at least some of included indications. This enables the access node to make more informed decisions regarding how to act to avoid RLFs while minimizing unnecessary actions (i.e., actions carried out based on predicted RLFs associated low confidence level or accuracy).

[0086] In some embodiments, the prediction report (i.e., message 204) may comprise at least one of: a float value (i.e., a scalar) that indicates the probability of the RLF in the upcoming time window (Tpredict),• float values (e.g., provided as a vector) that indicate the probabilities of the RLF in the upcoming one or more (consecutive) time windows (Tpredict,i, Tpredict,2, . . .) such as [0 - 100ms], [100 - 200ms], ..., [900- 1000ms], or• a 1-bit indication or an / V-bit indication indicating the probability of the RLF in the upcoming time window (Tpredict) or the upcoming one or more (consecutive) time windows (Tpredict,i , Tpredict, 2,. . .) (N being any positive integer).In the case of the last bullet point, the prediction report may be transmitted, e.g., as a medium access control control element (MAC CE) or via the physical uplink control channel (PUCCH).

[0087] In some embodiments, the prediction report (i.e., message 204) may comprise information on out-of-sync indications (or at least consecutive out-of-sync indications) detected by the terminal device. In some embodiments, the prediction report may further comprise information on in-sync indications.

[0088] In some embodiments, initiating or triggering of the transmission of the prediction report (i.e., message 204) may be carried out according to a reporting trigger condition. In other words, the transmission of message 204 may be carried out in response to the reporting trigger condition being satisfied. The reporting trigger condition may be based on (i.e., the satisfying of the reporting trigger condition may be dependent on) at least one of: a predicted probability for an UL RLF, a predicted probability for a DL RLF, an accuracy or confidence level of an UL RLF prediction or an accuracy or confidence level of a DL RLF prediction. The definition of the reporting trigger condition may be dependent on the RLF prediction mode used.

[0089] Specifically, if the RLF prediction in block 203 is performed using the UL mode, the reporting trigger condition may be based on the predicted probability for the UL RLF and / or the accuracy or confidence level of the UL RLF prediction. For example, the reporting trigger condition may be satisfied in response the predicted probability for the UL RLF exceeding a pre-defined probability threshold or in response the predicted probability for the UL RLF exceeding a pre-defined probability threshold while the accuracy or confidence level of the UL RLF prediction also exceeds a pre-defined accuracy or confidence level threshold.

[0090] If the RLF prediction in block 203 is performed using the DL mode, the reporting trigger condition may be based on the predicted probability for the DL RLF and / or the accuracy or confidence level of the DL RLF prediction. For example, the reporting trigger condition maybe satisfied in response the predicted probability for the DL RLF exceeding a pre-defined probability threshold or in response the predicted probability for the DL RLF exceeding a predefined probability threshold while the accuracy or confidence level of the DL RLF prediction also exceeds a pre-defined accuracy or confidence level threshold.

[0091] If the RLF prediction in block 203 is performed using the UL-DL mode, the reporting trigger condition may be based on at least one of the predicted probability for the UL RLF, the predicted probability for a DL RLF, the accuracy or confidence level of the UL RLF prediction or the accuracy or confidence level of the DL RLF prediction. For example, the reporting trigger condition may be satisfied in response the predicted probability for one or both of the UL RLF and DL RLF exceeding a pre-defined probability threshold or in response the predicted probability for one or both of the UL RLF and DL RLF exceeding a pre-defined probability threshold while the accuracy or confidence level of said one or both of the UL RLF and DL RLF also exceeds a pre-defined accuracy or confidence level threshold.

[0092] In some embodiments, the terminal device may include results of measurements of the serving and / or non-serving frequencies in the prediction report.

[0093] In some embodiments, the RLF prediction and the reporting of the prediction report may be carried out periodically or regularly.

[0094] In some embodiments, satisfying of the reporting trigger condition may also trigger periodic RLF predictions and reporting of said RLF predictions (that is, periodic repetitions of blocks 203, 204). Alternatively, the triggering of the periodic RLF predictions and reporting of said RLF predictions may be triggered based on a separate periodic reporting trigger condition. The periodic reporting trigger condition may be defined similar to as described for the reporting trigger condition though it may be defined to be more stringent compared to the reporting trigger condition (e.g., higher probability threshold may need to be exceeded for periodic reporting). The periodic reporting may be terminated in response to an RLF prediction failing to satisfy the reporting trigger condition. Alternatively, the periodic reporting may be terminated in response to an RLF satisfying a reporting termination condition. For example, the reporting trigger condition may be defined to be based on (i.e., dependent on) a predicted probability of the RLF prediction (and optionally the accuracy or confidence level of the RLF prediction) while the reporting termination condition may be defined to be based on (i.e., dependent on) the accuracy or confidence level of the RLF prediction. Thus, thetermination of the periodic reporting could occur, e.g., when the accuracy or confidence level falls below a pre-defined threshold.

[0095] The access node receives, in block 205, the prediction report from the terminal device. Thereafter, the access node may determine, in block 206, based at least on the results of the RLF prediction, that an RLF will (likely) occur. Specifically, the access node may determine, in block 206, based at least on the results of the RLF prediction, that an RLF will (likely) occur in UL and / or DL. The determination in block 206 may be further based on one or more pre-defined criteria for positive RLF detection. Said one or more pre-defined criteria may comprise, for example, a pre-defined threshold for a predicted probability indicative of the RLF. Additionally, the one or more pre-defined criteria may comprise a pre-defined threshold for acceptable accuracy or confidence level of the RLF prediction. Any of said predefined thresholds may be defined separately or jointly for UL and DL. Thus, in summary, the positive RLF detection for UL may be made if the predicted probability for the UL RLF and the accuracy or confidence level for that UL RLF prediction exceed respective pre-defined thresholds. Similarly, the positive RLF detection for DL may be made if both the predicted probability for the DL RLF and the accuracy or confidence level for that DL RLF prediction exceed respective pre-defined thresholds. In other embodiments, the accuracy or confidence levels may not be taken into account.

[0096] Based at least on the determining in block 206 (i.e., based on detecting an upcoming RLF in UL and / or DL), the access node may cause or trigger, in message 207, reconfiguration of the terminal device for minimizing a chance of the RLF occurring (in DL and / or UL). The terminal device carries out, in block 208, appropriate actions for executing the reconfiguration. Here, the reconfiguration may correspond, e.g., to switching of a primary cell of the terminal device or switching to use a supplementary uplink (SUL) configuration, instead of a primary uplink configuration. These functionalities are discussed in further detail in connection with FIGs. 8 & 9.

[0097] FIG. 3 illustrates one non-limiting example of an ML architecture 300 which may be used for performing the RLF prediction of embodiments (e.g., the RLF prediction of block 203 of FIG. 2). Specifically, FIG. 3 shows an example of neural network based supervised learning architecture 300 based on use of a gated recurrent unit (GRU) 303 usable for both UL and DL RLF detection. The architecture 300 comprises a hidden layer 301 , a first convolutional layer 302, a second convolutional layer 303, a GRU 304, a first linear layer 305 and a secondlinear layer 309. Elements 306 to 308 indicate N iterations in a time series as produced by the GRU 304.

[0098] The collection of data for training of the ML algorithm 300 may be carried out as follows. For DL RLF detection, a terminal device may be in charge of data collection for the training of the ML algorithm 300. Beam-specific received signal strength (e.g., beam-specific RSRP or RSRQ) for a beam codebook (e.g., number of antenna elements and their arrangement) and cell quality (e.g., cell-specific RSRP or RSRQ based on measurements for one or more beams of the cell) for a serving cell may be collected periodically (periodicity could be SSB periodicity e.g., 40 ms). These quantities serve as the features of the ML algorithm for DL in this non-limiting example. The DL data collection may be initiated or triggered upon a certain DL data collection condition being satisfied. For example, the data collection may be initiated or triggered:• upon receiving a pre-defined number of consecutive out-of-sync indications from the physical layer (as determined by the N310 counter for counting consecutive out-of-sync indications), or• upon a cell quality as determined based on measured or predicted reference signals (or symbols) falling below a pre-defined level, or• upon starting a beam failure recovery procedure.

[0099] At the time of (positive) RLF detection, the terminal device may label the associated previously collected data with an indication “RLF=TRUE”. If no RLF event occurs within a pre-defined time window starting from the RLF detection time instance (e.g., owing to cancellation due to receipt of N311 consecutive in-sync indications from the physical layer), the terminal device may mark the associated previously collected data with an indication “RLF=FALSE”.

[0100] For UL RLF prediction, the terminal device may be in charge of data collection though the data collected by the terminal device may be augmented further by the access node (as the access node also measures the UL quality). The UL data collection may be initiated or triggered upon a certain UL data collection condition being satisfied. The UL data collection condition may be satisfied when a number of RLC acknowledged mode (AM) retransmissions or number of random access attempts reaching a respective pre-defined threshold or powerheadroom falls below a pre-defined threshold. Collected measurements are the same as in the DL RLF case.

[0101] Values of one or more kinematic metrics of the terminal device may also be collected during the UL and / or DL RLF data collection. The one or more kinematic metrics of the terminal device may comprise at least one of a location, a speed, a velocity, an altitude or an acceleration.

[0102] Following the collection of the training data, the training of the ML architecture 300 may be carried out as follows. As mentioned above, in the data collection phase, training data samples marked DL RLF, UL RLF = TRUE or FALSE with associated time series measurements of cell and beams in given serving cell were collected. Hyperparameters of the ML algorithm 300 used for the training may comprise:• a first filtering time constant (e.g., 100 ms) for RSRP samples for cell level measurements,• a second filtering time constant (e.g., 40 ms) for RSRP samples for beam level measurements,• a number of cells (e.g., 16 or 32) and / or a number of beams,• a number of time steps determined by the amount of time the prediction is required to cover (e.g., for predicting 500 ms in advance, 16 time steps, i.e., 16 iterations, are required for the given SSB periodicity of 40 ms), and / or• an RLF timer and constants such as T310 = 500 msec and N310 and N311 selected as 3 or 5 each.It should be noted that the ML model 300 may need to be trained separately for each different set of hyperparameters.

[0103] As mentioned above, a GRU based NN architecture 300 is employed as the input data is a time series with a length N. The input size of the GRU 304 is number of beams and cell quality samples grouped in a tensor having a size of M X C, where M is the number of beams per cell (e.g., 32 or 64 beams) while C is the batch size, i.e., the number of cells. Optionally, position and / or velocity (and / or some other kinematic metric) embedding may also be employed.

[0104] The hidden layer 301 captures the relationship between the different iterations in the time series. The hidden layer 301 may, e.g., have a size of 20 (to capture the features of the time series).

[0105] The output of the hidden layer 301 is fed to the first convolution layer 302. The first and second convolution layers 302, 303 (which are connected to each in series) enable detection of features within the time series for a given serving cell by exploring the correlations between the beams for a given cell. The output of the second convolution layer 303 are fed to the first linear layer 305 (having output dimensions matching the input dimensions of the GRU 304) to allow for recurrent processing of the time series to explore short and long term correlations.

[0106] The number of iterations (“ITER”) 306, to 308 is equal to the number of time steps (e.g., / V= I 6). As this is a supervised learning example, a loss function (equally called a cost function) compares a binary cross entropy between the logits of the second linear layer 308 (i.e., output of the second linear array 309) to the training samples (i.e., the collected realized TRUE / FALSE values) for DL and UL RLF separately resulting in a trained model that acts as a classifier. If the DL and UL RLF need to be jointly processed (i.e., the ML algorithm should support UL-DL mode), the loss function may correspond to a sum of the loss functions for the DL and the UL.

[0107] Once the DL and UL RLF ML models have been trained, they may be used for inference (i.e., for RLF predictions) by inputting values of the beam-specific received signal strength (e.g., the beam-specific RSRP) and cell quality (e.g., the cell-specific RSRP) and optionally the one or more kinematic metrics such as the location and velocity of the terminal device. The logits need to be converted to probabilities (which allow also determining the confidence of the measurement) by using a softmax function. DL or UL or both RLF can be considered to be reported when the threshold of the likelihood exceeds a pre-defined threshold being, e.g., equal to 70%.

[0108] As mentioned above, one of RLF prediction modes supported by the ML algorithm may be the UL-DL mode where RLF prediction is carried out for both UL and DL directions. The UL-DL mode may employ both UL and DL ML models of the ML algorithm (or a single joint UL-DL ML model of the ML algorithm). In embodiments where the UL-DL mode is used for the RLF prediction, the interactions between the UL and DL ML models may be further investigated for enabling more robust decision-making. In fact, enabling both ULand DL options provide indications on the RLF predicted based, respectively, on UL and DL based information. Therefore, in these embodiments, additional aggregation step(s) may be introduced following the individual UL & DL based prediction steps. These aggregation step or steps may allow examination of results of both UL & DL RLF prediction functions and making a decision (e.g., a reporting decision), based on said results, following one or more predefined (deterministic) rules.

[0109] FIGs. 4A & 4B illustrate two alternative processes corresponding to alternative implementations of the aforementioned aggregation function according to embodiments. Both of the processes may be carried out by a terminal device or a part thereof. The terminal device may be any of the terminal devices 100, 102 of FIG. 1. The processes of FIGs. 4A & 4B may correspond to two more detailed implementations of a part of the process of FIG. 2 (namely, steps 203, 204) and, thus, any of the features and definitions provided in connection with FIG. 2 may apply, mutatis mutandis, also here.

[0110] Referring to FIG. 4A, the terminal device performs, in block 401 , RLF prediction using the UL-DL mode of the ML algorithm. It is assumed here that the RLF prediction is performed using an UL ML model and a DL ML model, as described in connection with FIG. 2. Then, the terminal device determines, in block 402, whether a confidence level associated with RLF prediction of each of the UL ML model and the DL ML model exceeds pre-defined confidence threshold (e.g., 60 %). In response to the confidence level for the RLF prediction of at least one of the UL ML model or the DL ML model falling below the pre-defined confidence threshold in block 403, the terminal device adjusts, in block 404, values of one or more hyperparameters of said at least one of the UL ML model or the DL ML model for increasing the confidence level (i.e., different parametrization is employed). Then, the terminal device repeats, in block 405, the RLF prediction for the at least one of the UL ML model or the DL ML model (i.e., for the ML model which did not exceed the pre-defined confidence threshold) using the adjusted values of the one or more hyperparameters. The one or more hyperparameters which are adjusted in block 404 may comprise, for example, at least one hyperparameter whose value indicates whether beam and / or cell measurements and / or measurement-based beam and / or cell predictions are to be used for obtaining values of the one or more DL radio link quality metrics and / or at least one hyperparameter whose value indicates a proportion of beam and / or cell measurements relative to measurement-based beam and / or cell predictions (or vice versa) to be used for obtaining values of the one or more DL radio link quality metrics. Thus, the adjustment of block 404 may entail using a higher proportion ofmeasured beams instead of measurement-based predicted beams (i.e., beams derived from measurements using, e.g., a suitable ML algorithm). Following the new RLF prediction, the process proceeds back to the evaluation of block 402 (which is now carried out for the new RLF prediction). This process is repeated until the pre-defined confidence threshold is reached (or exceeded) for both of the UL and DL ML models. Once the pre-defined confidence threshold is reached (or exceeded) for both of the UL and DL ML models in block 403, the terminal device transmits, in block 406, the prediction report comprising results of the RLF prediction for both of the UL and DL ML models, similar to as discussed in connection with message 204 of FIG. 2.

[0111] In some embodiments, the delayed joint prediction reporting of FIG. 4A is applied only if the UL RLF prediction (i.e., the RLF prediction of the UL ML model) is the first to satisfy the pre-defined confidence threshold in block 403 (or if the pre-defined confidence threshold is satisfied at the same time by both UL and DL RLF predictions). In other embodiments, the delayed joint prediction reporting of FIG. 4 A is applied only if the DL RLF prediction (i.e., the RLF prediction of the DL ML model) is the first to satisfy the predefined confidence threshold in block 403 (or if the pre-defined confidence threshold is satisfied at the same time by both UL and DL RLF predictions).

[0112] In summary, in the embodiment of FIG. 4A, the transmission of the prediction report is delayed until satisfactory (i.e., trustworthy) RLF prediction can be acquired for both UL and DL. As this may cause significant (unnecessary) delay for communication of the RLF prediction results for one of the UL or DL ML model for which satisfactory RLF prediction results were acquired first, this may not be optimal in all scenarios. FIG. 4B provides an alternative process for tackling this particular issue.

[0113] Referring to FIG. 4B, the terminal device performs, in block 411 , RLF prediction using the UL-DL mode of the ML algorithm. It is assumed here that the RLF prediction is performed using an UL ML model and a DL ML model, as described in connection with FIG. 2. Then, the terminal device determines, in block 412, whether a confidence level associated with RLF prediction of each of the UL ML model and the DL ML model exceeds a pre-defined confidence threshold (e.g., 60 %). In response to the confidence level for the RLF prediction of both of the UL ML model and the DL ML model reaching (i.e., exceeding or being equal to) the pre-defined confidence threshold in block 413, the terminal device transmits, in block414, to the access node, a prediction report comprising results of the RLF prediction for the UL and DL ML models.

[0114] In response to the confidence level for the RLF prediction of one or both of the UL and DL ML models falling below the pre-defined confidence threshold in block 413, the terminal device further checks, in block 415, whether the pre-defined confidence threshold is reached for at least one of the UL ML model or the DL ML model. In response to neither of the RLF predictions using the UL ML model and the DL model reaching the pre-defined confidence threshold, the terminal device adjusts, in block 422, values of one or more hyperparameters of both of the UL ML model or the DL ML model for increasing the confidence level (i.e., different parametrization is employed). This adjustment may be carried out similar to as discussed above in connection with block 404 of FIG. 4. Then, the terminal device repeats actions pertaining to blocks 411 to 413 (and possibly 415).

[0115] In response to the confidence level for the RLF prediction of exactly one of the UL ML model or the DL ML model reaching the pre-defined confidence threshold in block 415, the terminal device transmits, in block 416, to the access node, a first partial prediction report comprising results of said one of the UL ML model or the DL ML model (for which satisfactory RLF prediction is available). In other words, instead of waiting to acquire satisfactory RLF predictions for both UL and DL directions, the terminal device, first, transmits the first partial prediction report covering only one of the UL and DL directions. This enables the access node to perform any recovery actions in a timely manner for said one of the UL and DL directions. For example, the terminal device may trigger a serving cell change if the RLF in DL is detected based on the first partial prediction report. Alternatively, if the UL ML model confirms the RLF occurrence in the UL direction, the access node may choose to trigger SUL switch and / or wait to get a dependable DL prediction with a high confidence level and decide accordingly if need to trigger serving cell change.

[0116] Thereafter, the terminal device adjusts, in block 417, values of one or more hyperparameters of other of the UL ML model or the DL ML model (for which trustworthy RLF prediction is currently unavailable) so as to increase the confidence level of that RLF prediction and repeats, in block 418, the RLF prediction in the UL-DL mode for said other of the UL and DL ML models using the adjusted values of the one or more hyperparameters. The terminal device determines, in block 419, whether confidence level associated with the (latest)RLF prediction of said other of the UL ML model or the DL ML model exceeds the pre-defined confidence threshold.

[0117] Thereafter, actions pertaining to blocks 417 to 420 are repeated until the predefined confidence threshold is reached, in block 420, for the RLF prediction of said other of the UL and DL ML models. Once this happens, the terminal device transmits, in block 421, to the access node, a second partial prediction report comprising results of said other of the UL ML model or the DL ML model (the one for which trustworthy RLF prediction was obtained later).

[0118] In some embodiments, the first partial prediction report may comprise also latest results of said other of the UL ML model or the DL ML model (i.e., the one for which high- confidence RLF prediction is currently unavailable). In such a case, the first partial report may comprise the confidence level at least for said other of the UL ML model or the DL ML model or an indication that the confidence level for said other of the UL ML model or the DL ML model is not satisfactory.

[0119] In some embodiments relating to either of processes of FIGs. 4A or 4B, the predefined confidence threshold may be defined separately for the UL and DL ML models (i.e., separately for UL and DL RLF predictions).

[0120] In the procedure of FIG. 2, it was assumed that the access node is, at the onset of the procedure, aware of the capabilities of the terminal device (at least in terms of supported RLF prediction modes) and knows how the terminal device should be configured though it was not described therein how this information is shared between the terminal device and the access node. FIG. 5 illustrates signaling between a terminal device and an access node (or equally a network node or entity) for enabling the sharing of capabilities of the terminal device with the access node and carrying out the configuration of the terminal device accordingly according to some embodiments. The terminal device may be any of the terminal devices 100, 102 of FIG. 1. The access node may be a non-distributed access node such as the access node 104 of FIG. 1 or a distributed access node or a particular unit thereof (e.g., the DU 104 or the CU 108 of FIG. 1. In general, any of the features and definitions discussed in connection with FIG. 2 (and / or any of FIGs. 3, 4A & 4B) may apply, mutatis mutandis, also in the case of FIG. 5.

[0121] Referring to FIG. 5, it may be assumed, similar to as discussed in connection with FIG. 2, that the terminal device is maintaining, in at least one memory comprised in orconnected to the terminal device, an ML algorithm for predicting future RLFs in UL and / or DL directions based on at least on values of one or more radio link quality metrics (and optionally one or more kinematic metrics). Any of the features and definition provided in connection with FIG. 2 may apply, mutatis mutandis, also here.

[0122] Initially, the access node may transmit, in message 501, to the terminal device, a request for RLF prediction capabilities of the terminal device. In some embodiments, the request may explicitly specify which particular RLF prediction capabilities are requested. Thus, any of the specific types of RLF prediction capabilities to be discussed in connection with message 503 may be (explicitly) requested in the request. For example, the request may request RLF prediction capabilities relating to the UL mode, the DL mode or the UL-DL mode of the ML algorithm for RLF prediction.

[0123] In some embodiments, the request (i.e., message 501) may request RLF prediction capabilities of the terminal device specific to one or more frequency bands and / or one or more channel bandwidths (as specified in the request).

[0124] The terminal device receives, in block 502, the request for RLF prediction capabilities of the terminal device. Based on the request, the terminal device transmits, in message 503, to the access node, the RLF prediction capabilities of the terminal device. The RLF prediction capabilities may comprise capabilities relating directly to the RLF prediction and optionally also to the reporting of the results of the RLF prediction.

[0125] The transmitted RLF prediction capabilities may comprise at least one or more supported RLF prediction modes of the ML algorithm of the terminal device. The one or more supported RLF prediction modes may comprise at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions. In some embodiments, the one or more supported RLF prediction modes may comprise at least two or all of: the UL mode, the DL mode or the UL-DL mode.

[0126] In some embodiments, the RLF prediction capabilities transmitted in message 503 may comprise an indication whether the ML algorithm is applicable cell- specifically or geographic-area-specifically or network-specifically. In the first case, the ML algorithm may be applied for specific cell(s) whereas, in the latter two cases, the ML algorithm may be applied to the whole network or geographic area covered by a set of cells.

[0127] In some embodiments, the RLF prediction capabilities transmitted in message 503 may comprise one or more pre-defined future time windows used by the ML algorithm for the RLF prediction. The one or more pre-defined future time windows may comprise one or more first pre-defined future time windows for UL RLF prediction and / or one or more second pre-defined future time windows for DL RLF prediction. Alternatively, the one or more predefined future time windows may comprise one or more pre-defined future time windows usable for UL, DL and / or joint UL-DL RLF prediction. Specifically, the RLF prediction capabilities transmitted in message 503 may comprise a duration and / or the number of time steps of any of the listed future time windows used by the ML algorithm for the RLF prediction.

[0128] In some embodiments, the RLF prediction capabilities transmitted in message 503 may comprise an indication for supporting or not supporting periodic and / or aperiodic reporting of results of RLF prediction (i.e., periodic and / or aperiodic transmission of the prediction report). In some embodiments, said indication may be provided separately for UL and DL RLF prediction.

[0129] In some embodiments, the RLF prediction capabilities transmitted in message 503 may comprise an indication for supporting or not supporting performance monitoring of realized RLFs. The performance monitoring of the realized RLFs may comprise keeping track of whether a predicted RLF actually occurs, that is, whether the predicted RLF is subsequently matched with a realized RLF. It should be noted that this functionality may require the terminal device to keep the RLF detection functionalities active in spite of possible recovery actions configured by the network to avoid the RLF.

[0130] In some embodiments where the one or more supported RLF prediction modes comprise at least the UL-DL mode, the RLF prediction capabilities transmitted in message 503 may comprise one or more supported UL-DL mode reporting properties comprising at least one of:• an indication that joint UL-DL prediction reporting is triggered after a reporting trigger condition is satisfied for both of the UL and DL RLF predictions (e.g., the process of FIG. 4 A is employed),• an indication that joint UL-DL prediction reporting is triggered after the reporting trigger condition is satisfied for both of the UL and DL RLF predictions (e.g., the process of FIG. 4A is employed) at least in cases where the UL RLF prediction is first to satisfy the reporting trigger condition or the UL and DL RLF predictions satisfy thereporting trigger condition at the same time (i.e., terminal device waits for satisfactory DL RLF prediction if a satisfactory UL RLF prediction is acquired first),• an indication that joint UL-DL prediction reporting is triggered after the reporting trigger condition is satisfied for both of the UL and DL RLF predictions (e.g., the process of FIG. 4A is employed) at least in cases where the DL RLF prediction is first to satisfy the reporting trigger condition or the UL and DL RLF predictions satisfy the reporting trigger condition at the same time (i.e., terminal device waits for satisfactory UL RLF prediction if a satisfactory DL RLF prediction is acquired first),• an indication that prediction reporting is triggered separately for the UL and DL RLF predictions when the reporting trigger condition is satisfied for the UL and DL RLF predictions, respectively, (e.g., the process of FIG. 4B is employed), or• a supported time-to-trigger for the joint UL-DL prediction reporting (e.g., 100 ms or 200 ms).Here, the reporting trigger condition may comprise, e.g., the pre-defined confidence threshold, as discussed in connection with FIGs. 4A & 4B.

[0131] In some embodiments, the RLF prediction capabilities transmitted in message 503 may comprise one or more features (i.e., inputs) of the ML algorithm comprising at least the one or more radio link quality metrics, and / or one or more sets of one or more hyperparameters supported by the ML algorithm. The one or more radio link quality metrics indicated in message 503 may comprise one or more UL radio link quality metrics and / or one or more DL radio link quality metrics. The one or more UL radio link quality metrics and / or the one or more DL radio link quality metrics may be defined here as described in connection with FIG. 2. Each of the one or more sets of one or more hyperparameters may comprise, e.g., values of at least one of: the first filtering time constant, the second filtering time constant, the number of cells (i.e., C), the number of beams per cell (i.e., M), the number of time steps associated with at least one time window, an RLF timer type indicator, or an RLF timer limit. As mentioned previously, the one or more features of the ML algorithm may further comprise one or more kinematic metrics (e.g., a location and / or speed or velocity of the terminal device).

[0132] In some embodiments, the RLF prediction capabilities transmitted in message 503 may comprise at least the (aforementioned) one or more sets of one or more hyperparameters, the one or more supported RLF prediction modes may comprise at least one of the DL mode or the UL-DL mode, and the one or more radio link quality metrics maycomprise at least the (aforementioned) one or more DL radio link quality metrics. In such embodiments, at least one of the one or more sets of one or more hyperparameters may comprise at least one of:• at least one hyperparameter whose value indicates whether beam and / or cell measurements and / or measurement-based beam and / or cell predictions are used for obtaining values of the one or more DL radio link quality metrics (at the terminal device),• at least one hyperparameter whose value indicates a proportion of beam and / or cell measurements relative to measurement-based beam and / or cell predictions (or vice versa) to be used for obtaining values of the one or more DL radio link quality metrics,• at least one hyperparameter whose value indicates whether results of RLF predictions based on beam and / or cell measurements and / or measurement-based beam and / or cell predictions are included in the prediction report, or• a type of a DL reference signal (e.g., SSB or DMRS) used for DL RLF prediction.The definition and use of the predicted beam level or cell level measurements are discussed in further detail below in connection with FIG. 7.

[0133] In some embodiments, elements 501, 502 may be omitted. In such cases, the transmitting of the RLF prediction capabilities (i.e., message 503) may be triggered autonomously or independently by the terminal device (e.g., upon a handover or according to pre-defined periodic).

[0134] The access node receives, in block 504, the RLF prediction capabilities from the terminal device. Then, the access node may determine, in block 505, one or more configurations for RLF prediction of the terminal device based at least on the RLF prediction capabilities of the terminal device. Here, the one or more configurations (or at least one of them) comprise at least one of the one or more supported RLF prediction modes. In some embodiments, each or at least one of the one or more configurations may pertain, in addition to the RLF prediction itself, also to reporting of the results of the RLF prediction (i.e., transmission of the prediction report). In general, each (or at least one) of the one or more configurations may comprise (or indicate) a subset of the RLF prediction capabilities of the terminal device which were reported in message 503 and / or comprise configurationinformation not (explicitly) reported in message 503 (e.g., a supplementary uplink configuration).

[0135] To give an example, the access node may determine a particular supported set of one or more hyperparameters of the ML algorithm to enable configuration of the terminal device to perform RLF predictions for up to C cells (e.g., 16 cell) using up to M beams (e.g., 32 beams) per cell, providing reference signal configuration for some cells for all the M beams and in some cells asking the terminal device to first predict the reference signal from a subset of the beams (e.g., M / 2). The access node may select the RLF timers and constants (i.e., RLF timer limit values) from a set supported by the terminal device (e.g., T310 = 500 ms, N310 and / or N311 = 5). The configuration(s) pertaining to the prediction reporting may request the terminal device to prioritize UL RLF report in physical uplink control channel (PUCCH) (using uplink control information, UCI, for example). The configuration(s) may, additionally or alternatively, define that DL RLF prediction may be only reported using PUSCH with a semi- persistent or configured grant.

[0136] The access node may transmit, in message 506, to the terminal device, the one or more configurations. The terminal device may receive, in block 507, the one or more configurations. The message 506 may be, e.g., an RRC reconfiguration message. It should be noted that message 201 of FIG. 2 may be considered one special case of message 506.

[0137] In some embodiments, each or at least one of the one or more configurations transmitted in message 506 may comprise: an UL (prediction) trigger condition for initiating the RLF prediction in the UL direction and / or a DL (prediction) trigger condition for initiating the RLF prediction in the DL direction. The UL trigger condition may be based on at least one of: a number of attempted RLC retransmissions, a number of random access attempts or a power headroom. The DL trigger condition may be based on at least one of: a DL serving cell power, a DL target cell power or a number of consecutive out-of-sync indications associated with a DL reference signal. In general, the UL trigger condition (or conditions) and / or DL trigger condition (or conditions) may be defined as described above in connection with FIG. 2.

[0138] In some embodiments, each or at least one of the one or more configurations transmitted in message 506 may comprise a reporting trigger condition for triggering the transmitting of the prediction report. The reporting trigger condition may be based on at least one of: a predicted probability for an RLF (in either UL or DL) or an accuracy or confidence level of an RLF prediction. Alternatively, the reporting trigger condition may be based on atleast one of: a predicted probability for an UL RLF, a predicted probability for a DL RLF, an accuracy or confidence level of an UL RLF prediction or an accuracy or confidence level of a DL RLF prediction. The reporting trigger condition may comprise a separate threshold for any of the quantities listed in this paragraph. In general, the reporting trigger condition (or conditions) may be defined as described above in connection with FIG. 2.

[0139] In some embodiments, each or at least one of the one or more configurations transmitted in message 506 may comprise (or indicate) a set of one or more hyperparameters of the ML algorithm, where the set of one or more hyperparameters may be, e.g., one of the one or more sets of one or more hyperparameters included in the RLF prediction capabilities of message 503. Alternatively, each or at least one of the one or more configurations transmitted in message 506 may comprise at least identifying information (e.g., an identifier) on one of said one or more sets of one or more hyperparameters. Thus, each or at least one of the one or more configurations transmitted in message 506 may, e.g., comprise or indicate at least one of:• a value of at least one hyperparameter of the ML algorithm indicating whether beam or cell measurements and / or measurement-based beam or cell predictions are to be used for obtaining values of the one or more DL radio link quality metrics,• a value of at least one hyperparameter whose value indicates a proportion of beam and / or cell measurements relative to measurement-based beam and / or cell predictions (or vice versa) to be used for obtaining values of the one or more DL radio link quality metrics,• an indication to include, in a prediction report, an indication on whether reported results of the RLF prediction are based on the measured beam level or cell level measurements and / or predicted beam level or cell level measurements, or• a type of a DL reference signal (e.g., SSB or DMRS) to be used for DL RLF prediction.The definition and use of the predicted beam level or cell level measurements are discussed in further detail below in connection with FIG. 7.

[0140] In some embodiments, each or at least one of the one or more configurations transmitted in message 506 may comprise: an indication to include, in the prediction report, information on out-of-sync indications (or at least consecutive out-of-sync indications)detected by the terminal device and / or information on in-sync indications detected by the terminal device.

[0141] In some embodiments where the RLF prediction capabilities transmitted in message 503 comprised the one or more supported UL-DL mode reporting properties, each or at least one of the one or more configurations transmitted in message 506 may comprise or indicate at least one of the one or more supported UL-DL mode reporting properties.

[0142] Subsequently, the terminal device may perform RLF prediction using said one or more configurations or at least one of them in block 508, similar to block 203 of FIG. 2. Moreover, the terminal device may transmit, in message 509, a prediction report comprising at least some of the results of the RLF prediction, similar to as discussed in connection with message 204 of FIG. 2. The prediction reporting may, at least in some embodiments, also be influenced by the one or more configurations or at least one of them. The access node may receive, in block 510, the prediction report. Any of the subsequent actions described in connection with elements 206 to 208 of FIG. 2 may also be carried out here.

[0143] FIG. 6 illustrates supplementary signaling between a terminal device and an access node (or equally a network node or entity) for enabling selection of the most appropriate configuration for the terminal device according to some embodiments. The terminal device may be any of the terminal devices 100, 102 of FIG. 1. The access node may be a nondistributed access node such as the access node 104 of FIG. 1 or a distributed access node or a particular unit thereof (e.g., the DU 104 or the CU 108 of FIG. 1. In general, any of the features and definitions discussed in connection with any of FIGs. 2, 3, 4A, 4B and 5 may apply, mutatis mutandis, also in the case of FIG. 6.

[0144] Referring to FIG. 6, initial steps of the procedure of FIG. 6 in elements 601, 602 correspond fully to elements 506, 507 of FIG. 7. Thus, the access node transmits, in message 601, to the terminal device, one or more configurations for RLF prediction (and optionally prediction reporting), where the one or more configurations may have been derived via execution of the procedure of elements 502 to 505 or 501 to 505 of FIG. 5. The one or more configurations may comprise at least one of the one or more RLF prediction modes supported by the terminal device, as described in connection with previous embodiments. Each (or at least one) of the one or more configurations may comprise a subset of the RLF prediction capabilities of the terminal device which were reported to the access node earlier and / or configuration information not (explicitly) reported to the access node. The terminal device may receive, inblock 602, the one or more configurations. The message 601 may be an RRC reconfiguration message. In some embodiments, the one or more configurations may comprise a plurality of configurations.

[0145] Thereafter, instead of directly selecting one of the one or more configurations to use for the RLF prediction, a supplementary procedure for making a more informed decision regarding which configuration to use is carried out in elements 603 to 608.

[0146] First, the terminal device determines, in block 603, based on the one or more configurations, one or more configurations acceptable for use by the terminal device for the RLF prediction (and optionally for the prediction reporting). The determining may be further based on one or more properties of the terminal device and / or the ML algorithm. The one or more acceptable configurations are a subset of the one or more configurations transmitted in message 601. Here (and also in other parts of the application), the term “subset” includes also the alternative that the two sets are equal (i.e., the subset is not necessarily a proper or strict subset). To provide a concrete example of the determination of block 603, the terminal device may, for example, determine that some of the one or more configurations relate to performing DL RLF prediction using measurement-based predicted beams while the terminal device is only able perform the DL RLF prediction only using measured beams. Such configuration(s) may, thus, be omitted from the one or more acceptable configurations.

[0147] The terminal device transmits, in message 604, to the access node, information on the one or more acceptable configurations. Said information may comprise the configuration(s) themselves or simply an indication (e.g., a flag or a set of flags) indicating which of the previously transmitted one or more configurations are acceptable. The message 604 may be an RRC reconfiguration response.

[0148] The access node receives, in block 605, the information on the one or more configurations acceptable for use by the terminal device. The access node determines (or selects), in block 606, one of the one or more acceptable configurations for use by the terminal device and transmits, in message 607, to the terminal device, a request for activating said one of the one or more acceptable configurations. Said request may comprise the acceptable configuration to be activated or at least identifying information (e.g., an identifier) on the acceptable configuration to be activated.

[0149] Thereafter, the terminal device performs, in block 609, RLF prediction using the activated acceptable configuration. The terminal device may also transmit, in message 610, to the access node a prediction report comprising at least some of results of the RLF prediction. The transmission of the prediction report may, at least in some embodiments, be carried out also according to the activated acceptable configuration (namely, in cases where the activated acceptable configuration specifies how the prediction reporting is to be carried out). The access node may receive, in block 611, the prediction report. In general, the steps 609 to 611 may be carried out as described in connection with previous embodiments.

[0150] In the example scenario of FIG. 6, it is assumed that at least one of the one or more configurations transmitted in message 601 is acceptable for use by the terminal device. If this is not the case (i.e., none of the one or more configurations can be used), the terminal device may transmit, to the access node, a message indicating this fact (not shown in FIG. 6), instead of message 604. Obviously, the following steps of the procedure (i.e., steps 606 to 611) are not carried out in such a case.

[0151] As was mentioned above, the RLF prediction carried out using the ML algorithm may be based on beam or cell measurements and / or measurement-based beam or cell predictions. The one or more configurations transmitted from the access node to the terminal device (e.g., in message 201 of FIG. 2, in message 506 of FIG. 5 or in message 601 of FIG. 6) may specify whether beam or cell measurements and / or measurement-based beam or cell predictions are to be used for the RLF prediction. Support for beam or cell measurements and / or measurement-based beam or cell predictions may also be indicated in the RLF prediction capabilities transmitted from the terminal device to the access node (e.g., in message 503 of FIG. 5).

[0152] Said (direct) beam or cell measurements may be based, e.g., on legacy beam management procedures such as any of Pl & P2 coarse & narrow beam selection and update procedures and / or P3 beam refinement procedure as defined in TS 38.213 and TS 38.214 specifications.

[0153] Use of the measurement-based beam or cell predictions, instead of relying solely on direct measurements, provides the benefit of not needing to sweep through all beams leading to saving of time as well as resources and latency reduction. FIG. 7 illustrates one exemplary ML-based scheme 700 for performing said measurement-based beam or cell predictions in embodiments.

[0154] FIG. 7 shows a ML model 703 having at least one of two alternative inputs 701, 702 and an output 704. The inputs 701, 702 and the output 704 are represented in FIG. 7 as beam grids (with the active beam(s) being colored in black or with a pattern). In the illustrated example, the beam grid is an 8 x 8 grid of (narrow) beams. The ML model 703 may be implemented, for example, using supervised learning. For example, the ML model 703 may be based on at least one of: one or more RNNs, an artificial neural network based solution (e.g., multilayer perceptron), one or more SVMs, linear regression, logistic regression, one or more decision trees and a k-NN method.

[0155] The first alternative input 701 corresponds to (layer-1) RSRP measurements of wide beams while the second alternative input 702 corresponds to (layer- 1) RSRP measurements of a subset of narrow beams (that is, a subset of all beams of the beam grid). Additionally, the ML model 703 may take as input assistance information (comprising, e.g., beam shape information and / or beam identifiers) (not shown in FIG. 7).

[0156] The output 704 of the ML model 703 may comprise an identifier (e.g., a beam index) and / or RSRP of the best narrow beam and optionally an associated quality of service (QoS) value for beam selection optimizing specific target metric(s) such as throughput or packet delay.

[0157] In the case where the first alternative input 701 is used, the ML model 703 is trained to a predict narrow beam based on one or more measured coarse (i.e., wide) beams. In practice, this may correspond to prediction of a channel state information reference signal (CSI- RS) based on one or more SSBs. In FIG. 7, the coarse beams (e.g., SSB beams) of the input 701 are shown by square areas having different patterning. This alternative provides the benefit that the access node does not need to transmit any CSI-RSs.

[0158] In the case where the second alternative input 702 is used, the ML model 703 is trained to a predict a narrow beam based on a subset of narrow beams of the beam grid. In practice, this may correspond to prediction of a CSI-RS based on one or more (other) CSI-RSs. In FIG. 7, the narrow beams (e.g., CSI-RS beams) of the input 702 are shown as black square areas of the beam grid of the input element 702. This alternative provides the benefit that the access node can transmit a set of CSI-RSs with a reduced density.

[0159] FIGs. 8 & 9 illustrate signaling between a terminal device and an access node (or equally a network node or entity) for configuring the terminal device for performing RLFprediction, reporting the RLF prediction and carrying out one of three different specific procedures for avoiding or at least reducing the likelihood of the RLF according to some embodiments. The terminal device may be any of the terminal devices 100, 102 of FIG. 1. The access node may be a non-distributed access node such as the access node 104 of FIG. 1 or a distributed access node or a particular unit thereof (e.g., the DU 104 or the CU 108 of FIG. 1. In general, any of the features and definitions discussed in connection with any of FIGs. 2, 3, 4A, 4B, 5, 6 or 7 may apply, mutatis mutandis, also in the case of FIGs. 8 & 9.

[0160] Referring to FIG. 8, the actions relating to elements 801 to 806 may correspond corresponding actions described in connection with previous embodiments (e.g., in connection with elements 201 to 205 of FIG. 2 and / or elements 506 to 510 of FIG. 5). Thus, said actions are summarized in the following only briefly.

[0161] Initially, the access node may transmit, in message 801, to the terminal device, one or more configurations for RLF prediction using an ML algorithm. The terminal device may receive, in block 802, the one or more configurations. The ML algorithm is configured to predict future RLFs in UL and / or DL directions based at least on values of one or more radio link quality metrics (and optionally one or more kinematic metrics). The one or more configurations may comprise at least an RLF prediction mode of the ML algorithm to be used for the RLF prediction, where the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction, or an UL-DL mode for RLF prediction in UL and DL directions. In some embodiments, elements 801, 802 may be omitted (e.g., the terminal device may be considered pre-configured).

[0162] The terminal device performs, in block 803, RLF prediction using the RLF prediction mode of the ML algorithm.

[0163] The terminal device may determine, in block 804, that a reporting trigger condition is satisfied based on the results of the RLF prediction. If the RLF prediction mode used for the RLF prediction is the DL mode, the reporting trigger condition may be based on at least one of: a predicted probability for a DL RLF or an accuracy or confidence level of a DL RLF prediction. If the RLF prediction mode used for the RLF prediction is the UL mode, the reporting trigger condition may be based on at least one of: a predicted probability for an UL RLF or an accuracy or confidence level of an UL RLF prediction. If the RLF prediction mode used for the RLF prediction is the UL-DL mode, the reporting trigger condition may be based on at least one of: a predicted probability for an UL RLF, a predicted probability for aDL RLF, an accuracy or confidence level of an UL RLF prediction or an accuracy or confidence level of a DL RLF prediction.

[0164] In some embodiments, the determining of block 804 may be omitted (i.e., the process may proceed from block 803 directly to the transmission of message 805), similar to FIG. 2.

[0165] The terminal device transmits, in message 805, to the access node, a prediction report comprising at least some of results of the RLF prediction. If the determination of block 804 was carried out, the transmitting of the prediction report may be performed based on said determination. The contents of the prediction report may be defined as described in connection with previous embodiments. In some embodiments, the RLF prediction mode used for the RLF prediction is the DL mode or the UL-DL mode, and the prediction report comprises at least some of results of the RLF prediction at least for the DL direction. The access node receives, in block 806, the prediction report.

[0166] The access node determines, in block 807, based at least on the prediction report, that an RLF will (likely) occur. The determination of block 807 may be further based on one or more pre-defined criteria for positive RLF detection. The one or more pre-defined criteria for the positive RLF detection may be based at least on a predicted probability for an UL RLF or a combination of a predicted probability for an UL RLF and an accuracy or confidence level of an UL RLF prediction and / or a predicted probability for a DL RLF or a combination of a predicted probability for a DL RLF and an accuracy or confidence level of a DL RLF prediction (which may be provided in the prediction report). For example, the one or more pre-defined criteria may comprise a pre-defined threshold for at least one of the predicted probability for the UL RLF, the accuracy or confidence level of the UL RLF prediction, the predicted probability for the DL RLF or the accuracy or confidence level of an UL RLF prediction. In other words, said one or more pre-defined criteria may define, e.g., how high the probability of detection should be for positive UL and / or DL RLF detection and / or how high the confidence level or accuracy of the RLF prediction should be for positive UL and / or DL RLF detection (i.e., in order for the positive UL and / or DL RLF detection to be considered trustworthy).

[0167] Based on the determining in block 807, the access node transmits, in message 808, to the terminal device, a request for triggering a primary cell (PCell) handover for theterminal device. Upon receiving the request in block 809, the terminal device triggers, in block 810, the PCell handover according to the request.

[0168] In some embodiments, the request for triggering the PCell handover may be transmitted at least if the results of the RLF prediction indicate that a DL RLF (or both DL and UL RLFs) will (likely) occur. In other embodiments, the request for triggering the PCell handover may be transmitted also if the results of the RLF prediction indicate that an UL RLF will (likely) occur. It should be noted that, due to reciprocity of the radio channel, the UL RLF may be considered indicative of an RLF also in the DL direction.

[0169] Referring to FIG. 9, the actions relating to elements 901 to 903 may correspond corresponding actions described in connection with previous embodiments (e.g., in connection with elements 201 to 203 of FIG. 2 and / or elements 506 to 508 of FIG. 5). Thus, said actions are summarized in the following only briefly.

[0170] Initially, the access node may transmit, in message 901, to the terminal device, one or more configurations for RLF prediction using an ML algorithm. The terminal device may receive, in block 902, the one or more configurations. The ML algorithm is configured to predict future RLFs in UL and / or DL directions based at least on values of one or more radio link quality metrics (and optionally one or more kinematic metrics). The one or more configurations may comprise at least an RLF prediction mode of the ML algorithm to be used for the RLF prediction. Here, the RLF prediction mode is one of: an UL mode for RLF prediction in an UL direction, or an UL-DL mode for RLF prediction in UL and DL directions. In some embodiments, elements 901, 902 may be omitted (e.g., the terminal device may be considered pre-configured at least with said RLF prediction mode).

[0171] In some embodiments, the one or more configurations for the RLF prediction may comprise at least one of a supplementary uplink (SUL) configuration or an indication for enabling independent switching from the UL configuration to the SUL configuration (i.e., an indication defining which of alternatives 1 & 2 shown in FIG. 9 should be employed). According to a general definition, the SUL configuration is a supplementary UL configuration for enabling enhanced UL performance (compared to the primary UL configuration of the terminal device), particularly in challenging scenarios where UL coverage or capacity is limited. All or at least some of one or more frequency bands of the SUL configuration may cover lower frequencies compared to one or more frequency bands employed by the primary UL configuration for enabling said improved UL performance.

[0172] The terminal device performs, in block 903, RLF prediction using the RLF prediction mode (i.e., the UL or UL-DL mode) of the ML algorithm. It is assumed here that the terminal device is, at this point in time, configured to use a primary UL configuration of the terminal device.

[0173] The terminal device determines, in block 904, based on the results of the (UL) RLF prediction, that an RLF will (likely) occur for the UL direction. The determination of block 904 may be further based on one or more pre-defined criteria for positive UL RLF detection. The one or more pre-defined criteria for the positive UL RLF detection may be based at least on a predicted probability for an UL RLF or a combination of a predicted probability for an UL RLF and an accuracy or confidence level of an UL RLF prediction. For example, the one or more pre-defined criteria may comprise a pre-defined threshold for at least one of the predicted probability for the UL RLF or the accuracy or confidence level of the UL RLF prediction. In other words, said one or more pre-defined criteria may define, e.g., how high the probability of detection should be for positive UL RLF detection and / or how high the confidence level or accuracy of the RLF prediction should be for positive UL RLF detection (i.e., in order for the positive UL RLF detection to be considered trustworthy).

[0174] Based on the determining in block 904, the terminal device may initiate one of two alternative procedures (indicated as “Alt. 1” and “Alt. 2” in FIG. 9). As mentioned above, the terminal device may, in some embodiments, determine which of the two alternative procedures to carry out based on the one or more configurations received in block 902. Alternatively, the terminal device may be able to carry out only one of the two alternative procedures.

[0175] According to the first alternative, based on the determining in block 904, the terminal device switches, in block 905, (directly) from the primary UL configuration to a supplementary uplink (SUL) configuration to minimize a chance of the RLF occurring. As mentioned above, the SUL configuration may, at least in some embodiments, have been configured via message 901. Optionally, the terminal device may still transmit, in message 906, to the access node, a prediction report comprising at least some of results of the RLF prediction. Here, the transmission of the prediction report may be based on the determining in block 904 and / or based on determining that a reporting trigger condition is satisfied based on the results of the RLF prediction (being defined similar to as described in previous embodiments). The prediction report may be transmitted using the primary UL configuration and / or using the SULconfiguration. The contents of the prediction report may be defined as described in connection with previous embodiments The access node may receive, in block 907, at least the prediction report.

[0176] In some embodiments of the first alternative procedure, the terminal device may transmit, to the access node, an indication indicating that the terminal device will switch to use the SUL configuration before the switching of block 905. Additionally or alternatively, the terminal device may transmit, to the access node (e.g., in message 906), an indication indicating that the terminal device has switched to use the SUL configuration after the switching of block 905.

[0177] In some embodiments of the first alternative procedure, the terminal device may, before the switching to the SUL configuration in block 905, perform RLF prediction using the RLF prediction mode of the ML algorithm and the SUL configuration. Then, the terminal device may determine, based on the results of the RLF prediction performed using the SUL configuration (and optionally the one or more pre-defined criteria for the positive UL RLF detection), whether the RLF will (likely) occur for the UL direction when using the SUL configuration. In response to determining that the RLF will (likely) not occur when using the SUL configuration, the terminal device may perform the SUL switch of block 905. In response to determining that the RLF will likely occur also when using the SUL configuration, the terminal device may trigger an UL failure procedure (e.g., an RRC re-establishment procedure).

[0178] According to the second alternative, the terminal device does not directly switch to use the SUL configuration but first asks for permission from the network to do so. Namely, based on the determining in block 904, the terminal device transmits, in message 908, to the access node, a request requesting a switch from using the primary UL configuration to using the SUL configuration to minimize a chance of the RLF occurring. The terminal device may optionally also transmit, in message 909, to the access node, a prediction report comprising at least some of results of the RLF prediction. Here, the transmission of the prediction report may be based on the determining in block 904 and / or based on determining that a reporting trigger condition is satisfied based on the results of the RLF prediction (being defined similar to as described in previous embodiments). While FIG. 9 depicts transmission of message 908 occurring before transmission of message 909, in other embodiments, message 909 may be transmitted before message 908.

[0179] The access node receives, in block 910, the request and optionally the prediction report. The access node determines, in block 911, based on the request and optionally the prediction report, that the RLF will (likely) occur for the UL direction. In other words, the access node approves the requested SUL switch.

[0180] The determination of block 911 may be further based on one or more pre-defined criteria for positive RLF detection. The one or more pre-defined criteria for the positive UL RLF detection may be based at least on a predicted probability for an UL RLF or a combination of a predicted probability for an UL RLF and an accuracy or confidence level of an UL RLF prediction. For example, the one or more pre-defined criteria may comprise a pre-defined threshold for at least one of the predicted probability for the UL RLF or the accuracy or confidence level of the UL RLF prediction. In other words, said one or more pre-defined criteria may define, e.g., how high the probability of detection should be for positive UL RLF detection and / or how high the confidence level or accuracy of the RLF prediction should be for positive UL RLF detection (i.e., in order for the positive UL RLF detection to be considered trustworthy).

[0181] Based on the determination in block 911, the access node transmits, in message 912, to the terminal device, a confirmation for the switch to the SUL configuration. The terminal device receives, in block 913, from the access node, the confirmation and, based on said confirmation, switches, in block 914, from the primary UL configuration to the SUL configuration.

[0182] The blocks, related functions, and information exchanges described above by means of FIGs. 2, 3, 4A, 4B & 5 to 9 are in no absolute chronological order, and some of them may be performed simultaneously or in an order differing from the given one. Other functions can also be executed between them or within them, and other information may be sent, and / or other rules applied. Some of the blocks or part of the blocks or one or more pieces of information can also be left out or replaced by a corresponding block or part of the block or one or more pieces of information.

[0183] FIG. 10 provides an apparatus 1001 according to some embodiments. Specifically, FIG. 10 may illustrate an apparatus configured to carry out at least some of the functions described above. The apparatus 1001 may be or form a part of a terminal device. Alternative, the apparatus 1001 may be or form a part of a (non-distributed) access node or of at least one of an RU, a DU or a CU of a distributed access node.

[0184] The apparatus 1001 may comprise one or more communication control circuitry 1020, such as at least one processor, and at least one memory 1030, including one or more algorithms 1031, such as a computer program code (software) wherein the at least one memory and the computer program code (software) are configured, with the at least one processor, to cause the apparatus 1001 to carry out any one of the exemplified functionalities of the terminal device or the access node described above in connection with any of FIGs. 2, 3, 4A, 4B & 5 to 9. Said at least one memory 1030 may also comprise at least one database 1032.

[0185] When the one or more communication control circuitry 1020 comprises more than one processor, the apparatus 1001 may be a distributed device wherein processing of tasks takes place in more than one physical unit. Each of the at least one processor may comprise one or more processor cores. A processing core may comprise, for example, a Cortex-A12 processing core manufactured by ARM Holdings or a Zen processing core designed by Advanced Micro Devices Corporation. The one or more control circuitry 1020 may comprise at least one Qualcomm Snapdragon and / or Intel Atom processor.

[0186] Referring to FIG. 10, the one or more communication control circuitry 1020 of the apparatus 1001 is configured to carry out functionalities described above by means of any of elements of FIGs. 2, 3, 4A, 4B & 5 to 9 using one or more individual circuitries. It may also be feasible to use specific integrated circuits, such as DSP block, digital signal processor, application-specific integrated circuit (ASIC) or field-programmable gate array (FPGA), or other components and devices for implementing said functionalities in accordance with different embodiments.

[0187] Referring to FIG. 10, the apparatus 1001 may further comprise different interfaces 1010 such as one or more communication interfaces comprising hardware and / or software for realizing communication connectivity according to one or more communication protocols. For example, the one or more communication interfaces 1010 may comprise at least one interface enabling communication between the apparatus 1001 and one or more access nodes (if the apparatus is a terminal device or a part thereof) or between the apparatus and one or more terminal devices (if the apparatus is an access node or a part thereof). Additionally, if the apparatus is an access node, the one or more communication interfaces 1010 may comprise at least one interface enabling communication between the apparatus 1001 and at least one core network node and / or at least one interface enabling communication between the apparatus 1001 and one or more (other) access nodes. If the apparatus 1001 forms a part of a distributed accessnode, the one or more communication interfaces 1010 may comprise one or more interfaces providing one or more connections between the apparatus 1001 and other parts of the distributed access node.

[0188] Referring to FIG. 10, the memory 1030 may be implemented using any suitable data storage technology, such as semiconductor based memory devices, flash memory, magnetic memory devices and systems, optical memory devices and systems, fixed memory and removable memory.

[0189] 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 (and / or firmware), 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 terminal device or an access node, to perform various functions, and (c) hardware circuit(s) and 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. This definition of ‘circuitry’ applies to all uses of this term in this application, including 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 a portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware.

[0190] In an embodiment, at least some of the processes described in connection with FIGs. 2, 3, 4A, 4B & 5 to 9 may be carried out by an apparatus comprising corresponding means for carrying out at least some of the described processes. Some example means for carrying out the processes may include at least one of the following: detector, processor (including dual-core and multiple-core processors), digital signal processor, controller, receiver, transmitter, encoder, decoder, memory, register, multiply-accumulate (MAC) unit, delay element, RAM, ROM, software, firmware, display, user interface, display circuitry, user interface circuitry, user interface software, display software, circuit, filter (low-pass, high-pass, bandpass and / or bandstop), sensor, circuitry, inverter, capacitor, inductor, resistor, operational amplifier, diode and transistor. In some embodiments, at least some of the processes may be implemented using discrete components. In an embodiment, at least some of the processesdescribed in connection with FIGs. 2, 3, 4A, 4B & 5 to 9 may be carried out by an apparatus comprising corresponding hardware means for carrying out at least some of the described processes.

[0191] According to an embodiment, there is provided an apparatus (e.g., a terminal device) comprising means for performing: receiving, from an access node, a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm for predicting future RLFs in uplink, UL, and downlink, DL, directions based at least on values of one or more radio link quality metrics, wherein the RLF prediction mode is one of: an UL-DL mode for RLF prediction in the UL and DL directions, an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction; performing RLF prediction using the received RLF prediction mode of the ML algorithm; and transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction.

[0192] According to an embodiment, there is provided an apparatus (e.g., a network entity or node or an access node) comprising means for performing: transmitting, to a terminal device, a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm for predicting future RLFs in uplink, UL, and downlink, DL, directions based on values of one or more radio link quality metrics, wherein the RLF prediction mode is one of: an UL-DL mode for RLF prediction in the UL and DL directions, an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction; and receiving, from the terminal device, a prediction report comprising at least some of results of an RLF prediction carried out by the terminal device using the transmitted RLF prediction mode of the ML algorithm

[0193] According to an embodiment, there is provided an apparatus (e.g., a terminal device) comprising means for performing: transmitting, to an access node, radio link failure, RLF, prediction capabilities of the apparatus, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at leaston values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

[0194] According to an embodiment, there is provided an apparatus (e.g., a network entity or node or an access node) comprising means for performing: receiving, from a terminal device, radio link failure, RLF, prediction capabilities of the terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

[0195] According to an embodiment, there is provided an apparatus (e.g., a terminal device) comprising means for performing: performing RLF prediction using a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction, or an UL-DL mode for RLF prediction in UL and DL directions; determining that a reporting trigger condition is satisfied based on the results of the RLF prediction; based on the determining, transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction; and in response to receiving, from the access node, a request for triggering a primary cell handover for the apparatus, triggering the primary cell handover according to the request.

[0196] According to an embodiment, there is provided an apparatus (e.g., a terminal device) comprising means for performing: performing RLF prediction using a radio link failure, RLF, prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict futureRLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is an UL mode for RLF prediction in an UL direction, a DL mode for RLF prediction in a DL direction or an UL-DL mode for RLF prediction in UL and DL directions, the apparatus being configured to use a primary UL configuration of the apparatus; and determining, based on the results of the RLF prediction, that an RLF will likely occur for the UL direction; and based on the determining, either- switching from the primary UL configuration to a supplementary uplink, SUL, configuration to minimize a chance of the RLF occurring, or- transmitting, to the access node, a request requesting a switch from using the primary UL configuration to using a SUL configuration to minimize a chance of the RLF occurring, and, in response to receiving, from the access node, a confirmation for the switch to the SUL configuration, switching from the primary UL configuration to the SUL configuration.

[0197] According to an embodiment, there is provided an apparatus (e.g., a network entity or node or an access node) comprising means for performing: receiving, from a terminal device, a prediction report comprising at least some of results of RLF prediction performed, at the terminal device, using an RLF prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is one of: a DL mode for RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction or an UL-DL mode for RLF prediction in UL and DL directions; determining, based on the prediction report, that an RLF will likely occur; and based on the determining, transmitting, to the terminal device, a request for triggering a primary cell handover for the terminal device.

[0198] According to an embodiment, there is provided an apparatus (e.g., a network entity or node or an access node) comprising means for performing: receiving, from a terminal device, a prediction report comprising at least some of results of radio link failure, RLF, prediction performed, at the terminal device, using an RLF prediction mode of a machine-learning, ML, algorithm, wherein the ML algorithm is configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the RLF prediction mode is a DL modefor RLF prediction in a DL direction, an UL mode for RLF prediction in an UL direction or an UL-DL mode for RLF prediction in UL and DL directions; receiving, from the terminal device, a request requesting a switch from using a primary UL configuration of the terminal device to using a supplementary uplink, SUL, configuration to minimize a chance of an RLF occurring; determining, based on the prediction report and the request, that the RLF will likely occur for the UL direction; and transmitting, to the terminal device, a confirmation for the switch to the SUL configuration.

[0199] Embodiments as described above may also be carried out, fully or at least in part, in the form of a computer process defined by a computer program or portions thereof. Embodiments of the methods described in connection with FIGs. 2, 3, 4A, 4B & 5 to 9 may be carried out by executing at least one portion of a computer program comprising corresponding instructions. The computer program may be provided as a computer readable medium comprising program instructions stored thereon or as a non-transitory computer readable medium comprising program instructions stored thereon. The computer program may be in source code form, object code form, or in some intermediate form, and it may be stored in some sort of carrier, which may be any entity or device capable of carrying the program. For example, the computer program may be stored on a computer program distribution medium readable by a computer or a processor. The computer program medium may be, for example but not limited to, a record medium, computer memory, read-only memory, electrical carrier signal, telecommunications signal, and software distribution package, for example. The computer program medium may be a non-transitory medium. Coding of software for carrying out the embodiments as shown and described is well within the scope of a person of ordinary skill in the art.

[0200] The term “non-transitory”, as used herein, is a limitation of the medium itself (that is, tangible, not a signal) as opposed to a limitation on data storage persistency (for example, RAM vs. ROM).

[0201] Reference throughout this specification to one embodiment or an embodiment means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present solution. Thus, appearances of the phrases “in one embodiment” or “in an embodiment” in various places throughout thisspecification are not necessarily all referring to the same embodiment.

[0202] As used herein, a plurality of items, structural elements, compositional elements, and / or materials may be presented in a common list for convenience. However, these lists should be construed as though each member of the list is individually identified as a separate and unique member. Thus, no individual member of such list should be construed as a de facto equivalent of any other member of the same list solely based on their presentation in a common group without indications to the contrary. In addition, various embodiments and example of the present solution may be referred to herein along with alternatives for the various components thereof. It is understood that such embodiments, examples, and alternatives are not to be construed as de facto equivalents of one another, but are to be considered as separate and autonomous representations of the present solution.

[0203] Even though embodiments have been described above with reference to examples according to the accompanying drawings, it is clear that the embodiments are not restricted thereto but can be modified in several ways within the scope of the appended claims. Therefore, all words and expressions should be interpreted broadly and they are intended to illustrate, not to restrict, the embodiment. It will be obvious to a person skilled in the art that, as technology advances, the inventive concept can be implemented in various ways. Further, it is clear to a person skilled in the art that the described embodiments may, but are not required to, be combined with other embodiments in various ways.INDUSTRIAL APPLICABILITY

[0204] At least some embodiments find industrial application in wireless communications.

Claims

CLAIMS:

1. An apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: transmitting, to an access node, radio link failure, RLF, prediction capabilities of the apparatus, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

2. The apparatus of claim 1, wherein the one or more supported RLF prediction modes comprise at least two of: the UL-DL mode, the UL mode or the DL mode.

3. The apparatus of claim 1 or 2, wherein the at least one memory further storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform, before the transmitting of the RLF prediction capabilities: receiving, from the access node, a request for the RLF prediction capabilities of the apparatus, wherein the transmitting of the RLF prediction capabilities is performed based on the request.

4. The apparatus according to any preceding claim, wherein the RLF prediction capabilities further comprise at least one of: an indication whether the ML algorithm is applicable cell-specifically or geographic-area- specifically or network-specifically, one or more pre-defined future time windows used by the ML algorithm for the RLF prediction, an indication for supporting or not supporting periodic and / or aperiodic reporting of results of RLF prediction, oran indication for supporting or not supporting performance monitoring of realized RLFs.

5. The apparatus according to any preceding claim, wherein the RLF prediction capabilities further comprise at least one of: one or more features of the ML algorithm comprising at least the one or more radio link quality metrics, or one or more sets of one or more hyperparameters supported by the ML algorithm.

6. The apparatus of claim 5, wherein the RLF prediction capabilities comprise at least the one or more features of the ML algorithm comprising at least the one or more radio link quality metrics, the one or more radio link quality metrics comprising: one or more UL radio link quality metrics comprising at least one of: a number of attempted radio link control, RLC, retransmissions, a number of random access attempts, an RLC acknowledgment latency, power headroom, a hybrid automatic repeat request, HARQ, statistics metric or an UL power control parameter, and / or one or more DL radio link quality metrics comprising at least one of: a DL block error rate, a DL power control parameter, a DL beam measurement metric, a DL cell quality measurement metric, a reference signal received power, RSRP, per cell or beam, a reference signal received quality, RSRQ, per cell or beam or a number of consecutive out-of-sync indications associated with a DL reference signal.

7. The apparatus of claim 5 or 6, wherein the transmitted RLF prediction capabilities comprise at least the one or more sets of one or more hyperparameters, the one or more supported RLF prediction modes comprise at least one of the DL mode or the UL-DL mode, and the one or more radio link quality metrics comprise at least one or more DL radio link quality metrics, at least one of the one or more sets of one or more hyperparameters comprising at least one of: at least one hyperparameter whose value indicates whether beam or cell measurements and / or measurement-based beam or cell predictions are used for obtaining values of the one or more DL radio link quality metrics,at least one hyperparameter whose value indicates a proportion of beam and / or cell measurements relative to measurement-based beam and / or cell predictions to be used for obtaining values of the one or more DL radio link quality metrics, at least one hyperparameter whose value indicates whether results of RLF predictions based on beam and / or cell measurements and / or measurement-based beam and / or cell predictions are included in a prediction report transmitted by the apparatus, or a type of a DL reference signal used for DL RLF prediction.

8. The apparatus according to any of claims 5 to 7, wherein the one or more features of the ML algorithm, comprised in the transmitted RLF prediction capabilities, further comprise at least one of a location of the apparatus or a speed of the apparatus.

9. The apparatus according to any preceding claim, wherein the one or more supported RLF prediction modes comprise the UL-DL mode, and the transmitted RLF prediction capabilities comprise one or more supported UL-DL mode reporting properties comprising at least one of: an indication that joint UL-DL prediction reporting is triggered after a reporting trigger condition is satisfied for both of the UL and DL RLF predictions, an indication that joint UL-DL prediction reporting is triggered after the reporting trigger condition is satisfied for both of the UL and DL RLF predictions at least in cases where the UL RLF prediction is first to satisfy the reporting trigger condition or the UL and DL RLF predictions satisfy the reporting trigger condition at the same time, an indication that joint UL-DL prediction reporting is triggered after the reporting trigger condition is satisfied for both of the UL and DL RLF predictions at least in cases where the DL RLF prediction is first to satisfy the reporting trigger condition or the UL and DL RLF predictions satisfy the reporting trigger condition at the same time, an indication that prediction reporting is triggered separately for the UL and DL RLF predictions when the reporting trigger condition is satisfied for the UL and DL RLF predictions, respectively, or a supported time-to-trigger.

10. The apparatus according to any preceding claim, wherein the at least one memory further storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: receiving, from the access node, one or more configurations for RLF prediction determined based at least on the transmitted RLF prediction capabilities of the apparatus, wherein the one or more configurations comprise at least one of the one or more supported RLF prediction modes; performing the RLF prediction using the ML algorithm according to one of the received one or more configurations; and transmitting, to the access node, a prediction report comprising at least some of results of the RLF prediction.

11. The apparatus according to any preceding claim, wherein said at least some of the results of the RLF prediction included in the prediction report comprise one or more probabilities for an UL RLF within respective one or more first pre-defined future time windows and / or one or more probabilities for a DL RLF within respective one or more second pre-defined future time windows, and / or said at least some of the results of the RLF prediction included in the prediction report comprise one or more indications that an UL RLF or no UL RLF will likely occur within respective one or more first pre-defined future time windows and / or one or more indications that an DL RLF or no DL RLF will likely occur within respective one or more second predefined future time windows, and / or said at least some of the results of the RLF prediction included in the prediction report comprise one or more probabilities that an UL radio link quality falls below a first predefined radio link quality threshold within respective one or more first pre-defined future time window and / or one or more probabilities that a DL radio link quality falls below a second pre-defined radio link quality threshold within respective one or more second predefined future time window.

12. The apparatus of claim 10 or 11, wherein each or at least one of the one or more configurations comprises at least a subset of the RLF prediction capabilities of the apparatus to be used for the RLF prediction.

13. The apparatus according to any of claims 10 to 12, wherein each or at least one of the one or more configurations further comprises: an UL trigger condition for initiating the RLF prediction in the UL direction, wherein the UL trigger condition is based on at least one of: a number of attempted RLC retransmissions, a number of random access attempts or a power headroom, and / or a DL trigger condition for initiating the RLF prediction in the DL direction, wherein the DL trigger condition is based on at least one of: a DL serving cell power, a DL target cell power or a number of consecutive out-of-sync indications associated with a DL reference signal.

14. The apparatus according to any of claims 10 to 13, wherein each or at least one of the one or more configurations further comprises: a reporting trigger condition for triggering the transmitting of the prediction report, the reporting trigger condition being based on at least one of: a predicted probability for an UL RLF, a predicted probability for a DL RLF, an accuracy or confidence level of an UL RLF prediction or an accuracy or confidence level of a DL RLF prediction.

15. The apparatus according to any of claims 10 to 14, wherein the one or more received configurations comprise a plurality of received configurations, and the at least one memory further storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: determining, based on the plurality of configurations and properties of the apparatus and / or the ML algorithm, one or more configurations acceptable for use by the apparatus for the RLF prediction and the prediction reporting; transmitting, to the access node, information on the one or more acceptable configurations; receiving, from the access node, a request for activating one of the one or more acceptable configurations; and performing the RLF prediction and the transmitting of the prediction report according to the activated acceptable configuration.

16. An apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform:receiving, from a terminal device, radio link failure, RLF, prediction capabilities of the terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

17. The apparatus of claim 16, wherein the at least one memory further storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: determining one or more configurations for RLF prediction of the terminal device based at least on the RLF prediction capabilities of the terminal device, wherein the one or more configurations comprise at least one of the one or more supported RLF prediction modes; transmitting, to the terminal device, the one or more configurations; and receiving, from the terminal device, a prediction report comprising results of an RLF prediction carried out by the terminal device using the ML algorithm and one of the one or more configurations.

18. The apparatus of claim 17, wherein the at least one memory further storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: determining, based on the results of the RLF prediction, that an RLF will likely occur; and based at least on the determining, causing reconfiguration of the terminal device for minimizing a chance of the RLF occurring.

19. The apparatus according to any of claims 16 to 18, wherein the at least one memory further storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform, before the receiving of the RLF prediction capabilities: transmitting, to the terminal device, a request for the RLF prediction capabilities of the terminal device.

20. The apparatus according to any of claim 16 to 19, wherein the determined one or more configurations comprise a plurality of configurations, and the at least one memory furtherstoring instructions that, when executed by the at least one processor, cause the apparatus at least to perform: receiving, from the terminal device, information on one or more configurations acceptable for use by the terminal device for the RLF prediction, wherein the one or more configurations are a subset of the plurality of configurations; and transmitting, to the terminal device, a request for activating one of the one or more acceptable configurations.

21. A method, comprising: transmitting, to an access node, radio link failure, RLF, prediction capabilities wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

22. A computer program product comprising at least one non-transitory computer-readable storage medium having computer-executable program code instructions stored therein, the computer-executable program code instructions comprising program code instructions configured to: transmit, to an access node, radio link failure, RLF, prediction capabilities wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

23. An apparatus comprising:means for transmitting, to an access node, radio link failure, RLF, prediction capabilities of the apparatus, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction, a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

24. A method, comprising: receiving, from a terminal device, radio link failure, RLF, prediction capabilities of the terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

25. A computer program product comprising at least one non-transitory computer-readable storage medium having computer-executable program code instructions stored therein, the computer-executable program code instructions comprising program code instructions configured to: receive, from a terminal device, radio link failure, RLF, prediction capabilities of the terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

26. An apparatus comprising:means for receiving, from a terminal device, radio link failure, RLF, prediction capabilities of the terminal device, wherein the RLF prediction capabilities comprise at least one or more supported RLF prediction modes of a machine-learning, ML, algorithm, the ML algorithm being configured to predict future RLFs in uplink, UL, and / or downlink, DL, directions based at least on values of one or more radio link quality metrics, and the one or more supported RLF prediction modes comprising at least one of: an UL mode for RLF prediction in the UL direction or a DL mode for RLF prediction in the DL direction or an UL-DL mode for RLF prediction in the UL and DL directions.

Images