Methods, apparatuses and systems for machine learning-based radio link failure predictions

Abstract

Methods, apparatuses and systems for machine learning-based radio link failure predictions. In one embodiment, a method includes: receiving, at a wireless communication device, a first plurality of beams from a first plurality of wireless communication nodes, wherein the first plurality of wireless communication nodes includes a first wireless communication node and a second plurality of wireless communication nodes; transmitting, at the wireless communication device, a report to the first wireless communication node, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report includes: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

Description

F9125-62400 METHODS, APPARATUSES AND SYSTEMS FOR MACHINE LEARNING-BASED RADIO LINK FAILURE PREDICTIONS TECHNICAL FIELD

[0001] The disclosure relates generally to wireless communications and, more particularly, to methods, apparatuses and systems for machine learning-based radio link failure predictions. BACKGROUND

[0002] The development of 5G New Radio (NR) technology has introduced unprecedented capabilities in terms of data rates, latency, and connectivity. However, the reliance on high-frequency millimeter-wave (mmWave) bands, while providing significant bandwidth, poses substantial challenges in maintaining stable connections. These frequencies are particularly vulnerable to free space path loss, blockages, and environmental interference, which can result in rapid degradation of signal quality. As a result, both the base station (BS) and user equipment (UE) must constantly adapt their beam configurations to prevent radio link failures (RLFs) and ensure continuous connectivity, particularly in dense urban environments or high-speed mobility scenarios.

[0003] Traditionally, RLF and handover prediction methods rely on basic threshold-based measurements of signal strength and quality. These methods trigger handovers or RLF recovery only after signal quality has already deteriorated, often resulting in delayed responses that lead to service interruptions and degraded user experience. Furthermore, these conventional methods involve a high degree of signaling overhead, as the network must frequently request updates from the UE on multiple beams to make accurate predictions about potential connection losses. In a dynamic 5G network, this approach can lead to 1 DM2\20713381.2F9125-62400 increased latency and inefficient use of network resources, especially as the number of UEs grows and the network becomes increasingly dense. Therefore, there is a need to improve the conventional RLF and handover prediction methods to meet the demands of complex, high- speed, and high-capacity 5G networks. SUMMARY

[0004] The exemplary embodiments disclosed herein are directed to solving the issues relating to one or more of the problems presented in the prior art, as well as providing additional features that will become readily apparent by reference to the following detailed description when taken in conjunction with the accompany drawings. In accordance with various embodiments, exemplary systems, methods, devices and computer program products are disclosed herein. It is understood, however, that these embodiments are presented by way of example and not limitation, and it will be apparent to those of ordinary skill in the art who read the present disclosure that various modifications to the disclosed embodiments can be made while remaining within the scope of the present disclosure.

[0005] In some embodiments, a method includes: receiving, at a wireless communication device, a first plurality of beams from a first plurality of wireless communication nodes, wherein the first plurality of wireless communication nodes includes a first wireless communication node and a second plurality of wireless communication nodes; transmitting, at the wireless communication device, a report to the first wireless communication node, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report includes: an indication indicating that a radio link failure (RLF) occurs within a time period.

[0006] In some embodiments, the first plurality of beams includes: a plurality of channel state information reference signal (CSI-RS) beams from the first wireless communication 2 DM2\20713381.2F9125-62400 node; and a plurality of synchronization signal block (SSB) beams from the second plurality of wireless communication nodes.

[0007] In some embodiments, the method further includes: prior to receiving the first plurality of beams, receiving, at the wireless communication device, a pre-radio resource control (RRC) configuration message from the first wireless communication node, wherein the pre-RRC configuration message configures the first plurality of beams to be measured by the wireless communication device for RLF predictions or handover predictions. In some other embodiments, the pre-RRC configuration message may be an RRC Reconfiguration message, wherein the “pre” indicates the timing of when the RRC Reconfiguration message is sent.

[0008] In some embodiments, the report is generated based on the first plurality of measurements and a second plurality of measurements taken for a second plurality of beams, wherein the second plurality of beams is generated based on the first plurality of beams using a moving window technique. In some embodiments, the second plurality of beams is generated based on at least one of: a location of the wireless communication device; a speed of the wireless communication device; and a network configuration.

[0009] In some embodiments, the first plurality of measurements includes at least one of: a Reference Signal Received Power (RSRP); a Reference Signal Received Quality (RSRQ); a Signal-to-Interference-plus-Noise Ratio (SINR); a Channel Quality Indicator (CQI); a Beam Reference Signal Strength (BRSS); a Timing Advance (TA) Measurement; and an Angle of Arrival (AoA) or an Angle of Departure (AoD).

[0010] In some embodiments, the report is generated using a machine learning (ML) model, wherein the ML model is a deep neural network (DNN) model implemented in the wireless communication device. 3 DM2\20713381.2F9125-62400 BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Various exemplary embodiments of the present disclosure are described in detail below with reference to the following Figures. The drawings are provided for purposes of illustration only and merely depict exemplary embodiments of the present disclosure to facilitate the reader's understanding of the present disclosure. Therefore, the drawings should not be considered limiting of the breadth, scope, or applicability of the present disclosure. It should be noted that for clarity and ease of illustration these drawings are not necessarily drawn to scale.

[0012] FIG. 1A illustrates an exemplary wireless communication network, in accordance with some embodiments of the present disclosure.

[0013] FIG. 1B illustrates a block diagram of an exemplary wireless communication system, in accordance with some embodiments of the present disclosure.

[0014] FIG. 2A illustrates an exemplary design framework for machine learning-based radio link failure predictions, in accordance with some embodiments.

[0015] FIG. 2B illustrates an exemplary design framework for deriving cell quality from beam measurements and generating reports at different layers, in accordance with some embodiments.

[0016] FIG. 3A illustrates a high-level training stage procedure of a machine learning model used for radio link failure predictions, in accordance with some embodiments.

[0017] FIG. 3B illustrates a high-level inference stage procedure using a trained machine learning model for radio link failure predictions, in accordance with some embodiments.

[0018] FIG. 4 illustrates another exemplary design framework for handover predictions, in accordance with some embodiments of the present disclosure. 4 DM2\20713381.2F9125-62400

[0019] FIG. 5 illustrates a signaling diagram for handover failure HOF recovery based on pre-Radio Resource Control (RRC) configuration, in accordance with some embodiments of the present disclosure.

[0020] FIG. 6 illustrates a deep neural network (DNN) model used to implement a machine learning model, in accordance with some embodiments.

[0021] FIG. 7 illustrates an example method for machine learning-based radio link failure and handover predictions, in accordance with some embodiments. DETAILED DESCRIPTION OF EXEMPLARY EMBODIMENTS

[0022] Various exemplary embodiments of the present disclosure are described below with reference to the accompanying figures to enable a person of ordinary skill in the art to make and use the present disclosure. As would be apparent to those of ordinary skill in the art, after reading the present disclosure, various changes or modifications to the examples described herein can be made without departing from the scope of the present disclosure. Thus, the present disclosure is not limited to the exemplary embodiments and applications described and illustrated herein. Additionally, the specific order and / or hierarchy of steps in the methods disclosed herein are merely exemplary approaches. Based upon design preferences, the specific order or hierarchy of steps of the disclosed methods or processes can be re-arranged while remaining within the scope of the present disclosure. Thus, those of ordinary skill in the art will understand that the methods and techniques disclosed herein present various steps or acts in a sample order, and the present disclosure is not limited to the specific order or hierarchy presented unless expressly stated otherwise. 5 DM2\20713381.2F9125-62400

[0023] Figure 1A illustrates an exemplary wireless communication network 100, in accordance with some embodiments of the present disclosure. In a wireless communication system, a network side communication node or a base station (BS) 102 can be a node B, an E-UTRA Node B (also known as Evolved Node B, eNodeB or eNB), a New Generation eNB (ng-eNB), a gNodeB (also known as gNB) in new radio (NR) technology, a pico station, a femto station, or the like. A terminal side communication device or a user equipment (UE) 104 can be a long range communication system like a mobile phone, a smart phone, a personal digital assistant (PDA), tablet, laptop computer, or a short range communication system such as, for example a wearable device, a vehicle with a vehicular communication system and the like. A network communication node and a terminal side communication device are represented by a BS 102 and a UE 104, respectively, and in all the embodiments in this disclosure hereafter, and are generally referred to as “communication nodes” and “communication device” herein. Such communication nodes and communication devices may be capable of wireless and / or wired communications, in accordance with various embodiments of the invention. It is noted that all the embodiments are merely preferred examples, and are not intended to limit the present disclosure. Accordingly, it is understood that the system may include any desired combination of communication nodes and communication devices, while remaining within the scope of the present disclosure.

[0024] Referring to Figure 1A, the wireless communication network 100 includes a first BS 102-1, a second BS 102-2, a first UE 104-1, a second UE 104-2, a third UE 104-3, and a fourth UE 104-4. In some embodiments, the first BS 102-1 and the second BS 102-2 comprise a first plurality of antennas 106-1 to 106-n and a second plurality of antennas 116-1 to 116-n’, respectively. The first plurality of antennas 106-1 to 106-n may communicate with a plurality of UEs 104 to form a first multiple-input multiple-output (MIMO) system, and the 6 DM2\20713381.2F9125-62400 second plurality of antennas 116-1 to 116-n’ may communicate with the plurality of UEs 104 to form a second MIMO system.

[0025] In some embodiments, a plurality of UEs 104 may form direct communication (e.g., uplink) channels 103-1, 103-2, 103-3, and 103-4 with the first BS 102-1 and the second BS 102-2. In some embodiments, the plurality of UEs 104 may also form direct communication (e.g., downlink) channels 105-1, 105-2, 105-3, and 105-4 with the first BS 102-1 and the second BS 102-2. The direct communication channels between the plurality of UEs 104 and a distributed unit of the BS 102 can be through interfaces such as an Uu interface, which is also known as E-UTRAN air interface. In some embodiments, the UE 104 comprises a plurality of transceivers which enables the UE 104 to support multi connectivity so as to receive data simultaneously from the first BS 102-1 and the second BS 102-2. The first BS 102-1 and the second BS 102-2 each is connected to a core network (CN) 108 on a user plane (UP) through an external interface 107, e.g., an Iu interface, an NG-U interface, or an S1-U interface. In some embodiments, the CN 108 is one of the following: an Evolved Packet Core (EPC) and a 5G Core Network (5GC). In some embodiments, the CN 108 further comprises at least one of the following: Access and Mobility Management Function (AMF), User Plane Function (UPF), and System Management Function (SMF).

[0026] A direct communication channel 111 between the first BS 102-1 and the second 102-2 is through an X2 interface, in accordance with some embodiments. In some embodiments, a BS (gNB) is split into a Distributed Unit (DU) and a Central Unit (CU) on the UP, between which the direct communication is through a F1-U interface. In some embodiments, a CU of the second BS 102-2 can be further split into a Control Plane (CP) and a User Plane (UP), between which the direct communication is through an E1 interface. Hereinafter in the present disclosure, an Xx interface is used to describe one of the following 7 DM2\20713381.2F9125-62400 interfaces, the NG interface, the S1 interface, the X2 interface, the Xn interface, the F1 interface, and the E1 interface. When an Xx interface is established between two nodes, the two nodes can transmit control signaling on the CP and / or data on the UP. In some other embodiments, the BSs 102-1 and 102-2 are Transmit-Receive-Points (TRPs) connected to the same controller unit or DU.

[0027] Figure 1B illustrates a block diagram of an exemplary wireless communication system 150, in accordance with some embodiments of the present disclosure. The system 150 may include components and elements configured to support known or conventional operating features that need not be described in detail herein. In some embodiments, the system 150 can be used to transmit and receive data symbols in a wireless communication environment such as the wireless communication network 100 of Figure 1A, as described above.

[0028] The system 150 generally includes a first BS 102-1, a second BS 102-2, and a UE 104, collectively referred to as BS 102 and UE 104 below for ease of discussion. The first BS 102-1 and the second BS 102-2 each comprises a BS transceiver module 152, a BS antenna array 154, a BS memory module 156, a BS processor module 158, and a network interface 160. In the illustrated embodiment, each module of the BS 102 is coupled and interconnected with one another as necessary via a data communication bus 180. The UE 104 comprises a UE transceiver module 162, a UE antenna 164, a UE memory module 166, a UE processor module 168, and an I / O interface 169. In the illustrated embodiment, each module of the UE 104 is coupled and interconnected with one another as necessary via a date communication bus 190. The BS 102 communicates with the UE 104 via a communication channel 192, which can be any wireless channel or other medium known in the art suitable for transmission of data as described herein. 8 DM2\20713381.2F9125-62400

[0029] As would be understood by persons of ordinary skill in the art, the system 150 may further include any number of modules other than the modules shown in Figure 1B. Those skilled in the art will understand that the various illustrative blocks, modules, circuits, and processing logic described in connection with the embodiments disclosed herein may be implemented in hardware, computer-readable software, firmware, or any practical combination thereof. To clearly illustrate this interchangeability and compatibility of hardware, firmware, and software, various illustrative components, blocks, modules, circuits, and steps are described generally in terms of their functionality. Whether such functionality is implemented as hardware, firmware, or software depends upon the particular application and design constraints imposed on the overall system. Those familiar with the concepts described herein may implement such functionality in a suitable manner for each particular application, but such implementation decisions should not be interpreted as limiting the scope of the present invention.

[0030] A wireless transmission from a transmitting antenna of the UE 104 to a receiving antenna of the BS 102 is known as an uplink (UL) transmission, and a wireless transmission from a transmitting antenna of the BS 102 to a receiving antenna of the UE 104 is known as a downlink (DL) transmission. In accordance with some embodiments, the UE transceiver 162 may be referred to herein as an “uplink” transceiver 162 that includes a radio frequency (RF) transmitter and receiver circuitry that is each coupled to the UE antenna 164. A duplex switch (not shown) may alternatively couple the uplink transmitter or receiver to the uplink antenna in time duplex fashion. Similarly, in accordance with some embodiments, the BS transceiver 152 may be referred to herein as a “downlink” transceiver 152 that includes RF transmitter and receiver circuitry that are each coupled to the antenna array 154. A downlink duplex switch may alternatively couple the downlink transmitter or receiver to the downlink antenna array 154 in time duplex fashion. The operations of the two transceivers 152 and 162 are 9 DM2\20713381.2F9125-62400 coordinated in time such that the uplink receiver is coupled to the uplink UE antenna 164 for reception of transmissions over the wireless communication channel 192 at the same time that the downlink transmitter is coupled to the downlink antenna array 154. Preferably, there is close synchronization timing with only a minimal guard time between changes in duplex direction. The UE transceiver 162 communicates through the UE antenna 164 with the BS 102 via the wireless communication channel 192. The BS transceiver 152 communications through the BS antenna 154 of a BS (e.g., the first BS 102-1) with the other BS (e.g., the second BS 102-2) via a wireless communication channel 196. The wireless communication channel 196 can be any wireless channel or other medium known in the art suitable for direct communication between BSs.

[0031] The UE transceiver 162 and the BS transceiver 152 are configured to communicate via the wireless data communication channel 192, and cooperate with a suitably configured RF antenna arrangement 154 / 164 that can support a particular wireless communication protocol and modulation scheme. In some exemplary embodiments, the UE transceiver 162 and the BS transceiver 152 are configured to support industry standards such as the Long Term Evolution (LTE) and emerging 5G standards (e.g., NR), and the like. It is understood, however, that the invention is not necessarily limited in application to a particular standard and associated protocols. Rather, the UE transceiver 162 and the BS transceiver 152 may be configured to support alternate, or additional, wireless data communication protocols, including future standards or variations thereof.

[0032] The processor modules 158 and 168 may be implemented, or realized, with a general purpose processor, a content addressable memory, a digital signal processor, an application specific integrated circuit, a field programmable gate array, any suitable programmable logic device, discrete gate or transistor logic, discrete hardware components, 10 DM2\20713381.2F9125-62400 or any combination thereof, designed to perform the functions described herein. In this manner, a processor module may be realized as a microprocessor, a controller, a microcontroller, a state machine, or the like. A processor module may also be implemented as a combination of computing devices, e.g., a combination of a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a digital signal processor core, or any other such configuration.

[0033] Furthermore, the steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in firmware, in a software module executed by processor modules 158 and 168, respectively, or in any practical combination thereof. The memory modules 156 and 166 may be realized as RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. In this regard, the memory modules 156 and 166 may be coupled to the processor modules 158 and 168, respectively, such that the processors modules 158 and 168 can read information from, and write information to, memory modules 156 and 166, respectively. The memory modules 156 and 166 may also be integrated into their respective processor modules 158 and 168. In some embodiments, the memory modules 156 and 166 may each include a cache memory for storing temporary variables or other intermediate information during execution of instructions to be executed by processor modules 158 and 168, respectively. The memory modules 156 and 166 may also each include non-volatile memory for storing instructions to be executed by the processor modules 158 and 168, respectively.

[0034] The network interface 160 generally represents the hardware, software, firmware, processing logic, and / or other components of the base station 102 that enable bi-directional communication between BS transceiver 152 and other network components and 11 DM2\20713381.2F9125-62400 communication nodes configured to communication with the BS 102. For example, network interface 160 may be configured to support internet or WiMAX traffic. In a typical deployment, without limitation, network interface 160 provides an 802.3 Ethernet interface such that BS transceiver 152 can communicate with a conventional Ethernet based computer network. In this manner, the network interface 160 may include a physical interface for connection to the computer network (e.g., Mobile Switching Center (MSC)). The terms “configured for” or “configured to” as used herein with respect to a specified operation or function refers to a device, component, circuit, structure, machine, signal, etc. that is physically constructed, programmed, formatted and / or arranged to perform the specified operation or function. The network interface 160 could allow the BS 102 to communicate with other BSs or a CN over a wired or wireless connection.

[0035] Referring again to Figure 1A, as mentioned above, the BS 102 repeatedly broadcasts system information associated with the BS 102 to one or more UEs 104 so as to allow the UEs 104 to access the network within the cells where the BS 102 is located, and in general, to operate properly within the cell. Plural information such as, for example, downlink and uplink cell bandwidths, downlink and uplink configuration, cell information, configuration for random access, etc., can be included in the system information. Typically, the BS 102 broadcasts a first signal carrying some major system information, for example, configuration of the cell where the BS 102 is located through a Physical Broadcast Channel (PBCH). For purposes of clarity of illustration, such a broadcasted first signal is herein referred to as “first broadcast signal.” It is noted that the BS 102 may subsequently broadcast one or more signals carrying some other system information through respective channels (e.g., a Physical Downlink Shared Channel (PDSCH)). 12 DM2\20713381.2F9125-62400

[0036] Referring again to Figure 1B, in some embodiments, the major system information carried by the first broadcast signal may be transmitted by the BS 102 in a symbol format via the communication channel 192 (e.g., a PBCH). In accordance with some embodiments, an original form of the major system information may be presented as one or more sequences of digital bits and the one or more sequences of digital bits may be processed through plural steps (e.g., coding, scrambling, modulation, mapping steps, etc.), all of which can be processed by the BS processor module 158, to become the first broadcast signal. Similarly, when the UE 104 receives the first broadcast signal (in the symbol format) using the UE transceiver 162, in accordance with some embodiments, the UE processor module 168 may perform plural steps (de-mapping, demodulation, decoding steps, etc.) to estimate the major system information such as, for example, bit locations, bit numbers, etc., of the bits of the major system information. The UE processor module 168 is also coupled to the I / O interface 169, which provides the UE 104 with the ability to connect to other devices such as computers. The I / O interface 169 is the communication path between these accessories and the UE processor module 168.

[0037] Referring again to Figure 1A, during the transmission of signals between the BS 102 and the UE 104, the established wireless transmission channels between the BS 102 and the UE 104 may introduce various impairments and distortions to the transmitted signals due to factors such as fading, interference, and noise. Channel estimation can be performed to estimate the characteristics of the communication channel between the BS 102 and the UE 104 to optimize wireless communication system performance and to improve the reliability of communication. A conventional way to perform channel estimation is to use channel reciprocity for MIMO precoding in the downlink by estimating the UL channel based on the symmetrical properties between the UL and DL channels. That is, the UE 104 can periodically transmit pilot signals or Sounding Reference Signals (SRSs) during specific time 13 DM2\20713381.2F9125-62400 slots allocated for UL channel sounding, and the corresponding BS 102 can measure the received SRSs to estimate the UL channel characteristics such as channel gains and phases. In case of a time-division duplexing (TDD) transmission, there is channel reciprocity between the UL and DL channels. This means that the UL and DL channel responses are related, allowing information obtained from UL measurements to be used for DL transmission. For example, the BS 102 can perform DL MIMO precoding based on the extracted UL channel state information (CSI) from the received pilot signals or SRSs. Once the DL MIMO precoding matrix is determined, the BS 102 can use it to precode the DL data transmission, which helps in mitigating the effects of channel fading and interference and improving the quality of the received signal at the UEs.

[0038] In some embodiments, a massive MIMO system is used with a number of antennas at the BS 102 to enhance data throughput and spectrum efficiency. However, the implementation of MIMO systems necessitates accurate CSI acquisition at the BS and UE transmitters, which can be achieved through codebook-based feedback in Frequency Division Duplexing (FDD) networks or reciprocity-based sounding in Time Division Duplexing (TDD) networks. Despite these methods, the exhaustive beam search required for selecting optimal transmit-receive (Tx-Rx) beam pairs in 5G New Radio (NR) technology results in significant signaling overhead and delays, thus necessitating more efficient approaches.

[0039] In some embodiments, a handover (HO) is needed in the wireless communication network 100 when certain conditions related to signal quality, user movement, or network optimization are met. These conditions can trigger a handover decision to maintain service continuity and provide an optimal connection. Example of common scenarios when a handover is needed include: serving cell signal degradation, better signal quality from a neighboring cell, and high UE mobility. In some embodiments, HO procedures can be 14 DM2\20713381.2F9125-62400 categorized into two types: L3-based HO and L1 / L2-triggered mobility (LTM), as defined in 3GPP TS 38.300 and TS 38.331. L3-based HOs rely on higher-layer decisions based on signal strength thresholds for the serving and neighboring cells, whereas LTM-based HOs rely on lower-layer triggers for faster, more dynamic handover responses. Each of these HO types can be further optimized through Conditional Handover (CHO) if supported by both the network and the UE. However, regardless of the specific HO type, the network’s decision to initiate an HO is based on beam measurements provided by the UE, typically in the form of periodic, aperiodic, or event-triggered measurement reports. In some embodiments, event- triggered measurement reports may be employed for performing HO predictions, which allow the UE to inform the network when predefined conditions are met, such as those defined by the 3GPP Events A1 to A6. These events serve as key triggers for the network’s decision- making and are based on comparisons of the UE’s measured beam or cell quality against configured thresholds.

[0040] In some embodiments, the CHO may employ a proactive handover technique where the network provides the UE with a list of conditional configurations for potential handovers. This list may include:

[0041] Candidate cells or beams that the UE can hand over to when certain conditions are met.

[0042] Threshold values (e.g., RSRP, RSRQ, SINR) for triggering the handover.

[0043] TTT values that define how long the signal quality must remain above the threshold before the handover is triggered.

[0044] The CHO may allow the UE to autonomously execute the handover when the configured conditions are satisfied, improving handover latency and resilience. 15 DM2\20713381.2F9125-62400

[0045] FIG. 2A illustrates an exemplary design framework 200 for machine learning- based radio link failure (RLF) predictions, in accordance with some embodiments of the present disclosure. In some embodiments, a UE may be in communication with a plurality of BSs or TRPs, wherein the plurality of BSs comprises a serving BS and a plurality of neighboring BSs orTRPs. In one embodiment, the UE may be configured to receive a plurality of beams from the serving BS, wherein the serving BS uses beamforming techniques to create the plurality of beams. Examples of beams that can be received by the UE include but not limited to: Synchronization Signal Block (SSB) beams, Channel State Information Reference Signal (CSI-RS) beams, and handover beams (e.g. a strong beam from the serving cell prior to the handover or a high-quality beam from the target cell after the handover initiation). In another embodiment, the UE may be configured to receive the plurality of beams from the plurality of BSs, wherein each of the plurality of beams is generated by a respective one of the plurality of BSs. For example, the network (e.g. the serving BS) may configure the UE to receive and measure a plurality of M beams denoted by{^^, … , ^^, … , ^^}, where ^^ denotes the i-th beam from the plurality of M beams.

[0046] In some embodiments, the UE may receive the plurality of M beams{^^, … , ^^, … , ^^} at different time points. As illustrated in FIG. 2A, a total of N time points atwhich the UE receives the plurality of M beams are denoted by 204-1 to 204-N in a time axis 202. For instance, at time point 204-1, the UE may receive the plurality of M beams{^^, … , ^^, … , ^^} and take beam measurements for each of the plurality of M beams asillustrated by the blocks vertically aligned with the time point 204-1 in FIG. 2A. Examples of beam measurements that can be taken by the UE include but not limited to: Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), Signal-to- Interference-plus-Noise Ratio (SINR), Channel Quality Indicator (CQI), Beam Reference Signal Strength (BRSS), and Angle of Arrival (AoA). In some embodiments, the beam 16 DM2\20713381.2F9125-62400 measurements of the plurality of M beams taken at the plurality of N time points are reported to a machine learning (ML) model 206 located at the network side , wherein the ML model 206 uses the reported beam measurements as inputs to predict future beam measurements foreach of the plurality of M beams for at least one future time point denoted by 204-P with ^ >^. In some embodiments, the network configures the UE to measure the plurality of M beams from the serving cell (e.g. the serving BS) only. In some other embodiments, the network configures the UE to measure the plurality of M beams from the neighboring cells (e.g. the plurality of neighboring BSs and / or TRPs) only. In yet some other embodiments, the network configures the UE to measure the plurality of M beams from both the serving cell and the neighboring cells.

[0047] In some embodiments, the exemplary design framework 200 illustrated in FIG. 2A may incur large signaling overhead because the UE must periodically report the beam measurements (e.g., Layer 1 Reference Signal Received Power (L1-RSRP) measurements and / or RSRQ measurements) for all the M beams at each time point. In the training stage of the ML model 206, the UE must measure and report all the available M beams to provide a comprehensive dataset for the ML model 206. The beam measurement reporting from the UE allows the ML model 206 to learn patterns across a wide range of beams. This repeated periodic reporting may consume bandwidth, add to network traffic, and require more energy from the UE. In some embodiments, to manage this overhead during the inference stage of the ML model 206 after the ML model 206 has been trained, the network can optimize which beams are measured and reported by the UE. That is, instead of requiring reports on all the considered M beams, the network may only request measurements from a subset of the M beams based on previous reports and predicted beam quality trajectory from the ML model 206. This subset-based reporting approach can reduce the number of measurements the UE need to provide to the network, thereby reducing the overall signaling burden on the network. 17 DM2\20713381.2F9125-62400

[0048] In some embodiments, during the inference stage of the ML model 206 wherein the ML model 206 is configured to generate the beam predictions at the at least one time point 204-P, the network can track the UE’s trajectory and request measurements only for a subset of the M beams based on the previous reports. For example, the network can start byrequesting the UE to report a first subset of beams {^^, … , ^^} with ^ < ^. Then based on thereport for the first subset of beams, the next request to the UE is to report a second subset ofbeams {^^, … , ^^^^}. Next based on the report for the first and / or the second subsets ofbeams, the network may request the UE to report a third subset of beams {^^, … , ^^^^}, andso on. In some embodiments, a portion of the first, second and third subsets of beams is used for predicting RLF events using the ML model 206. In some other embodiments, all the first, second and third subsets of beams is used for predicting RLF events using the ML model 206. This moving window technique allows the network to track how the UE’s radio environment changes over time without needing measurements on every beam at every time instance.

[0049] In some embodiments, while the network does not track the geographical location of the UE, the network can use the signal strength patterns as a proxy for UE’s trajectory. That is, as the UE moves, some beams may weaken while others may become stronger in signal strength, reflecting changes in the UE’s position. Therefore, the network may discard beams with low signal strength from future measurement requests and adds new candidate beams that might have improved signal quality. This approach allows the network to dynamically adjust the measurement set based on the UE’s signal quality patterns, not its exact location. In some embodiments, the size and selection of the subsets of beams used for inference may impact the performance of the ML model 206. For example, if the selected subset(s) is too small, the ML model 206 may not capture the full range of beam quality 18 DM2\20713381.2F9125-62400 changes, potentially missing out on the optimal beams. On the other hand, if the selected subset(s) is too large, signaling overhead may be reintroduced.

[0050] In some embodiments, the moving window subset of beams can be determined based on the sequential number of beams (e.g. beams that are adjacent in angle or location) or based on quality metrics such as RSRP ranking. In one embodiment, the network may request only the beams with the highest RSRP values within the window, focusing on the most promising candidates. In another embodiment, the network may start with a subset of beamssuch as {^^, ^^, ^^} and, after receiving the measurement reports for the subset of beams{^^, ^^, ^^} at different previous time points, the network may determine that ^^ is weakeningwhile ^^is strengthening. For the next measurement at the next time point, the network maythen request {^^, ^^, ^^}, following the trend of the UE’s movement represented by beamsignal strength variations. This sequence allows the network to maintain a window of high- quality beams without overwhelming the UE with excessive reporting requirements. In some other embodiments, the network may rank the beams based on their RSRP values over a spatial grid, effectively generating a heat map of signal quality. In some embodiments, the generated heat map can be used to track the UE’s movement based on signal strength variations, without relying on geolocation data.

[0051] In some embodiments, the functionality of the ML model 206 is denoted by ^(∙), wherein ^(∙) aims to calculate the predicted probability for RLF at a future time point ^^, as shown below:

[0052] Prob{RLF occurs

[0053] where +^, / denotes the beam measurement(s) performed for the i-th beam at the j- th previous time point, M denotes the total number of considered beams, N denotes the total 19 DM2\20713381.2F9125-62400 number of previous time points used to predict the RLF, and ^(denotes a future time pointwith ^ > ^. In some embodiments, each measurement +^, / is calculated as: +^, / = 2^, / −24567859:;, wherein 2^, / denotes the raw measurement (e.g. RSRP or SINR) for the <-th beam taken at time ^ / , and the parameter 24567859:;may be a signal quality threshold that represents the minimum acceptable quality level for stable connectivity. For example, +^, / may measure how much the beam quality deviates from the threshold 24567859:;. Positive values of +^, / (when 2^, / > 24567859:;) may indicate acceptable quality, while negative values of +^, / maysuggest degraded quality below the threshold. In some embodiments, the threshold 24567859:;determines the lower bound of the measurement. Measurement values below the threshold may be meaningless such that a measurement signal is discarded if below the threshold. In some other embodiments, the threshold value is the same for stationary UEs and UEs under mobility. In yet some other embodiments, the threshold value is different for stationary UEs and UEs under mobility

[0054] The RLF prediction scheme outlines above provide an autonomous approach for capable UEs to manage beam measurements and handle potential RLF scenarios. In this setup, the UE can take on more responsibility for beam selection and measurement management. That is, instead of waiting for continuous configuration requests from the network, the UE itself can decide which beams to measure based on its current speed and signal quality trends. For example, the UE may begin with an initial set of beams and periodically measure their signal strength. If the UE detects that certain beams in the set have signal strengths below a threshold, it discards those weaker beams and replaces them with new candidate beams in the next measurement cycle. Therefore, this method reduces signaling overhead since the network no longer needs to repeatedly instruct the UE on which 20 DM2\20713381.2F9125-62400 beams to measure. The UE can adapt its beam measurement set on its own, using speed and signal strength information to dynamically adjust the list of candidate beams.

[0055] In some embodiments, while the UE may take the lead in managing beam measurements, the network can still configure the overall setup. The network may provide guidelines for how the UE should handle beam measurements, such as which thresholds to use or how often to adjust the beam set. This configuration allows for a consistent approach for beam management across various UEs, ensuring the UEs follow network-defined rules for optimizing connectivity and reducing overhead. In some other embodiments, the ML model 206 is implemented in the network side, and the network may use the UE-reported beam measurements as inputs to the ML model 206 that predicts RLF. If the ML model 206 indicates that all beams in the UE’s reported set are likely to fall below the signal threshold at a future time ^(, then the network may predict that the UE will experience an RLF at ^(. In this case, the network may send an “RRCReEstablishConnection” configuration message to the UE, wherein the “RRCReEstablishConnection” message includes reconnection information that the UE can use to reconnect to the same network after the RLF occurs, streamlining the process of finding and connecting to a new beam.

[0056] In some other embodiments, the network may reuse the RRC Reconfiguration message to configure the UE for reconnection after an RLF occurs. For example, the RRC Reconfiguration message may be extended with additional Information Elements (IEs) specifically for post-RLF recovery, wherein the additional IEs may comprise recovery beam list IE, RLF timing and action information IE, and / or reconnection priority flag. By incorporating these IEs, the RRC Reconfiguration message can support the reconnection process after an RLF, thereby streamlining the recovery without introducing a new message type. In this approach, the network avoids the need to send a dedicated 21 DM2\20713381.2F9125-62400 "RRCReEstablishConnection" message while ensuring that the UE is guided through the reconnection process efficiently.

[0057] In some embodiments, the “RRCReEstablishConnection” message may comprise a list of candidate beams (e.g. SSBs and / or CSI-RS) for the UE to search after the RLF. This list may be based on the beams the UE reported just before the RLF, supplemented by additional beams predicted by the ML model 206 to be potential reconnection candidates. Inone embodiment, the UE’s last report before an RLF included the beams {^^, ^^^^, … , ^^},then the network might include a broader set {^^=^, ^^=^, ^^, ^^^^, … , ^^, ^^^^, ^^^^, ^^^^} inthe “RRCReEstablishConnection” message. This extended list including the new beams^^^^, ^^^^} predicted by the ML model 206 may be made available to theUE for reestablishing connection with the network after the RLF. By storing the “RRCReEstablishConnection” message and the candidate beam list, the UE can quickly reconnect after the RLF, bypassing the need to re-initiate a new search. The network’s inclusion of predicted beams makes it easier for the UE to find a suitable reconnection point.

[0058] In some embodiments, the ML model 206 is trained to map the post-RLF or the re-establishment candidate beam set to a specific pre-RLF beam set. This mapping allows the network to make informed decisions about which additional candidate beams to include in the reconnection message (e.g. “RRCReEstablishConnection” message), improving the UE’s chances of successful reconnection. In some other embodiments, the UE may proactively measure the post-RLF candidate beam set before an RLF occurs, allowing the UE to switch to a new beam in time to avoid the RLF entirely. In this case, the UE connects to the strongest available candidate beam before the connection fails, so that the downtime can be reduced. In some other embodiments, the UE may be pre-configured by the BS with a plurality of beam sets to support proactive recovery in case of an RLF, wherein the plurality 22 DM2\20713381.2F9125-62400 of beam sets may comprise: a primary beam set used for regular connectivity and mobility operations under normal conditions, and a recovery beam set comprising a broader set of candidate beams, wherein the broader set of candidate beams comprises beams from the serving cell and neighboring cells that the UE can autonomously monitor and switch to if an RLF is detected. The BS can configure these beam sets in advance as part of the RRC Reconfiguration message, ensuring that the UE has access to all necessary recovery beams without requiring a new configuration message during the RLF recovery process. This approach reduces the dependency on receiving real-time instructions from the BS when the connection is unstable. In some embodiments, the UE may proactively measure the recovery beam set before an RLF occurs to identify the strongest available candidate beam. If the signal quality of the current beam degrades below a threshold, the UE may autonomously switch to the strongest candidate beam within the recovery beam set. This allows the UE to maintain connectivity and avoid downtime without waiting for a new configuration message from the BS.

[0059] In some embodiments, the ML model 206 illustrated in FIG. 2A allows the UE to predict RLF events in advance, rather than reacting to signal degradation after the RLF occurs. On the other hand, traditional RLF techniques often rely on fixed thresholds for metrics such as RSRP or SINR and react to these metrics only once they cross predefined limits, which can lead to abrupt connection drops or suboptimal handover timing. The ML- based framework illustrated herein, however, enables anticipatory actions, allowing the network and UE to initiate handovers or reconnection processes proactively, reducing the likelihood of dropped connections. In addition, prior art methods generally require the UE to measure all the configured beams continuously, resulting in high signaling overhead and increased UE power consumption. The moving window approach incorporated in the design framework 200 herein allows the UE to select and report only the most relevant subset of 23 DM2\20713381.2F9125-62400 beams over time. In this way, the UE can focus on a manageable number of beams while still providing adequate coverage for RLF predictions.

[0060] In some embodiments, the UE may derive the quality of its serving and neighboring cells based on multiple beam measurements. The UE may collect beam-specific samples (e.g., RSRP or SINR) from SSB and / or CSI-RS resources configured by the BS. These beam measurements can be processed at different layers (Physical Layer 1 and Radio Resource Control (RRC) Layer 3) before being consolidated into a final report sent to the BS.

[0061] FIG. 2B illustrates an exemplary design framework 210 for deriving cell quality from beam measurements and generating reports at different layers (L1 and L3), in accordance with some embodiments. In some embodiments, beam measurement samples are filtered, consolidated, and evaluated against thresholds before the UE sends reports to the network (e.g. the BS). As illustrated in FIG. 2B, a UE may receive a plurality of beams (e.g. Beam 1 to Beam K) at point 212 and take beam-specific measurements at the physical layer (e.g., RSRP, SINR). These raw measurements represent the initial sampling of signal quality for each beam. Then the raw measurements at point 212 may undergo internal filtering at Layer 1 to derive smoothed values for each beam at point 214. In some embodiments, the specific implementation of the Layer 1 filtering is UE-specific.

[0062] In some embodiments, the filtered measurements may be passed from Layer 1 to Layer 3 as beam-specific measurements. At Layer 3, the beam-specific measurements may be consolidated to derive an overall “cell quality” metric illustrated by the “beam consolidation / selection” block and point 216 in FIG. 2B. This step combines multiple beam measurements to provide a single quality score for the cell. The configuration of this beam consolidation module may be provided by RRC signaling, and the reporting period at point 216 may equal one measurement period at point 214. In some embodiments, at Layer 3, 24 DM2\20713381.2F9125-62400 additional filtering may be applied to smooth the cell quality measurement before it is evaluated for reporting, as illustrated by the “layer 3 filtering for cell quality” block. In some other embodiments, the filtered cell quality value at point 218 is the final input for determining whether a measurement report should be triggered. In some embodiments, Layer 3 filtering for cell quality and related parameters used do not introduce any delay in the sample availability between point 216 and 218. The configuration of this Layer 3 filtering for cell quality module may be provided by RRC signaling, and the reporting period at point 218 may equal one measurement period at point 216. The UE may evaluate whether the filtered cell quality measurement meets the criteria to trigger a report, as illustrated by the “evaluation of reporting criteria” block. This evaluation may be based on the thresholds and events configured by the network (e.g., TS 38.331 events A2 when the signal quality of the serving cell becomes worse than a threshold, A3 when the signal quality of a neighboring cell becomes better than the serving cell by a defined offset, or A5 when the signal quality of the serving cell becomes worse than a first threshold, and the signal quality of a neighboring cell becomes better than a second threshold). If the reporting criteria are met, the UE may send a measurement report comprising the cell and beam quality metrics to the network at point 222. In some embodiments, the evaluation may consider additional inputs 220, wherein the additional inputs 220 may comprise cell quality derived from a neighboring cell’s beams. The configuration of this evaluation of reporting criteria module may be provided by RRC signaling, and the UE may evaluate the reporting criteria at least every time a new measurement result is reported at point 218 and 220.

[0063] In some embodiments, in parallel with the cell quality derivation, the beam- specific measurements are further filtered at Layer 3 to prepare for beam-specific reporting. For example, each of the beam measurements filtered at Layer 1 may be processed by L3 beam filtering to produce a plurality of L3-filtered beams at point 224, wherein the plurality 25 DM2\20713381.2F9125-62400of L3-filtered beams at point 224 is used to select the top > (1 ≤ > ≤ A) beams to beincluded in the report, wherein the top > best beams may be selected based on their signal quality (e.g., highest RSRP values), as illustrated by the “beam selection for reporting” block. The configurations of the L3 beam filtering module and the beam selection and reporting module may be provided by RRC signaling, and the reporting period at point 224 may equal one measurement period at point 214. In some embodiments, information (e.g. beam IDs and measurement results) of the selected > top beams may be included in the final report sent to the network. In some other embodiments, cell and beam measurement quantities to be included in measurement reports are configured by the network. In yet some other embodiment, the number of non-serving cells to be reported may be limited through configuration by the network. In still some other embodiments, Layer 3 filtering for cell quality and related parameters used do not introduce any delay in the sample availability between point 224 and 226.

[0064] In some embodiments, the network can configure the UE to filter specific cells when performing event evaluations and generating measurement reports. For example, cells belonging to an exclude-list configured by the network may be ignored during event evaluation and reporting. Conversely, if an allow-list is configured, only the cells specified in the allow-list are considered during event evaluation and reporting. Additionally, the network may configure the content of beam measurements to be included in the UE’s measurement reports. The configuration may specify: reporting only the beam identifier, reporting the measurement result along with the beam identifier, or omitting beam-specific reporting entirely (i.e., no beam details included in the report). This flexibility in configuration allows the network to control which cells and beam measurements are considered, helping to optimize signaling overhead and ensure that only relevant information is reported by the UE. In some other embodiments, how filtered beam measurements are handled according to the 26 DM2\20713381.2F9125-62400 allow-list / exclude-list may be indicated at point 218 in FIG. 2B. In yet some other embodiments, determination of the structure of the measurement report (e.g., beam ID only, beam ID with results, or no beam reporting) by the network is performed at point 222 in FIG. 2B.

[0065] In some embodiments, the exemplary machine learning-based RLF prediction scheme comprises two stages: a training stage and an inference stage. FIG. 3A illustrates a high-level training stage procedure 300 of an ML model 306 for RLF predictions, in accordance with some embodiments. In some embodiments, a BS 302 may be in communication with a UE 304, wherein the BS 302 transmits a beam measurement configuration message to the UE 304, wherein the beam measurement configuration message comprises information on at least one of: beam measurement type(s) (e.g. RSRP, RSRQ, SINR), beam measurement frequency, and beam measurement time point(s) and duration(s). By configuring the UE 304 (and other UEs covered by the BS 302) with this information, the BS 302 can receive consistent and standardized beam measurements comparable across different UEs.

[0066] Upon receiving the beam measurement configuration message, the UE 304 may perform the required beam measurements as specified by the BS 302 and periodically transmit a beam measurement report back to the BS 302. This report may include the signal quality metrics for each configured beam, such as RSRP, RSRQ and SINR, as well as time- stamps to notify the beam measurements and RLF occurrences. These measurements provide the BS 302 with insights into the quality and performance of various beams over time. This data can be used for training the ML model 306, as it reflects the signal conditions that the UE 304 experiences. In some embodiments, if the UE 304 experiences an RLF during the training period, the UE 304 may transmit an RLF report to the BS 302 after re-establishing a 27 DM2\20713381.2F9125-62400 connection, wherein the RLF report includes information about the circumstances leading up to the RLF, such as times of the RLF occurrence and beam measurements prior to the RLF. This RLF report, combined with the beam measurements, can allow the BS 302 to compile a dataset that links specific beam conditions with RLF events. This dataset may be then used to train the ML model 306 to recognize patterns that indicate RLF.

[0067] In some embodiments, the RLF Report may be reused as specified in Release 16 (TS 38.331) and sent using either UEInformationResponse message in response to a UEInformationRequest from the BS 302, or RRCReconfigurationComplete message if the UE 304 has successfully received an RRC reconfiguration for re-establishment. The RLF Report may include standard fields such as rlf-Cause-r16 comprising ENUMERATED values indicating the cause of the RLF, such as t310-Expiry, beamFailureRecoveryFailure, rlc- MaxNumRetx, lbtFailure-r16, bh-rlfRecoveryFailure, and other causes as specified in TS 38.331. In addition to the standard fields, the RLF Report may be extended to include additional information not currently available in Release 16, such as: beam-specific measurements prior to RLF (e.g. detailed RSRP, RSRQ, or SINR values for the failed beams leading up to the RLF), pre-failure timing and conditions (e.g. the specific time when the RLF occurred and the duration of degraded connectivity prior to the failure), mobility context information (e.g. details on the UE’s speed, location, and handover-related context at the time of the RLF), recovery attempts (e.g. information on any beam failure recovery attempts made before the RLF occurred, including whether a fallback beam switch was attempted). This additional information can help the BS 302 better understand the cause of the RLF and improve its ML-based training process by associating specific RLF causes and beam conditions with network performance patterns. By reusing the existing RLF-Report structure from TS 38.331 and leveraging the UE Information Request / Response and / or 28 DM2\20713381.2F9125-62400 RRCReconfigurationComplete messages, this approach reduces signaling overhead and maintains compatibility with existing 5G RRC procedures.

[0068] In some other embodiments, the RLF Report may include a random access report (RA-Report) to specify the purpose of the random access procedure that failed. The RA- Report may indicate the purpose of the RA procedure using the raPurpose-r16 ENUMERATED field, which can include the following values: accessRelated (e.g. random access initiated for general access purposes), beamFailureRecovery (e.g. random access triggered due to a beam failure recovery procedure), reconfigurationWithSync (e.g. random access for reconfiguration requiring synchronization), ulUnSynchronized (e.g. random access due to uplink desynchronization), schedulingRequestFailure (e.g. random access due to a failure to obtain a scheduling request, noPUCCHResourceAvailable (e.g. No PUCCH resource was available for uplink signaling), requestForOtherSI (e.g. request for other system information), msg3RequestForOtherSI-r17 (e.g. request for system information in the msg3 procedure), lbtFailure-r18 (e.g. random access triggered due to a listen-before-talk failure in unlicensed bands). The inclusion of the RA-Report provides additional context regarding why the random access procedure was initiated and why it may have failed. For example, the beamFailureRecovery value indicates that the failure was linked to a beam recovery attempt, whereas schedulingRequestFailure points to a failure to obtain an uplink grant. This additional information can improve the training of the ML model 306 by providing insights into the underlying reasons for random access-related failures, allowing the network to distinguish between RLFs caused by beam-related issues versus those related to resource allocation or scheduling constraints.

[0069] In some embodiments, the ML model 306 is implemented as a network-side model. In some other embodiments, the ML model 306 is implemented as a UE-side model 29 DM2\20713381.2F9125-62400 since the ML model 306 can be used at the UE side during the inference stage. However, in these embodiments, the model training for the ML model 306 can occur at the network side (e.g. the BS side). In some embodiments, during the training stage, a plurality of UEs comprising the UE 304 may be within the cell covered by the BS 302, and the plurality of UEs may be configured to report their respective beam measurements and any RLF events to the network (e.g. the BS 302). The respective report transmitted by each of the plurality of UEs may comprise information such as: the time of the RLF and the beam measurements just prior to the RLF. The BS 302 may then collect these reports and use them to create a comprehensive dataset that includes both beam measurement trends and the actual RLF occurrences. The ML model 306 can be then trained to recognize patterns that lead to RLF, effectively mapping the reported beam measurements to the observed RLF events. This allows the ML model 306 to learn which signal quality trends are likely to precede an RLF. In some embodiments, the ML model 306 is designed to be cell-specific, meaning that the model parameters (such as the number of layers, weights, and other hyperparameters) are tuned specifically for each cell’s radio environment. This ensures that the ML model 306 can accurately reflect the unique conditions of each cell. Because each cell may have different radio characteristics and beam configurations, customized model parameters are important to providing reliable predictions.

[0070] In some embodiments, in order to enhance the accuracy of the ML model 306, the ML model 306 can require a mixed training dataset that includes beam measurements and RLF data from UEs served by multiple neighboring cells. This broader dataset enables the ML model 306 to understand variations in signal quality and beam performance beyond a single cell. However, using data from too many cells can significantly increase the model’s complexity and size, making it harder to deploy on the UE side, where processing power and memory are limited. Therefore, a practical trade-off may be to train the ML model 306 on 30 DM2\20713381.2F9125-62400 data from 2 or 3 neighboring cells. This setup provides enough diversity in the dataset for the ML model 306 to generalize well, without making the model too large or too complex for the UE 304 to run effectively.

[0071] Once the ML model 306 is trained, it may be deployed to the UE 304 for performing predictions in the inference stage. During this stage, the UE 304 may use the trained ML model 306 to predict potential RLF events based on real-time beam measurements. In some embodiments, as the UE 304 continuously measures signal quality across a subset of beams, the UE 304 inputs these beam measurements into the ML model 306. The ML model 306 may then assess the likelihood of RLF based on the patterns it learned during training. By performing inference on the UE side, the ML model 306 can make quick, real-time predictions about the UE’s connectivity, allowing it to proactively switch beams or prepare for a handover before a connection failure occurs.

[0072] In some embodiments, a grid of beams approach may be used to divide the total set of beams into disjoint subsets that cover distinct spatial areas or directions within a cell. The grid in this approach may be a predefined pattern of beams that the network uses to cover the entire cell area systematically. By organizing the beams into these subsets, each subset represents a specific section of the cell’s coverage area. The beams in each subset can be spatially fixed, meaning they consistently target the same area within the cell regardless of the UE’s movement or the network’s dynamic conditions. In some embodiments, in this grid of beams approach, explicit information on the UE’s speed and location is not required. Instead, the ML model 306 indirectly captures the effects of speed and location through the beam measurement reports from the UE 304. As the UE 304 moves, the signal quality of certain beams will increase while others decrease, depending on the UE’s proximity to those beams in the grid. This pattern of beam quality changes implicitly reflects the UE’s 31 DM2\20713381.2F9125-62400 movement within the cell and provides the ML model 306 with insights into the UE’s relative location and trajectory. For example, if a UE moves towards a particular beam subset in the grid, the ML model 306 can detect this through the increase in signal quality for beams in that subset, while beams in the previous subset might degrade. This information helps the ML model 306 learn which beam subsets are relevant to track as the UE 304 moves through the cell.

[0073] In some embodiments, the training stage procedure 300 illustrated in FIG. 3A may offer advantages over prior art in RLF prediction and management by incorporating ML techniques that are adaptive, data-driven, and capable of making proactive predictions. That is, the ML model 306 disclosed herein can predict RLF events before they occur, allowing the UE or the network to take preventative measures, such as switching to a more stable beam or preparing for a handover. On the other hand, traditional beam management methods are typically reactive, relying on pre-set thresholds (e.g., signal strength) that trigger an action only after signal quality has already deteriorated, which often leads to abrupt connection losses. The proactive approach of this ML-based design can reduce RLF occurrences and enhance connection continuity.

[0074] FIG. 3B illustrates a high-level inference stage procedure 310 using a trained ML model 316 for RLF predictions, in accordance with some embodiments. In some embodiments, the trained ML 316 may be trained using the procedure outlined in FIG. 3A. In some embodiments, when a UE 314 enters the serving cell’s coverage provided by a BS 312, the BS 312 may be configured to transmit information of the trained ML model 316 to the UE 314 using a model transfer signal. For example, the BS 312 may transmit information about the trained ML model 316 via System Information Block (SIB) messages, wherein the SIB messages comprise: a model ID to identify the specific version of the trained ML model 32 DM2\20713381.2F9125-62400 316, or a directive to download the parameters of the trained ML model 316 from a specified network location. The UE 314 may then store the parameters of the trained ML model 316 locally and use the trained ML model 316 to perform RLF predictions in real time.

[0075] In some embodiments, if the trained ML model 316 predicts an RLF for a future time, the UE 314 may be triggered to send an RLF prediction message to the network (e.g. to the BS 312), indicating that an RLF is likely to occur within the predicted time frame, as illustrated by the “predicting RLF” signal in FIG 3B. The RLF prediction message may comprise: the probability of RLF occurrence expressed as a percentage, and the predicted time to RLF indicating when the RLF is expected (e.g., 10ms later, or at a future time point ^(). Additional details on the actual measured or predicted signal quality metrics (such as RSRP, RSRQ, or SINR) can also be provided to the network with context about the beam conditions that are leading to the predicted RLF.

[0076] In some embodiments, upon receiving the RLF prediction message, the network (e.g. the BS 312) may respond by sending a reconnect message to the UE 314. This reconnect message may provide instructions to help the UE 314 re-establish its connection if the RLF occurs. In some embodiments, the reconnect message may include instructions for the UE 314 on what steps to take following the RLF, such as initiating a Random Access Channel (RACH) process. This process allows the UE 314 to reconnect to the network by requesting resources on a dedicated random access channel. The reconnect message may also include a list of candidate beams or cells that the UE 314 should search for when re-establishing the connection. These candidate beams can be selected based on: predicted beam performance or alternative cell coverage that may support the UE 314 after the RLF. By proactively providing reconnection instructions, the network ensures that the UE 314 is prepared to 33 DM2\20713381.2F9125-62400 quickly recover from the RLF, reducing potential downtime and enhancing service continuity.

[0077] In legacy systems, when the UE 314 detects an RLF, it initiates the RRC Re- establishment procedure, wherein the RRC Re-establishment procedure comprises: cell selection (e.g. the UE scans for and selects a suitable candidate cell) and RRC Re- establishment Request (e.g. the UE sends this request to the selected cell to re-establish its connection). If the selected cell is one of the configured target cells (e.g., in Conditional HO (CHO) or Conditional Layer 1 / Layer 2 triggered mobility (C-LTM) scenarios), the handover can be completed in the target cell without requiring a full RRC re-establishment.

[0078] In some embodiments, to support and enhance the legacy systems, the reconnect message sent in response to the RLF prediction may include the following information: candidate target cell / beam list which specifies the prioritized cells or beams for re- establishment or CHO completion, particularly in cases where conditional handovers are pre- configured, pre-configured CHO / C-LTM information which reinforces the UE’s ability to seamlessly complete the handover to a pre-configured target cell without requiring full re- establishment, and fallback RACH Instructions which provides steps to initiate a RACH process for re-establishment if the pre-configured target cells are unavailable. This reconnect message may also reduce re-establishment delays by narrowing the beam / cell selection process, especially in high-mobility scenarios. For example, if the selected target cell from the CHO configuration is known to provide reliable connectivity in the UE’s predicted movement direction, the UE can immediately proceed with the handover rather than performing additional measurements for cell selection. This approach ensures that the UE leverages pre-configured CHO or C-LTM information when available, while still providing 34 DM2\20713381.2F9125-62400 backup options to handle RLF scenarios where the pre-configured target cells may be unreachable.

[0079] In some embodiments, the network may configure the UE 314 with a set of nbeams {B^, B^, … , BC} (with B^ denoting the i-th beam) which the UE 314 needs to measureand report, and a set of window parameters such as window length “w” and moving steps “s”, wherein the window length “w” defines the number of beams that the UE 314 will measure and report in each cycle, and the moving steps “s” defines how many beams that the window will shift forward in each measurement cycle. In some embodiments, UE 314 is configured to maintain an internal counter <, which starts at 0 and helps track which subset of beams is currently being measured. In some embodiments, the UE 314 may start bymeasuring and reporting a subset of beams {B^^^, B^^^, … , B^^D}, where the initial counter isset to < = 0. That is, at the initial measurement cycle, the UE 314 may measure beamstoBD. After completing the first measurement and report, the counter < may be increased by astep size F (i.e. < = < + F), causing the measurement window to shift forward by F beams. Forthe next cycle, the UE 314 may measure and report the beams {B^^H^^, B^^H^^, … , B^^H^D},covering a new subset. This process can be repeated, allowing the UE 314 to gradually measure all beams in smaller, more focused subsets. This moving window provides a sequential and targeted beam measurement approach, which reduces signaling overhead by focusing on a manageable subset of beams in each cycle. From a control perspective, the network can activate or deactivate this moving window mechanism based on the operational needs. For example, when the moving window is deactivated, the UE 314 may measure andreport all the beams {B^, B^, … , BC} for the training of the model. When the moving windowis activated, the UE 314 may measure and report subsets of the beams for the inference of the model. 35 DM2\20713381.2F9125-62400

[0080] In some embodiments, the inference stage procedure 310 illustrated in FIG. 3B may provide enhanced reconnection process with pre-configured instructions. That is, prior art RLF handling techniques may require the UE to conduct a full beam search after an RLF occurs, leading to extended reconnection times. In the inference stage procedure 310 illustrated herein, if an RLF is predicted, the network can send a reconnect message to the UE in advance, with instructions such as initiating the RACH process and identifying candidate beams or cells for reconnection. By providing pre-configured reconnection paths, the ML- based approach disclosed herein may reduce downtime and enable a faster return to service.

[0081] FIG. 4 illustrates another exemplary design framework 400 for handover (HO) predictions, in accordance with some embodiments. In some embodiments, a UE may receivea plurality of M beams {^^, … , ^^, … , ^^} at a plurality of N time points from 404-1 to 404-Nas illustrated by the time axis 402 FIG. 4. In some embodiments, the beam measurements of the plurality of M beams taken at the plurality of N time points are reported to an ML model 406, wherein the ML model 406 uses the reported beam measurements as inputs to predict the HO probability that the UE should hand over to at least one cell (e.g. the cell j) at a futuretime point 404-P (^ > ^). In some embodiments, the ML model 406 at the UE may predictthe HO probability based on beam measurements and related signal conditions. However, the final handover decision may also be made based on additional information available at the network, such as the load conditions of the candidate cells and overall resource availability. In such cases, the predicted HO probability may serve as a recommendation that the network may consider before initiating the handover.

[0082] In some embodiments, as illustrated in FIG. 2B, the plurality of M beams{^^, … , ^^, … , ^^} may comprise multiple layers of filtering based on the measurement flowdefined in TS 38.300: 36 DM2\20713381.2F9125-62400

[0083] Layer 1 filtering: Beam-specific measurements at the physical layer (e.g., RSRP) are averaged internally to derive beam quality metrics.

[0084] Layer 3 filtering: These beam-specific measurements are aggregated and filtered at the Radio Resource Control (RRC) layer to derive cell-level quality metrics.

[0085] Beam Consolidation: The UE may select the top X beams based on their quality for reporting purposes. In some embodiments, the UE may consolidate multiple beam measurements at different layers (Layer 1 and Layer 3) before transmitting reports to the network. The design framework 400 may leverage these consolidated measurements, which may include the > best beams from the serving and neighboring cells as selected by the UE based on its beam quality filtering configuration. By selecting and reporting the most relevant beams, the UE may ensure that the ML model 406 operates on high-quality input data while minimizing measurement and reporting overhead.

[0086] In some embodiments, the design framework 400 leverages the beam measurements described above at various levels to generate predictions on future RLF or HO events. The ML model 406 may utilize filtered Layer 1 and Layer 3 measurements for input, including the reported beam-specific and cell-specific qualities for both the serving and neighboring cells. The use of filtered measurements reduces noise and ensures that the predictions are based on consistent trends in the radio environment.

[0087] In one embodiment, the ML model 406 is implemented in the UE. In another embodiment, the ML model 406 is implemented in the serving BS. In some embodiments, the serving cell (e.g. the serving BS) may have a plurality of neighboring cells, and the Synchronization Signal Block (SSB) measurements from the plurality of neighboring cells may also be incorporated as inputs of the ML model 406 to enhance the handover prediction accuracy. In some embodiments, the ML model 406 adds SSB measurements from the 37 DM2\20713381.2F9125-62400plurality of neighboring cells (denoted as cell j, where J = 1, … , K) to the input data, whereinthe SSB measurements from the plurality of neighboring cells provide information on alternative beams that the UE might connect to if the current serving cell’s signal weakens. Including these neighboring cells’ SSB measurements also allows the ML model 406 to predict the respective handover probability for each of the plurality of neighboring cells (i.e.the handover probability for cell i, where < = 1, … , K). For example, based on the predictionresults, the ML model 406 may generate a prediction report comprising an indication to indicate which neighboring cell that the UE should hand over to in a future time point, and the UE may transmit the prediction report to the serving cell.

[0088] In some embodiments, during the training stage of the ML model 406, the beam measurements from the serving cell (e.g. the serving BS) along with the neighboring cells’ SSB measurements are used as inputs of the ML model 406 and the network (e.g. the serving BS) may provide the handover decisions as part of the training data set. That is, the network may use the beam measurements along with the neighboring cells’ SSB measurements as inputs to train the ML model 406 by considering data on actual handover decisions as the corresponding outputs. For each successful handover in the training data, the ML model 406 learns to associate specific beam measurements (from the serving cell) and neighboring cells’ SSB measurements with the outcome of an associated handover decision.

[0089] In some embodiments, in the inference stage of the ML model 406, the trained ML model 406 may be run on the UE, wherein the UE continuously inputs its current beam measurements from the serving cell, along with the SSB measurements from the neighboring cells, into the trained ML model 406. If the trained ML model 406 predicts that the conditions are suitable for a handover to a specific neighboring cell (e.g., cell J that has the highest predicted HO probability among all the neighboring cells), the UE may generate and transmit 38 DM2\20713381.2F9125-62400 a handover request message to the network. This request indicates a potential handover target and provides information for the network to make an informed decision on whether to proceed with the handover. The handover request message from the UE may include: measured and predicted signal quality metrics (such as RSRP, RSRQ, SINR) for both the serving cell and the neighboring cells, the recommended target cell(s) with the predicted probability of successful handover to each target, and / or best timing for handover execution (e.g. a future time point ^() based on when the trained ML model 406 predicts the UE should transition to the neighboring cell to maintain optimal connectivity. With this information, the serving BS can make a more informed decision on how to proceed with the handover. In some embodiments, the serving BS takes into account the UE’s recommended target cell and timing, along with any additional network conditions, to determine the best moment to send a Handover Command to the UE, wherein the Handover Command specifies the exact timing and target cell for the handover.

[0090] In some embodiments, the functionality of the ML model 406 is denoted by ^(∙), wherein ^(∙) aims to predict the HO probability that the UE should hand over to a specific cell j at a future time ^(, wherein ^(∙) can be expresses as follows:

[0092] where +^, / denotes the beam measurement(s) performed for the i-th beam at the j- th time point from the serving cell, X^, / denotes the SSB measurement(s) performed for the i- th neighboring cell at the j-th time point, M denotes the total number of beams transmitted from the serving cell, N denotes the total number of considered time points used for HO 39 DM2\20713381.2F9125-62400 probability predictions, J denotes the total number of neighboring cells, and ^(denotes afuture time point for predicting the HO probability with ^ > ^. In some embodiments, X^, / ismeasured periodically from time points ^,to ^-.

[0093] In some embodiments, each measurement +^, / is calculated as: +^, / = 2^, / −24567859:;, wherein 2^, / denotes the raw measurement (e.g. RSRP or SINR) for the <-th beam from the serving cell taken at time ^ / , the parameter 24567859:;denotes a signal quality threshold that represents the minimum acceptable quality level for stable connectivity as described as above with reference to FIG. 2A. This computation of measurement +^, / serves as an example of a data preprocessing technique that can be applied before inputting the measurements into the ML model. However, other preprocessing approaches may also be used, and this computation is not a mandatory form of input. In some embodiments, the SSB measurement(s) performed for the i-th neighboring cell at the j-th time point X^, / can beexpressed as: = 2^, / − 2876abcd=e7:: −denotes the raw SSB signalstrength measuremen(s) for the i-th neighboring cell at the j-th time point, 2876abcd=e7::denotes the raw measurement (e.g. RSRP) for the <-th beam from the serving cell taken at the time point ^ / , and ∆ghHijkjH^Hdenotes a small value used to implement hysteresis in handover decisions. That is, in some embodiments, hysteresis may be employed to prevent rapid and frequent handovers (known as the “ping-pong” effect) due to threshold discontinuities, as illustrated by the parameter ∆ghHijkjH^Hin the equation of X^, / above. The “ping-pong” effect may happen when a UE repeatedly switches between two or more neighboring cells due to minor fluctuations in signal strength around a threshold. By applying the hysteresis, a margin or buffer can be added to the handover decision thresholds. For example, a UE can be configured not to initiate a handover to a neighboring cell until the signal from that neighboring cell exceeds the serving cell’s signal by a certain margin (e.g. the hysteresis 40 DM2\20713381.2F9125-62400 value denoted by ∆ghHijkjH^H). This ensures that small, temporary changes in signal quality do not trigger unnecessary handovers. In this way, the signaling overhead can be reduced and the network stability can be improved.

[0094] In legacy L3 event measurements, such as Event A3, an offset is defined to compare the neighbor cell’s signal strength against the serving cell’s signal. Similarly, the hysteresis value ∆ghHijkjH^Hdescribed herein can serve as an offset-like value. However, in this context, the hysteresis is primarily used to prevent rapid and frequent handovers (known as the “ping-pong” effect) by ensuring that the neighbor cell’s signal strength consistently surpasses the serving cell’s signal by a specified margin before a handover is triggered. This approach reduces signaling overhead and improves network stability by avoiding unnecessary handover attempts.

[0095] In some embodiments, the ML model 406 may be configured to map HO prediction probabilities to the corresponding 3GPP measurement events that typically trigger handover decisions, such as Events A2, A3, and A5. These events are defined as follows:

[0096] Event A2: Triggered when the serving cell signal quality becomes worse than a configured threshold.

[0097] Event A3: Triggered when a neighboring cell signal quality becomes a certain offset better than the serving cell.

[0098] Event A5: Triggered when the serving cell signal quality becomes worse than a first threshold, and a neighboring cell signal quality becomes better than a second threshold.

[0099] In some embodiments, the ML model 406 may predict the likelihood that the signal quality metrics (e.g., RSRP) for a neighboring cell will cross the configured thresholds for these events. In some other embodiments, by incorporating Time-to-Trigger (TTT) timers, 41 DM2\20713381.2F9125-62400 the ML model 406 can ensure that the predicted signal quality remains above or below the thresholds consistently before recommending a handover, thereby preventing rapid and unnecessary handovers (i.e., ping-pong effects). The TTT timers may enable the ML model 406 to reflect stable measurement patterns rather than reacting to temporary fluctuations.

[0100] In some embodiments, the ML model 406 may be configured to enhance CHO- based handovers by predicting when to activate or de-activate CHO conditions. For example, the ML model 406 may predict whether the TTT conditions for an Event A3 handover will be met within a given time frame and preemptively recommend candidate cells to optimize the CHO decision-making process.

[0101] In some embodiments, the exemplary design framework 400 provides enhanced handover prediction accuracy, as the design framework 400 incorporates both the serving cell beam measurements and the neighboring cell SSB measurements into the ML model 406, therefore providing a richer dataset for making handover decisions. On the other hand, traditional wireless communication systems may focus only on the serving cell and the strongest neighboring cell, overlooking the broader network context. By including measurements from multiple neighboring cells, the exemplary design framework 400 can improve handover accuracy by selecting the most suitable target cell from a wider pool of options.

[0102] FIG. 5 illustrates a signaling diagram for handover failure (HOF) recovery based on pre-Radio Resource Control (RRC) configuration, in accordance with some embodiments of the present disclosure. In some embodiments, the signaling diagram illustrated in FIG. 5 provides details on the recovery procedure in the event of an HOF as predicted by the ML model 406 in FIG. 4, or in the event of an RLF as predicted by the ML model 316 in FIG. 3B. In some embodiments, a UE 504 may be in communication with a BS 502-1, wherein the BS 42 DM2\20713381.2F9125-62400 502-1 may have a plurality of neighboring BSs 502-2 to 502-3. In some embodiments, the BS 502-1 that is in communication with the UE 504 may be referred to as the “serving cell” or “source cell”, and the BSs 502-2 and 502-3 may be referred to as “neighboring cells”. In some embodiments, the UE 504 may start the procedure by applying its ML model to predict potential HO events. This prediction may be based on an RRC reconfiguration message the UE 504 received from the serving cell BS 502-1, wherein the RRC reconfiguration message includes the necessary parameters and configuration for HO predictions as illustrated above with reference to FIG. 4. In some embodiments, the ML model implemented in the UE 504 may analyze current beam measurements, including the SSB measurements from the neighboring cells, to assess the likelihood of an RLF or the need for a handover.

[0103] In some embodiments, the ML model used for HOF predictions may consider both physical layer signal measurements and network configuration parameters as inputs, such as offset, threshold, and Time-to-Trigger (TTT) values used in event-based mobility reporting. The ML model may also include historical handover success and failure data to account for the impact of network policies and varying load conditions. In some other embodiments, the ML model predicts the likelihood of HOF by incorporating expected network behavior, which may be informed by prior mobility decisions, load-balancing strategies, and recent configuration changes. This enables the UE 504 to infer how the network is likely to behave during a handover attempt and adjust its behavior accordingly. In yet some other embodiments, the network may implement a feedback loop where the UE 504 provides reports indicating the frequency of HOFs and correlated network configurations. This feedback helps the network refine mobility control policies and avoid non-optimal configurations that could lead to HOFs. 43 DM2\20713381.2F9125-62400

[0104] In some embodiments, each of the neighboring cells BSs 502-2 and 502-3 may transmit a respective SSB configuration message to the serving cell BS 502-1, wherein the respective SSB configuration message comprises updated SSBs based on changes in the respective neighboring cell’s configuration or conditions. The updated SSBs provide the serving cell BS 502-1 with the latest beam information from the neighboring cells BSs 502-2 and 502-3, helping maintain an up-to-date configuration that the UE 504 can use when performing measurements or considering handovers.

[0105] Upon receiving the SSB configuration messages from the neighboring cells, the serving cell BS 502-1 may transmit a pre-RRC configuration message to the UE 504, wherein the pre-RRC configuration message comprises at least one of: CSI-RS beams from the serving cell BS 502-1, SSBs from neighboring cells BSs 502-2 and 502-3, wherein the UE 504 may be configured to measure the CSI-RS beams from the serving cell and the SSBs from the neighboring cells and evaluate the quality of potential handover candidates. In addition, the pre-RRC configuration message may include full configurations for the serving and / or neighboring cells, similar to the RRC reconfiguration message, but without conditional reconfiguration details for conditional handover (CHO) or Layer 1 / Layer 2 triggered mobility (LTM). In some embodiments, the step of pre-RRC configuration message transmission may be performed when the UE 504 predicts an RLF or an HOF and reports this prediction to the serving cell BS 502-1. The pre-RRC configuration message may also indicate a list of candidate beams to the UE 504 to ensure a smooth transition to a target cell.

[0106] In some embodiments, after receiving the pre-RRC configuration message, the UE 504 may predict an RLF or an HOF using the process outlined in FIG. 2A or 4. In some embodiments, if an RLF has already occurred (as per legacy RLF procedures), the UE 504 may use the configuration specified in the pre-RRC configuration message to select a new 44 DM2\20713381.2F9125-62400 cell. At this point, the UE 504 may activate and apply the pre-RRC configuration specified in the pre-RRC configuration message, and use it as a guide for selecting: neighboring cell beams to measure, and a target cell in the event of a handover. Based on the pre-RRC configuration, the UE 504 may perform beam measurements on the newly defined candidate beams from both the serving and neighboring cells. These beam measurements can help the UE 504 evaluate the best potential targets for handover by comparing the signal quality of the new beams against the serving cell’s beams.

[0107] In some embodiments, the UE 504 may identify the SSB(s) from a specific neighboring cell (e.g. BS 502-2) from the beam measurements as the target beam and determine the specific neighboring cell (e.g. BS 502-2) to be the best candidate for a handover. This decision may be made based on the relative strength, quality, and stability of the neighboring cell’s signal compared to the serving cell. Then at the predicted optimal time, the UE 504 may initiate the handover procedure to the target cell (e.g. BS 502-2). This can be done through one of the following two mechanisms: CHO-like mechanism or handover command from source cell mechanism. In the CHO-like mechanism, the UE 504 may proactively initiate the handover (e.g. handover to the BS 502-2) without waiting for explicit commands from the serving cell BS 502-1. This allows the UE 504 to move to the target cell as soon as it detects that the signal quality has crossed a favorable threshold. In the handover command from source cell mechanism, the UE 504 may report the target beam (SSB) back to the serving cell BS 502-1, which then transmits an HO command to the UE 504 to execute the handover. This command provides additional instructions for completing the handover to the neighboring cell. This procedure combines predictive modeling and proactive configuration to facilitate seamless and timely handovers. By anticipating the need for a handover and pre-configuring the UE 504 with candidate beams, the network can reduce service disruptions and improve connectivity continuity. This approach is valuable in 45 DM2\20713381.2F9125-62400 scenarios where frequent movement or signal fluctuations make traditional handover methods less effective.

[0108] In some embodiments, the UE 504 continuously monitors the RSRP of both the serving cell and neighboring cells (e.g., through SSB signals). When the difference in signal strength between a neighboring cell and the serving cell surpasses a configured threshold, the serving cell BS 502-1 may initiate a handover to the neighboring cell. To overcome the ping- pong effect, several techniques can be used, such as hysteresis, L3 filtering, and time-to- trigger, wherein the hysteresis may be referred to as a buffer value to prevent rapid handover triggers, the L3 filtering may be referred to as a process to smoothen the measurement reports over time, and the time-to-trigger may be a timer to ensure that the threshold condition is sustained for a certain duration before the handover is triggered. In 5G NR, 3GPP has specified various conditions and thresholds to enhance these approaches, allowing operators to adjust the parameters based on deployment scenarios. In some embodiments, in the training stage of the ML model implemented in the UE 504, similar threshold conditions can be applied to create a dataset for the ML model. That is, the UE 504 may measure and report the signal strengths of beams and neighboring SSB measurements whenever the threshold condition for a potential handover is met. If a handover occurs, the UE 504 may report the signal measurements to the network along with information on handover success or failure. This reported dataset may be used to train the ML model by allowing it to learn the conditions under which handovers are successful versus when the handovers fail. Over time, the ML model can be “fine-tuned” to determine the optimal thresholds for triggering handovers in the inference stage, balancing handover accuracy and stability.

[0109] In some embodiments, a reinforcement learning (RL) scheme or model can be used to allow the ML model implemented in the UE 504 to learn through trial and error by 46 DM2\20713381.2F9125-62400 interacting with the environment and receiving feedback in the form of rewards. In some embodiments, in the RL scheme, an autonomous agent (e.g. the ML model implemented in the UE 504) may be initially trained on synthetic data that includes various beam / SSB strength values and the corresponding handover probabilities. Then during the filed of operation, the RL scheme may perform an action-reward process. That is, in the action process, the RL scheme may decide whether to initiate a handover when it detects that the neighboring cell’s RSRP (or other signal quality metric) is greater than that of the serving cell. Then in the reward process, after the handover is completed (or if it fails), the RL scheme receives feedback based on the RSRP or other signal strength metric of the new serving cell’s beam(s). A successful handover that improves signal quality would result in a positive reward, while a failed or suboptimal handover would result in a lower reward. This action-reward feedback allows the RL scheme to adjust its parameters based on the outcomes, improving its decision-making process over time. As the RL scheme gains experience, it learns to optimize handover conditions to improve connection quality while reducing unnecessary handovers, therefore generating more accurate handover decisions.

[0110] In some embodiments, handover decisions may incorporate a more adaptive threshold based not only on the signal strength (e.g. RSRP) of the serving and target cells but also on the variability in these signals over time. For example, the ML model implemented in the UE 504 may learn to apply offsets to the standard handover criteria based on various conditions. That is, the ML model may add a dynamic offset when the serving cell’s signal strength is below a specific threshold, prompting earlier handover actions to improve connection stability. In some other embodiments, the ML model may hold off on the handover if the UE 504 is moving at a speed higher than a threshold, thereby reducing the likelihood of unnecessary handovers and improving handover success rates. In yet some other 47 DM2\20713381.2F9125-62400 embodiments, a handover action or decision can be made when the serving cell’s signal strength is expressed according to the following inequality:

[0112] where lml^Hjkn^Co=pjqqdenotes the RSRP of the serving cell, lml^irkoji=pjqqdenotes the RSRP of the target neighboring cell, ∆ represents an offset that adjusts the handover threshold dynamically, wherein ∆ may be calculated as a function of the signalvariability for both the serving and target cells: ∆= stuv^<wu(xHjkn^Co=pjqq, xirkoji=pjqq),wherein xHjkn^Co=pjqqdenotes the standard deviation of the serving cell’s received signal strength (e.g. RSRP) computed over several recent measurements, xirkoji=pjqqdenotes the standard deviation of the target cell’s received signal strength (e.g. RSRP) computed over several recent measurements, and stuv^<wu denotes the function that defines how the offset ∆ is calculated based on the standard deviations of the serving cell’s and the neighboring cell’s signal strengths. Examples of the function used calculate the offset ∆ include: linearfunctions such as: ∆= ^^ ∙ xHjkn^Co=pjqq + ^^ ∙ xirkoji=pjqq with ^^ and ^^ being the weightparameters, non-linear functions such as ∆= yx^ ^Hjkn^Co=pjqq + xirkoji=pjqq , or conditionalfunctions such as: ∆= xHjkn^Co=pjqq − xirkoji=pjqq if xHjkn^Co=pjqq > xirkoji=pjqq and ∆= 0otherwise. By considering the standard deviations of signal strength, this method adapts the handover threshold based on how stable or unstable the signals are for both the serving and the target cells. This approach can be useful in scenarios where signal strength varies due to environmental changes, interference, or UE movement. In some embodiments, the specific method for deriving the offset ∆ may be implementation-dependent, allowing flexibility for the UE 504 to optimize the calculation based on design choices and performance goals. 48 DM2\20713381.2F9125-62400

[0113] In some embodiments, the values of xHjkn^Co=pjqqand xirkoji=pjqqare provided as inputs to the ML model implemented in the UE 504, enabling the ML model to evaluate the stability of the candidate cells. Then the ML model may use the provided inputs to adjust the handover decision dynamically based on the observed variability of the signals. In some embodiments, the ML model learns specific patterns of signal fluctuation that are associated with successful or unsuccessful handovers, which allows the ML model to refine the value of Δ adaptively, optimizing the ML model for different conditions and reducing unnecessary handovers caused by transient signal variations.

[0114] In some embodiments, the ML model implemented in the UE 504 can be enhanced by training the ML model with data collected from the UE 504 while the UE 504 is moving at different speeds. During the training stage, the ML model may be provided with data sets that include high mobility UE measurements (e.g., UE is moving at fast speeds, such as in vehicles) alongside low mobility UE measurements (e.g., UE is moving at low speeds or stationary). This variety of mobility scenarios allows the ML model to learn the signal characteristics and handover patterns associated with different speeds, increasing the ML model’s versatility and accuracy across a range of mobility conditions. By training on this mixed data set, the ML model may predict the RLF or HO events for high-speed UEs by gaining a better understanding of how rapid signal fluctuations correlate with RLFs or HOs, allowing for more accurate predictions in high-mobility situations. In some embodiments, a larger, more diverse training data set inherently increases the ML model’s size and complexity, as the ML model needs to learn a broader set of conditions and adapt its parameters to a wider range of scenarios. This increased complexity may lead to higher computational requirements, which can impact the UE’s processing resources and battery life during inference. As a result, trade-offs may be necessary to balance model accuracy and 49 DM2\20713381.2F9125-62400 resource efficiency, especially in scenarios where the ML model is run on resource- constrained devices such as mobile UEs.

[0115] In some embodiments, to address the challenges associated with increased complexity, the network can reduce the number of required measurements by limiting the candidate beams and SSBs that the UE 504 needs to monitor. This can be achieved by shortening the measurement window size. By narrowing the measurement scope, the network can reduce the data input into the ML model, thereby reducing the processing load on the UE 504. For example, the network may provide assistance information to the UE 504, wherein the assistance information comprises: limiting the number of beams or SSBs to those most relevant for the UE’s current location within the cell, and / or focusing on beams that are more likely to provide stable signals for the UE 504 in specific regions, and reducing the need for the UE 504 to measure all possible beams. In some embodiments, the network may be configured to optimize measurement requirements by providing location-based assistance information to the UE 504. For example, if the UE 504 is near the cell boundary, the network might instruct the UE 504 to focus on neighboring cell SSBs and beams likely to be handover candidates. As another example, if the UE 504 is in a high-mobility scenario, the network can restrict measurements to key beams known to perform well in such conditions, reducing unnecessary measurements and saving resources. This approach ensures that the UE 504 focuses its measurements on the most relevant signals, improving efficiency while maintaining prediction accuracy.

[0116] In some embodiments, when an RLF or HOF occurs, the UE 504 may be responsible for reporting these events to the serving cell BS 502-1. For example, when an RLF occurs, the UE 504 may lose its connection to the serving cell, and the UE 504 may be configured to store relevant information about the RLF event, including radio measurements 50 DM2\20713381.2F9125-62400 (e.g., RSRP, RSRQ, SINR) just prior to the failure. Once the UE 504 successfully reconnects to the network, the UE 504 may report the RLF details to the network. This post-reconnection report may include: the radio measurements leading up to the RLF, and / or additional context such as the UE’s speed, location, and time of the RLF. These details help the network analyze the cause of the failure and identify potential areas for improvement. These stored data provide the network with insights into the conditions under which the RLF occurred, enabling more targeted optimizations and enhancements to the network’s handover and connection stability mechanisms.

[0117] In some embodiments, when an HOF occurs, the UE 504 may be unable to successfully complete a handover to a target cell. In such a case, if the UE 504 is still connected to the source cell (i.e., the serving cell connection remains intact despite the failed handover attempt), the UE 504 can report the HOF directly to the serving cell. On the other hand, if the UE’s connection to the source cell was also lost, the UE 504 can initiate a connection reestablishment procedure to restore communication with the network. After reestablishing the connection, the UE 504 may then report the HOF event to the network. Similar to RLF reporting, the UE 504 may also provide contextual information such as speed, location, and the time of the HOF event. This helps the network understand the mobility and environmental conditions during the failure.

[0118] In some embodiments, the threshold used for handover or RLF detection might be dynamic based on factors such as UE’s speed or scaling requirements. If the threshold is dynamic, the UE 504 may include the specific threshold value in its HOF or RLF report to the network. Based on the reported threshold, the network can evaluate whether the HOF or RLF might have been due to an inappropriate or overly sensitive threshold setting and can adjust accordingly for future scenarios. Upon receiving the RLF or HOF report, the network 51 DM2\20713381.2F9125-62400 (e.g. the serving cell 502-1) may respond by instructing the UE 504 to use an alternate Model ID. This might involve switching to a different version of the ML model, for example one that is better suited for the UE’s mobility conditions or the specific cell environment. The network can also use the information of the RLF or HOF report to fine-tune the existing ML model or select a model optimized for the conditions that led to the RLF or HOF. This adaptability allows the network to continuously refine its prediction accuracy and improve connectivity.

[0119] In some embodiments, the signaling diagram illustrated in FIG. 5 offers a proactive HOF or RLF recovery approach by providing the UE 504 with necessary candidate beams and configuration data in advance of potential failures. On the other hand, traditional methods may rely on reactive responses after a failure occurs, requiring the UE to initiate a full reestablishment process, leading to service interruptions and increased latency. Therefore, the approach shown in the signaling diagram in FIG. 5 can reduce recovery time and ensures seamless connectivity.

[0120] FIG. 6 illustrates a deep neural network (DNN) model 600 used to implement the ML models 206, 306, 316 or 406, in accordance with some embodiments of the present disclosure. Although a DNN example is illustrated in FIG. 6, the ML model in the present disclosure is not limited to DNN implementation, and can take any other forms of ML model, such as multilayer perceptron, feedforward neural networks, convolutional neural networks, recurrent neural networks, autoencoder, generative adversarial networks, and long short-term memory. In some embodiments, in the ML models 206, 306, 316 or 406, the layer 410-koutput can be arranged in the vector {(|) = [w^, … , wC~] with layer ^ consisting of u^neurons and the input layer corresponding to ^ = 0. The relationship between the outputs oflayers ^ − 1 and ^ can be represented as w(^) = F(w(^ − 1)^^), where F(. ) is a non-linear52 DM2\20713381.2F9125-62400 activation function applied elementwise to the input argument and ^^is a weight matrix that connects layers 610-(k) and 610-(k-1).

[0121] In some embodiments, the DNN model 600 is trained using a plurality of training samples, wherein each training sample comprises an input vector and a corresponding output vector. In one embodiment, to find the optimal values of the weight matrices ^^to ^|^^during the training of the DNN model 600, a back propagation algorithm is used by taking an error rate of a forward propagation and feeding this loss backward through the layers of the DNN model 600 to fine-tune the weights. In another embodiment, to find the optimal values of the weight matrices ^^to ^|^^during the training of the ANN model 600, a weight perturbation technique can be used. The weight perturbation technique may be applied in an iterative manner for a plurality of iterations, wherein in each of the plurality of iterations, a weight variation of random sign is added to each of the elements in the weight matrices ^^to ^|^^and a corresponding training error is observed. If the training error is increased in a given iteration, then the elements in the weight matrices ^^to ^|^^will be changed to the opposite directions of the weight variations; if the training error is decreased in a given iteration, then the elements in the weight matrices ^^to ^|^^will be changed to the same directions of the weight variations. This iterative training can be stopped if at least one of the following conditions is met: the training error becomes smaller than a predetermined error threshold value, a maximum number of iterations is reached, and the training error does not decrease for a predetermined number of iterations. In some embodiments, a dynamic weight perturbation technique can be applied to train the DNN model 600 by decreasing the amount of weight variations in each iteration, such that the DNN model 600 is fine-tuned towards the end of the training process. In one embodiment, the weight variation in the ^-th iteration ^ican be calculated as: ^i = ^, / (^^), where ^, is an initial weight variation amount, and ^ is auser-defined parameter which controls the decrease rate of ^i. 53 DM2\20713381.2F9125-62400

[0122] FIG. 7 illustrates an example method 700 for machine learning-based radio link failure and handover predictions, in accordance with some embodiments. The operations of method 700 presented below are intended to be illustrative. In some embodiments, method 700 may be accomplished with one or more additional operations not described and / or without one or more of the operations discussed. Additionally, the order in which the operations of method 700 are illustrated in FIG. 7 and described below is not intended to be limiting.

[0123] At step 702, a network (e.g. a serving BS) may transmit a first signal to a UE, wherein the first signal may configure a set of beams from the serving cell (i.e. the serving BS) and a set of SSBs from a plurality of neighboring cells to be measured by the UE for RLF or HO predictions. In some embodiments, the first signal comprises a pre-RRC configuration message for the UE, wherein the pre-RRC configuration message comprises CSI-RS beams from the serving BS and SSBs from the plurality of neighboring cells. In some embodiments, the UE is configured to measure the set of beams from the serving cell and the set of SSBs from the plurality of neighboring cells, and evaluate the quality of potential handover candidates. In some other embodiments, the pre-RRC configuration message may comprise full configurations for the serving and / or neighboring cells, similar to an RRC reconfiguration message, but without conditional reconfiguration details for conditional handover or Layer 1 / Layer 2 triggered mobility. In some embodiments, step 702 may be performed when the UE predicts an RLF or an HOF and reports this prediction to the serving BS. The pre-RRC configuration message may also indicate a list of candidate beams to the UE to ensure a smooth transition to a target cell if needed.

[0124] At step 704, the serving BS may update the set of beams and the set of SSBs to be measured by the UE. In some embodiments, the serving BS updates a subset of beams and 54 DM2\20713381.2F9125-62400 SSBs based on previous measurement reports received from the UE. In some embodiments, as the UE transmits periodic reports on the measured signal strengths (e.g., RSRP, RSRQ) for its current subset of beams and SSBs, the serving BS may analyze these reports to determine the quality trends of the monitored signals. Based on the analysis, the serving BS may adjust the UE’s measurement configuration to focus on a new subset of beams and SSBs. This adjustment ensures that the UE measures signals that are most likely to provide relevant information for upcoming RLF or HO predictions. In some embodiments, the set of beams and SSBs is updated using a moving window that shifts across all possible beams / SSBs. This method provides the UE with a focused subset of beams / SSBs in each measurement cycle, which is updated as the window moves. This moving window approach ensures comprehensive coverage over time, as the UE will eventually measure all beams. However, in any given cycle, the UE only focuses on a specific subset of beams / SSBs, reducing measurement overhead while still gathering necessary data. In some embodiments, to reduce the signaling overhead, the UE may autonomously apply the “moving window” technique to report the subset of beams / SSBs to the network. In the “moving window” technique, the UE may be configured to adjust the measurements of beams / SSBs (e.g. updating one subset of beams to another subset of beams) based on its location, speed, and / or network configuration.

[0125] At step 706, the UE may use an ML model implemented in the UE to predict an upcoming RLF or HO event and report this prediction to the network (e.g. the serving BS). In some embodiments, the prediction of the RLF or HO may be based on an RRC reconfiguration message that the UE receives from the serving BS, wherein the RRC reconfiguration message comprises the necessary parameters and configuration for RLF or HO predictions. 55 DM2\20713381.2F9125-62400

[0126] In some embodiments, the ML model implemented in the UE may determine whether the beam measurements meet the conditions for any of the configured event types (e.g., A2, A3, A5). Once the relevant measurements exceed the configured thresholds and remain stable for the duration of the Time-to-Trigger (TTT) timer, the ML model can generate an RLF or HO prediction report. This report may indicate the event type, such as a predicted Event A3 (indicating that a neighboring cell’s signal is stronger than the serving cell by a configured offset), along with the associated probability and timing. By mapping the RLF or HO prediction to the corresponding measurement event, the network can make informed decisions on whether to proceed with the handover.

[0127] At step 708, the UE may transmit an RLF or HO prediction report to the network, wherein the report comprises an indication to indicate the probability of RLF or HO completion (e.g. 70%, 80%, 90%, etc.) and a subsequent time when the RLF or HO occurs (e.g., 10 ms later, 10 time-slot units later, or within a subsequent time period). The probability of HO completion may be indicated for a specific time period of the handover execution for the target neighbor cell. Additionally, the actual measured and the predicted RSRP / RSRQ / SINR values for the considered beams / SSBs may be included in the report in order for the network to make decisions. In some embodiments, the subsequent time in the indication of the HO prediction report is an elapsed time between generating the indication and a start of a handover timer. In some other embodiments, the indication further comprises a time period indicating a validity of a handover execution period. In yet some other embodiments, the indication further comprises a cell identification (ID) of at least one neighbor cell for the handover execution.

[0128] In some embodiments, upon receiving the RLF prediction report, the network (e.g. the serving BS) can respond by transmitting a second signal comprising a Reconnect message 56 DM2\20713381.2F9125-62400 to the UE. In some embodiments, the Reconnect message comprises at least one of: reconnection instructions (e.g. RACH procedure), candidate beams and cells, and action timings on when to initiate the specified actions. Upon receiving the reconnect message, the UE may follow the instructions outlined in the message to initiate the reconnection procedure, monitor the specified beams or cells for a suitable signal to re-establish communication, and / or use the RACH procedure or other specified reconnection methods to regain access to the network.

[0129] In some other embodiments, when an HO prediction is made and an HOF occurs, the UE may be unable to successfully complete the HO to a target cell. In such a case, if the UE is still connected to the serving cell despite the failed HO attempt, the UE can report the HOF directly to the serving cell. On the other hand, if the UE’s connection to the serving cell was also lost, the UE can initiate a connection reestablishment procedure to restore communication with the network. After reestablishing the connection, the UE may then report the HOF event to the network. Similar to the RLF reporting, the UE may also provide contextual information such as speed, location, and the time of the HOF event, which helps the network understand the mobility and environmental conditions during the HOF.

[0130] While various embodiments of the present disclosure have been described above, it should be understood that they have been presented by way of example only, and not by way of limitation. Likewise, the various diagrams may depict an example architectural or configuration, which are provided to enable persons of ordinary skill in the art to understand exemplary features and functions of the present disclosure. Such persons would understand, however, that the present disclosure is not restricted to the illustrated example architectures or configurations, but can be implemented using a variety of alternative architectures and configurations. Additionally, as would be understood by persons of ordinary skill in the art, 57 DM2\20713381.2F9125-62400 one or more features of one embodiment can be combined with one or more features of another embodiment described herein. Thus, the breadth and scope of the present disclosure should not be limited by any of the above-described exemplary embodiments.

[0131] It is also understood that any reference to an element herein using a designation such as "first," "second," and so forth does not generally limit the quantity or order of those elements. Rather, these designations can be used herein as a convenient means of distinguishing between two or more elements or instances of an element. Thus, a reference to first and second elements does not mean that only two elements can be employed, or that the first element must precede the second element in some manner.

[0132] Additionally, a person having ordinary skill in the art would understand that information and signals can be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits and symbols, for example, which may be referenced in the above description can be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.

[0133] A person of ordinary skill in the art would further appreciate that any of the various illustrative logical blocks, modules, processors, means, circuits, methods and functions described in connection with the aspects disclosed herein can be implemented by electronic hardware (e.g., a digital implementation, an analog implementation, or a combination of the two), firmware, various forms of program or design code incorporating instructions (which can be referred to herein, for convenience, as "software" or a "software module), or any combination of these techniques.

[0134] To clearly illustrate this interchangeability of hardware, firmware and software, various illustrative components, blocks, modules, circuits, and steps have been described 58 DM2\20713381.2F9125-62400 above generally in terms of their functionality. Whether such functionality is implemented as hardware, firmware or software, or a combination of these techniques, depends upon the particular application and design constraints imposed on the overall system. Skilled artisans can implement the described functionality in various ways for each particular application, but such implementation decisions do not cause a departure from the scope of the present disclosure. In accordance with various embodiments, a processor, device, component, circuit, structure, machine, module, etc. can be configured to perform one or more of the functions described herein. The term “configured to” or “configured for” as used herein with respect to a specified operation or function refers to a processor, device, component, circuit, structure, machine, module, etc. that is physically constructed, programmed and / or arranged to perform the specified operation or function.

[0135] Furthermore, a person of ordinary skill in the art would understand that various illustrative logical blocks, modules, devices, components and functions described herein can be implemented within or performed by one or more circuits or circuitry. As used herein, the term “circuitry” refers to and includes any one or more of the following: discrete circuit components or devices coupled to each other to form circuit, logic circuitry, integrated circuits, application specific integrated circuits, state machines, general purpose processors, special purpose processors, digital signal processors (DSP), microprocessors, field programmable gate arrays (FPGA) or other programmable logic devices, or any combination thereof. Circuitry can further include antennas, reflectors, transmitters, receivers and / or transceivers to communicate with various components, devices or nodes within a communication network. As used herein, the term “processor” refers to a combination of structures including processing circuitry, a memory coupled to the processing circuitry, and executable code stored in the memory that when executed by the processing circuitry perform the functions or operations instructed by the executable code. 59 DM2\20713381.2F9125-62400

[0136] If implemented in software, the functions can be stored as one or more instructions or code on a computer-readable medium. Thus, the steps of a method or algorithm disclosed herein can be implemented as software stored on a computer-readable medium. Computer- readable media includes both computer storage media and communication media including any medium that can be enabled to transfer a computer program or code from one place to another. A storage media can be any available media that can be accessed by a computer. By way of example, and not limitation, such non-transitory computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer.

[0137] In this document, the term "module" as used herein, refers to software, firmware, hardware, and any combination of these elements for performing the associated functions described herein. Additionally, for purpose of discussion, the various modules are described as discrete modules; however, as would be apparent to one of ordinary skill in the art, two or more modules may be combined to form a single module that performs the associated functions according embodiments of the present disclosure.

[0138] Additionally, memory or other storage, as well as communication components, may be employed in embodiments of the present disclosure. It will be appreciated that, for clarity purposes, the above description has described embodiments of the present disclosure with reference to different functional units and processors. However, it will be apparent that any suitable distribution of functionality between different functional units, processing logic elements or domains may be used without detracting from the present disclosure. For example, functionality illustrated to be performed by separate processing logic elements, or 60 DM2\20713381.2F9125-62400 controllers, may be performed by the same processing logic element, or controller. Hence, references to specific functional units are only references to a suitable means for providing the described functionality, rather than indicative of a strict logical or physical structure or organization.

[0139] Various modifications to the implementations described in this disclosure will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other implementations without departing from the scope of this disclosure. Thus, the disclosure is not intended to be limited to the implementations shown herein, but is to be accorded the widest scope consistent with the novel features and principles disclosed herein, as recited in the claims below. 61 DM2\20713381.2

Claims

F9125-62400 CLAIMS What is claimed is:

1. A method comprising: receiving, by a wireless communication device, a first plurality of beams from a first plurality of wireless communication nodes, wherein the first plurality of wireless communication nodes comprises a first wireless communication node and a second plurality of wireless communication nodes; transmitting, by the wireless communication device, a report to the first wireless communication node, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

2. The method of claim 1, wherein the first plurality of beams comprises: a plurality of channel state information reference signal (CSI-RS) beams from the first wireless communication node; and a plurality of synchronization signal block (SSB) beams from the second plurality of wireless communication nodes.

3. The method of claim 1, further comprising: prior to receiving the first plurality of beams, receiving, by the wireless communication device, a pre-radio resource control (RRC) configuration message from the first wireless communication node, wherein the pre-RRC configuration message configures the first plurality of beams to be measured by the wireless communication device for RLF predictions. 62 DM2\20713381.2F9125-62400 4. The method of claim 1, wherein the report is generated based on the first plurality of measurements and a second plurality of measurements taken for a second plurality of beams, wherein the second plurality of beams is generated based on the first plurality of beams using a moving window technique.

5. The method of claim 4, wherein the second plurality of beams is generated based on at least one of: a location of the wireless communication device; a speed of the wireless communication device; and a network configuration.

6. The method of claim 1, wherein the first plurality of measurements comprises at least one of: a Reference Signal Received Power (RSRP); a Reference Signal Received Quality (RSRQ); a Signal-to-Interference-plus-Noise Ratio (SINR); a Channel Quality Indicator (CQI); a Beam Reference Signal Strength (BRSS); and an Angle of Arrival (AoA).

7. The method of claim 1, wherein the subsequent time period comprises a first number of time-slot units or a second number of milliseconds.

8. A wireless communication device comprising: a transceiver configured to: 63 DM2\20713381.2F9125-62400 receive a first plurality of beams from a first plurality of wireless communication nodes, wherein the first plurality of wireless communication nodes comprises a first wireless communication node and a second plurality of wireless communication nodes, transmit a report to the first wireless communication node, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

9. A non-transitory computer readable medium storing computer-executable instructions which when executed perform a method comprising: receiving, by a wireless communication device, a first plurality of beams from a first plurality of wireless communication nodes, wherein the first plurality of wireless communication nodes comprises a first wireless communication node and a second plurality of wireless communication nodes; transmitting, by the wireless communication device, a report to the first wireless communication node, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

10. Circuitry configured to perform a method, the method comprising: receiving, by a wireless communication device, a first plurality of beams from a first plurality of wireless communication nodes, wherein the first plurality of wireless communication nodes comprises a first wireless communication node and a second plurality of wireless communication nodes; 64 DM2\20713381.2F9125-62400 transmitting, by the wireless communication device, a report to the first wireless communication node, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

11. A method comprising: transmitting, by a first plurality of wireless communication nodes, a first plurality of beams to a wireless communication device, wherein the first plurality of wireless communication nodes comprises a first wireless communication node and a second plurality of wireless communication nodes; receiving, by the first wireless communication node, a report from the wireless communication device, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

12. The method of claim 11, wherein the first plurality of beams comprises: a plurality of channel state information reference signal (CSI-RS) beams from the first wireless communication node; and a plurality of synchronization signal block (SSB) beams from the second plurality of wireless communication nodes.

13. The method of claim 11, further comprising: prior to transmitting the first plurality of beams, transmitting, by the first wireless communication node, a pre-radio resource control (RRC) configuration message to the 65 DM2\20713381.2F9125-62400 wireless communication device, wherein the pre-RRC configuration message configures the first plurality of beams to be measured by the wireless communication device for RLF predictions or handover predictions.

14. The method of claim 11, wherein the report is generated based on the first plurality of measurements and a second plurality of measurements taken for a second plurality of beams, wherein the second plurality of beams is generated based on the first plurality of beams using a moving window technique.

15. The method of claim 14, wherein the second plurality of beams is generated based on at least one of: a location of the wireless communication device; a speed of the wireless communication device; and a network configuration.

16. The method of claim 11, wherein the first plurality of measurements comprises at least one of: a Reference Signal Received Power (RSRP); a Reference Signal Received Quality (RSRQ); a Signal-to-Interference-plus-Noise Ratio (SINR); a Channel Quality Indicator (CQI); a Beam Reference Signal Strength (BRSS); and an Angle of Arrival (AoA).

17. A first of wireless communication node comprising: a transceiver configured to: 66 DM2\20713381.2F9125-62400 transmit a plurality of channel state information reference signal (CSI-RS) beams in a first plurality of beams to a wireless communication device, wherein the first plurality of beams is transmitted to the wireless communication device, and wherein the first plurality of beams further comprises a plurality of synchronization signal block (SSB) beams transmitted from a first plurality of wireless communication nodes; receive a report from the wireless communication device, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

18. A non-transitory computer readable medium storing computer-executable instructions which when executed perform a method comprising: transmitting, by a first plurality of wireless communication nodes, a first plurality of beams to a wireless communication device, wherein the first plurality of wireless communication nodes comprises a first wireless communication node and a second plurality of wireless communication nodes; receiving, by the first wireless communication node, a report from the wireless communication device, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

19. Circuitry configured to perform a method, the method comprising: transmitting, by a first plurality of wireless communication nodes, a first plurality of beams to a wireless communication device, wherein the first plurality of wireless 67 DM2\20713381.2F9125-62400 communication nodes comprises a first wireless communication node and a second plurality of wireless communication nodes; receiving, by the first wireless communication node, a report from the wireless communication device, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period.

20. A communication system comprising a wireless communication device and a first wireless communication node, wherein: the wireless communication device comprises a first transceiver configured to receive a first plurality of beams from a first plurality of wireless communication nodes, wherein the first plurality of wireless communication nodes comprises the first wireless communication node and a second plurality of wireless communication nodes; the first wireless communication node comprises a second transceiver configured to receive a report from the wireless communication device, wherein the report is generated based on a first plurality of measurements taken for the first plurality of beams, and wherein the report comprises: an indication indicating that a radio link failure (RLF) occurs within a subsequent time period. 68 DM2\20713381.2

Images