Methods and devices enabling user-provided machine learning-based network configuration recommendations

Abstract

Methods and devices in a wireless network enable a user equipment (UE) to provide machine learning-based network configuration recommendations. In a UE wireless communication method (300), the UE (102) generates (305) channel quality predictions based on channel quality measurements using a machine learning module, and then transmits (306), to a network node (104), a recommendation to update the network configuration, the recommendation being based on the channel quality predictions.

Description

METHODS AND DEVICES ENABLING USER-PROVIDEDMACHINE LEARNING-BASED NETWORK CONFIGURATION RECOMMENDATIONSFIELD OF THE DISCLOSURE

[0001] This document generally describes methods and devices operating in wireless communication systems such as (but not limited to) the ones described in 5G standard documents, known as third generation partnership (3GPP) communication systems. More particularly, this document refers to systems in which a wireless network user (a.k.a. user equipment (UE)) provides a network configuration recommendation based on machine learning (ML) channel quality predictions.BACKGROUND

[0002] Developing ML-based algorithms to predict wireless channel metrics (thus, implicitly channel quality) is an ongoing worldwide project as described, for example, in 3GPP Technical Report (TR) 38.843 (entitled “Study on Artificial Intelligence (Al) / Machine Learning (ML) for NR air interface”). The predicted metrics, which use current and, optionally, historical channel measurements, may be channel coefficients, channel quality indicators (e.g., reference signal received power (RSRP), received signal strength indicator (RSSI), Signal-to-lnterference-plus-Noise Ratio (SINR)), and beam related metrics (e.g., a beam index) with a particular focus for Frequency Range 2 (FR2) (i.e., approximately between 24 and 71 GHz). The predicted channel metrics may correspond to a channel in a future time (i.e., time domain predictions), a not-currently- measured channel at present (i.e., spatial domain prediction), or a combination of time and space domain predictions. One example of spatial domain prediction is when the UE measures the RSRP in beamforming direction A to predict the RSRP in another beamforming direction B.

[0003] A UE running ML algorithms includes an ML prediction module, that is, software and / or hardware support for an ML function. The ML prediction module may be software provided by the network (NW) vendor, a UE vendor, or a third party serviceprovider. The input of the ML prediction module includes UE measurements of reference signals (RSs) transmitted from NW nodes (NNs, that is, a device performing a network function, such as, a base station). The input may be a multi-dimensional vector with each dimension corresponding to an RS. The input may include one or more time sequences of the measurements, where different elements in the same time sequence correspond to different measurement times. The UE may report the predictions to the NW, or the UE may use the predictions locally (e.g., to identify a beam for receiving DL transmissions).

[0004] Another ongoing goal of worldwide research is saving power at the NW (infrastructure) side. At times (e.g., when the user load is relatively low), the NW may use power saving node techniques (e.g. turning off certain NW nodes such as a base station or a repeater) or refraining from using certain beams (and stop sending reference signals such as synchronization signal blocks (SSBs) for these beams) that serve few UEs or whose service can be replaced by another node, by using a smaller array / lower Tx power, etc. Such power saving techniques may be triggered when the number of served UEs is low, and the UE service requirements are not rigid. The condition to apply a power saving NW configuration (NWC) depends not only on the associated load, but also on the quality of service observed by the currently served UEs. In other words, the NW power saving techniques should apply only to the extent of satisfying certain quality of service (QoS) requirements. Additionally, reversing a NWC change may be problematic. Even after a served UE reports certain QoS issues, switching between different NWCs (e.g., turning on certain SSBs / nodes) requires a certain amount of time during which served UEs will experience inadequate NW service.SUMMARY

[0005] In various embodiments, a UE configured to predict non-measured channel metrics using ML (e.g., using an ML module that may be configured by the network), sends, to the NN, an NWC recommendation (e.g., to turn on / off certain cells or NN components, to adjust transmission power, to switch beams, to change discontinuous transmission and reception cycle parameters, etc.). Besides the predictions regarding traffic and channel quality, the NWC recommendation may also be based on additionallocal information (e.g., other UE predicted channel quality metrics, UE applications running, UE mobility, UE power level, UE processing capabilities, UE assistance-type information, and / or UE location information) that the UE does not contemporaneously share with the network. The NWC recommendation may be explicit (e.g., indicating the manner of changing a parameter) or implicit indicating a specific NWC when associated conditions are met. The UE may receive from the NN a list of acceptable NWC recommendations and / or respective conditions associated with the NWC recommendations. In various aspects, the UE may send the NWC recommendation using a random access (RA) procedure (i.e. , on a physical random access channel (PRACH)), a medium access control (MAC) control element (CE), or a preconfigured report as uplink control information (UCI) (i.e., on a physical uplink control channel (PUCCH) or physical uplink shared channel (PUSCH)). In view of the NWC recommendation, the NN may explicitly indicate adopting the recommended NWC. Alternatively, the UE may assume that the recommended NWC is applied after a predetermined time interval (having a duration that may depend on the specific NWC recommendation) after the recommendation transmission. The UE and / or the NN may measure channel qualities after applying the recommended NWC.

[0006] The UE providing a NWC recommendation based on ML channel quality predictions reduces the overhead and delay for making the analysis and taking a decision at the network side, meanwhile promoting energy savings and traffic improvement.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate one or more embodiments and, together with the description, explain these embodiments.

[0008] Fig. 1 is a block diagram of a wireless communication system including a UE and an NN able to perform methods according to various embodiments.

[0009] Fig. 2 is a signal diagram illustrating a network node configuration change triggered by a UE recommendation based on an ML output predicting channel metrics.

[0010] Fig. 3 is a flowchart of a UE method for recommending a network configuration (NWC) change based on ML output according to an embodiment.

[0011] Fig. 4 is a flowchart of an NN method for changing configuration upon receiving an ML-based recommendation from a UE according to an embodiment.

[0012] Fig. 5 is a dataflow in a UE outputting a NWC recommendation identifier according to an embodiment.

[0013] Fig. 6A is a timeline of a UE triggering an NWC update based on ML channel quality predictions according to an embodiment.

[0014] Fig. 6B is a timeline of a UE triggering an NWC update based on ML channel quality predictions according to another embodiment.DETAILED DESCRIPTION

[0015] Methods and devices described in this section embody techniques related to network configuration recommendations provided by UEs based on ML channel quality predictions. The embodiment descriptions in this section refer to the accompanying drawings. The same reference numbers in different drawings identify the same or similar elements. The detailed descriptions do not preclude other embodiments within the scope of the appended claims. The embodiments are not limited to the described configurations but may be extended to other arrangements.

[0016] Conventionally, a UE that predicts channel metrics in the time domain and / or the spatial domain using ML-based techniques reports the ML predictions to the NN. In view of these UE predictions, the NN may then adjust its configuration to save power, optimize traffic, etc. This conventional approach leads to delay in actually implementing such an adjusted configuration in a timely manner.

[0017] According to various embodiments described in this section, a UE that predicts channel metrics in the time domain and / or the spatial domain using ML-based techniques outputs an NWC recommendation based on these predictions. The UE’s NWC recommendation alerts the NW that the recommended NWC seems beneficial thereby reducing the overhead and delay relative to conventional NW-side analysis and decision, meanwhile promoting energy savings and traffic improvement. Subject toNW’s ultimate decision, the UE’s NWC recommendation may trigger configuration changes at both NW and UE sides.

[0018] Before describing in detail aspects related to the UE-provided NWC recommendation, FIG. 1 illustrates a wireless communication system 100 configured to perform such methods. The wireless communication system 100 includes a UE 102, base stations (BSs) 104 and 106 and a core network (CN) 110. The UE 102 connects to the BS 104 and may operate in dual connectivity (DC) with the BSs 104 and 106. The UE 102 may communicate with the BSs using the same radio access technologies (RATs), such as, EUTRA or NR, or different RATs.

[0019] The BS 104 is equipped with processing hardware 130 that includes one or more general-purpose processors (e.g., CPUs) 132 and a non-transitory computer- readable memory 138 storing instructions that the one or more general-purpose processors execute. Additionally, or alternatively, the processing hardware 130 may include special-purpose processing units. The processing hardware 130 includes a transmitter 134 and a receiver 136 (which may be called collectively “transceiver”). The transmitter 134 is configured to transmit data and control signals on physical downlink (DL) channels and DL reference signals with one or more user devices (e.g., UE 102) via one or more cells (e.g., the cell(s) 124 and / or 125). The receiver 136 is configured to receive data and control signals on physical uplink (UL) channels and / or UL reference signals with the one or more user devices via one or more cells. The second BS 106 includes similar processing hardware (not illustrated).

[0020] The CN 110 may be an evolved packet core (EPC) 111 or a fifth-generation core (5GC) 160 (both cores being illustrated in Fig. 1 ) and may be hosted on one or more physical devices (i.e., hardware, not illustrated in Fig. 1 ) with processors, wireless communication hardware (e.g., transceivers), and memories storing executable codes. The BS 104 may be an eNB supporting an S1 interface for communicating with the EPC 111 , an ng-eNB supporting an NG interface for communicating with the 5GC 160, or a gNB that supports an NR radio interface as well as an NG interface for communicating with the 5GC 160. To directly exchange messages with each other during the embodiments discussed below, the BSs 104 and 106 may support an X2 or Xn interface.

[0021] Among other components (not illustrated), the EPC 111 includes a Mobility Management Entity (MME) 112, a Serving Gateway (SGW) 114, and a Packet Data Network Gateway (PGW) 116. The MME 112 manages authentication, registration, paging, and other related functions. The SGW 114 transfers user-plane packets related to audio calls, video calls, Internet traffic, etc. The PGW 116 provides connectivity from the UE to one or more external packet data networks (e.g., an Internet network and / or an Internet Protocol (IP) Multimedia Subsystem (IMS) network).

[0022] Among other functions (not shown), the 5GC 160 provides an Access and Mobility Management (AMF) 162, a Session Management Function (SMF) 164, and a User Plane Function (UPF) 166. The AMF 162 manages authentication, registration, paging, and other related functions. The SMF 164 manages packet data unit (PDU) sessions. The UPF 166 transfers user-plane packets related to audio calls, video calls, Internet traffic, etc.

[0023] As also illustrated in FIG. 1 , the BS 104 supports cells 124 and 125, while the BS 106 supports cell 126. The cells 124 and 125 partially overlap, so that the UE 102 may communicate using carrier aggregation (CA) with the first BS 104. These cells also overlap with the cell 126 so the UE 102 may communicate in DC with the BS 104 and the BS 106 (where one of the BSs 104 and 106 is a master node (MN) and the other is a secondary node (SN)). The BSs 104 and 106 may support additional cell(s). The wireless communication network 100 may include any suitable number of BSs supporting NR cells and / or EUTRA cells. More particularly, the EPC 111 or the 5GC 160 may be connected to any suitable number of BSs supporting NR cells and / or EUTRA cells. Although the examples below refer specifically to specific CN types (EPC, 5GC) and RAT types (5G NR and EUTRA), the techniques of this disclosure also may apply to other suitable radio access and / or core network technologies such as sixth generation (6G) radio access and / or 6G core network or 5G NR-6G DC.

[0024] The UE 102 is equipped with processing hardware 140 that includes one or more (general-purpose and / or special-purpose) processors 142 such as CPUs and a non-transitory computer-readable memory 148 storing machine-readable instructions executable on the one or more processors. The transmitter 144 and the receiver 146 (which may be called collectively “transceiver”) enable the UE to receive data andcontrol signals on physical DL channels and / or DL reference signals with the BSs 104 and / or 106 via one or more cells and to transmit data and control signals on physical UL channels and / or UL reference signals with the BSs 104 or 106 via one or more cells.

[0025] Focusing now on the techniques related to UE providing a NWC recommendation, Fig. 2 is a signal diagram illustrating a NWC change triggered by a UE recommendation based on ML-predicted channel metrics at the UE. The UE 102 may first indicate 202, to an NN (such as, a BS) 104, its ability to provide NWC recommendations. That is, the UE sends UE capability information for providing NWC recommendations. This step is optional (as suggested by using dashed line in Fig. 2) and may occur during the UE’s registration procedure. Alternatively, the network may retrieve an indication about the UE’s ability to provide NWC recommendations from the user’s subscription. Also optional, the NN 104 may then provide 204 an ML configuration to the UE 102. The ML configuration directs the UE to predict nonmeasured channel metrics in time and / or space domains (i.e., corresponding to either a future time or to another spatial direction than measured directions).

[0026] The NN 104 may also send 206, to the UE 102, a message indicating one or more allowed NWC recommendations (e.g., valid NWC candidates for the NN). For example, the NW thus configures multiple NWC recommendations which correspond to valid NWCs, and the UE may only provide a NWC recommendation among these multiple NWC recommendations. The NW may send the multiple recommendation candidates by indicating pre-defined mapping rules, a range of NWC parameters available for the UE to select from, or a list of NWC parameters available for the UE to select from. Additionally and also optionally, the NN 104 informs 208 the UE 102 about triggering conditions for the UE to provide an NWC recommendation. The triggering conditions may be associated with the NWC candidates, respectively.

[0027] The NN 104 then sends 210 an NWC recommendation based on ML predictions output by an ML module at the UE that may have been configured according to the network-provided ML configuration. Here, the ML module refers to hardware and / or software that provides channel metric predictions based on current measurements at the UE and optionally, historical channel quality-related data. The MLmodule may be a separate component of the UE or embedded in the components illustrated in Fig. 1 .

[0028] The NWC recommendation may include: (i) to turn on / off certain NW nodes or reference signals (e.g., SSBs); (ii) to increase or decrease transmission power; (iii) to change antenna elements and / or codebooks; and (iv) to adjust cell discontinuous transmission / reception configuration (e.g., the duration of ON time, the duty cycle duration, the duration of inactivity timer), etc. The UE may provide such NWC recommendations explicitly or merely indicate one or more configuration identifiers or parameters. The UE may indicate a certain configuration identifier or parameter when one or more corresponding trigger conditions are met.

[0029] The NWC recommendation may be based also on local information such as predicted channel quality metrics not concurrently communicated to the network, UE’s predicted traffic (e.g., based on the foreseeable data volume of applications the UE is running), UE power level, UE processing capabilities, UE assistance-type information, and / or metrics based on UE’s mobility and location information.

[0030] The NWC recommendation may be explicit (e.g., indicating parameter values such as transmission power), implicit (e.g., based on a signaling such as 204 and 206 and / or a pre-defined mapping rule) or a combination of explicit and implicit. The predefined mapping rule may be reported by the UE to the NW, configured by the NW at the UE, or predefined in standards (and stored by both the UE and the NW). The signaling may include a UE report associated with the ML module. For example, the NW may preconfigure multiple NWC recommendations (e.g., as later described based on Table 1 ) at the UE (see also message 206), and the UE provides a recommendation ID based on the output of ML module at the UE. In another example, the UE reports a predicted RSRP, and then, based on the pre-defined mapping rule, the NN converts the reported RSRP value to different NWCs (e.g., the rule instructs that if RSRP>X dB, switching the current transmission configuration indicator (TCI) state to another TCI state).

[0031] As already mentioned, in some embodiments, the UE sends an NWC recommendation when one or more corresponding conditions are met. For example, the UE is configured to predict RSRP of a cell / beam (for a currently inactive cell); whenthe predicted RSRP is X dB better than RSRP of the currently serving cell / beam, the UE sends an NWC recommendation to turn on the currently inactive cell. The threshold X may be configured by the NW.

[0032] Table 1 below exemplary illustrates conditions for triggering different NWC recommendations (NWCR) according to an embodiment.

[0033] Relative to the UE’s behavior when NWCR I D= 1 , the handover may be triggered by Layer 3 signaling or by lower layer (1 or 2) signaling (i.e. , a lower layer triggered mobility (LTM) procedure). The handover may use predetermined resources (e.g., a predefined random access channel (RACH) resource).

[0034] Relative to the UE’s behavior when NWCR I D=3, optionally the UE may use an updated transmission power (as recommended) to compute pathloss used to determine UL Tx power of beam 1 in the closed loop UL power control. The Tx power adjustment may be indicated to UE via explicit signaling or by indicating a switch of current TCI, where different TCIs are preconfigured to be associated with different Tx powers or NW arrays.

[0035] The UE may send 210 the NWC recommendation using a reserved PRACH resource or a reserved preamble sequence. The reserved PRACH resource may be common for multiple recommendations. The UE may explicitly indicate the NWC recommendation via a subsequent message (e.g., a Msg 3 of a four-step random access (RA) procedure with the PRACH preamble as Msg 1 ). Alternatively, different PRACH resources / preambles may be reserved for different NWC recommendations.

[0036] In some embodiments, the NWC recommendation is sent using a MAC CE. In one embodiment, step 210 includes two steps: during step 1 , the UE sends a scheduling request (e.g., on a reserved scheduling request resource) asking for physical uplink shared channel (PUSCH) resources for submitting the NWC recommendation, and then, during step 2, the NW responds with an UL grant, and the UE sends the NWC recommendation using the UL grant-specified resources.

[0037] In other embodiments, the UE sends the NWC recommendation in a preconfigured report, which may be periodic. The periodic report may be configured for the UE to send its prediction(s) periodically (e.g., as UCI). When the reported predictions meet one or more predefined conditions, the report triggers a NWC change known to both the UE and the NN. In this embodiment, the NWC recommendation is implicit, the UE and NN having the same understanding as to whether the reported prediction(s) satisfies the condition triggering the NWC change. The condition triggering the NWC change may be signaled to the UE by the NW or predefined in technical specifications.

[0038] Returning now to Fig. 2, the NN 104 may send 212, to the UE 102, a response to the NWC recommendation. In some embodiments, after sending the recommendation, the UE receives a message indicating whether (“Yes” or “No”) the NN has adopted the NWC recommendation (i.e. , the NWC recommendation has triggered the recommended change). In other embodiments, absent receiving, during a predetermined time interval, a message indicating that the NN has declined to adopt the recommended change, the UE assumes that the recommended change was adopted. Finally, both the NN 104 and the UE 102 may then update their configuration. That is, the NN 104 updates 214 the NWC based on the NWC recommendation, and the UE 102 updates 215 its UE configuration according to the NWC recommendation. The UE 102 and the NN 104 may then exchange 216 communications based on the updated configurations.

[0039] Fig. 3 is a flowchart of a UE method 300 for providing an NWC recommendation based on ML output according to an embodiment. The method 300 includes optionally sending 302, to the NW, UE capability information on providing NWC recommendations. Also optionally, the method 300 includes obtaining 304 an ML configuration for an ML module (e.g., receiving the ML configuration from the NN or another network device, retrieving the ML configuration, etc.). The method 300 further includes the UE generating 305 channel quality predictions (i.e., metrics as discussed above) in time and / or space domains, using the ML module (as optionally configured). Before or after step 305, the method 300 includes receiving 306 allowable NWC candidates and receiving 308 triggering conditions for providing an NWC recommendation. In one embodiment, triggering conditions are specifically associated with NWC candidates.

[0040] The method 300 then includes the UE indicating 310 an NWC recommendation based on the ML predictions. Here, the UE may transmit NWC parameters or a NWC identifier (RID as illustrated in Fig. 5). The method 300 may also include receiving 312 a response to the indicated NWC recommendation, updating 315 the UE configuration according to the NWC recommendation after a predetermined timeinterval and / or in view of the response, and communicating 316 with the NW using the updated UE configuration.

[0041] Thus, according to some embodiments, a wireless communication method performed by a UE (such as UE 102 in Figs. 1 and 2) includes at least (a) obtaining (as in step 304) an ML configuration for an ML module of the UE (e.g., 520 in Fig.5), (b) generating (as in step 305) channel quality predictions by inputting channel quality measurements to the ML module, and (c) transmitting (as in step 310), to a NN (e.g., 104 in Figs. 1 and 2), a recommendation, based on the channel quality predictions, to update the current NWC. Updating the current NWC may include: (i) switching on or off a network entity, a cell, or a channel, (ii) using another beam different from a currently used beam for NN-UE communications, (iii) adjusting the transmission power used by the NN for the NN-UE communications, or (iv) changing parameters of a discontinuous transmission and reception operation mode. In addition to the channel quality predictions, the recommendation may be based on local information not concurrently shared with the NN. This local information may include at least one of ML predicted values, UE future application-related traffic information, UE power level information, UE mobility information, and UE location information. The recommendation may indicate a recommended parameter value associated with the NWC. The recommendation may indicate a predefined configuration when a triggering condition associated with the predefined configuration is met. The predefined configuration may be one of NWC candidates received by the UE. The UE may also receive triggering conditions for one or more of the NWC candidates, respectively. The UE may receive a response to the recommendation, the response indicating whether the NWC is updated according to the recommendation. In one embodiment, the UE receives a network message specifying allowed NWC updates.

[0042] The recommendation may indicate the NN to use a specific cell to serve the UE. Such a recommendation may be triggered by a predicted RSRP value of a serving beam being a predetermined amount lower than an RSRP value associated with the specific cell. The UE may initiate a lower layer triggered mobility procedure to switch to the specific cell. Alternatively, the recommendation may suggest using another beam different from the current serving beam. Such a recommendation may be triggered by apredicted RSRP of the currently serving beam being within a predetermined range around the RSRP value of the other beam. Alternatively or additionally to the above-described recommendations, the recommendation may be to decrease the power level of the current serving beam or to use fewer array elements than currently used for beamforming. The UE may transmit a time indication related to updating the NWC, with the recommendation. The UE may transmit the recommendation in a RACH message, using a MAC CE, or in uplink control information, UCI, via a preconfigured uplink resource.

[0043] The method may also include communicating with the NN using a UE configuration updated according to the recommendation a predetermined time after the transmitting of the recommendation, or when receiving a confirmation that the recommendation has been adopted. The method may further include measuring updated configuration signals different from initial signals measured for the channel quality measurements input to the ML module, and reporting, to the NN, the measurement of the updated configuration signals. In one embodiment the method also includes sending, to the NN, UE capability information indicating support for providing an NWC-related recommendation.

[0044] Fig. 4 is a flowchart of a NN method 400 for changing configuration upon receiving an ML-based recommendation from a UE according to an embodiment. The method 400 includes optionally receiving 402 UE capability information regarding providing NWC recommendations. Also optionally, the method 400 includes sending 404 an ML configuration for an ML module of the UE. The method 400 may (i.e. , optionally) further include sending 406 NWC candidates and sending 408 triggering conditions for providing an NWC recommendation. In one embodiment, triggering conditions are specifically associated with individual or a set of NWC candidates.

[0045] A wireless communication device (e.g., 102) including a processor (e.g., 142), a transceiver (e.g., 144, 146), and computer readable recording medium (148) storing executable codes that, when executed by the processor in collaboration with the transceiver, make the wireless communication device to perform method 300 and variants thereof described above.

[0046] The method 400 then includes receiving 410, from the UE, an indication of an NWC recommendation based on the ML predictions and then determining 411 whether to adopt the recommended NWC. The method 400 may also include sending 412 a response to the indicated NWC recommendation indicating whether the recommended NWC is adopted, and then optionally updating 414 the NWC, and communicating 416 with the UE using the updated NWC.

[0047] According to some embodiments, a wireless communication method performed by an NN (such as NN 104 in Figs. 1 and 2) includes at least (a) receiving (as in step 410) a recommendation for updating NWC based on ML channel quality predictions, from a UE (such as UE 102 in Figs. 1 and 2), and (b) transmitting (as in step 412) a response to the recommendation and / or updating (as in step 414) the NWC based on the recommendation. The NN may update at a time indicated in the recommendation. Alternatively or additionally, the NN may update the NWC is according to one or more predefined rules. The recommendation for updating the NWC directs one of (a) switching on or off a network entity, a cell, or a channel, (b) using another beam different from the currently used beam for NN-UE communications, (c) adjusting a transmission power for NN-to-UE communications, or (d) changing parameters of a discontinuous transmission and reception operation mode. The recommendation may explicitly indicate a recommended parameter value. The recommendation may indicate a predefined configuration. The NN may transmit NWC candidates, the predefined configuration being one of the NWC candidates. The NN may also transmit triggering conditions for one or more of the NWC candidates, respectively. The method may further include sending, to the UE, a message confirming an NWC update triggered by the recommendation. The recommendation may be one of (i) using a specific cell to serve the UE, (ii) using another beam than a currently serving beam to serve the UE, (iii) lowering power level of the currently serving beam, or (iv) using fewer array elements than currently used to form a beam serving the UE. The method may also include sending, to the UE, an ML configuration, and / or receiving, from the UE, UE capability information indicating support for providing an NWC-related recommendation.

[0048] A wireless communication device (e.g., 104) including a processor (e.g., 132), a transceiver (e.g., 134, 136), and computer readable recording medium (148) storing executable codes that, when executed by the processor in collaboration with the transceiver, make the wireless communication device to perform method 400 and variants thereof described above.

[0049] Fig. 5 illustrates a dataflow in a UE outputting an NWC recommendation identifier according to an embodiment. The ML module 520 receives measurements related to one or more reference signals (i.e. , RS1 , RS2, etc.) and outputs channel metric predictions (P1 , P2, etc.). The ML module 520 may be trained in a conventional manner or using federated learning. The data for training is made of instances the input to the ML module plus other measurements (e.g. the measured RSRP or performance metrics such as data rate / throughput, or block error rate (BLER)) corresponding to the prediction / output of the ML module. Conventionally, the UE stores the data for training, and sends the data to a NW central server. The central server performs the training and returns the updated ML model to the UE.

[0050] In federated learning, each UE uses the locally stored data for training to compute a local update of the ML module. Then all UEs send their local updated ML model to the NW central server. The NW central server aggregates the local updates and determines a central update of the ML model. The central update then is sent to all UEs, which then use this central update of the ML model.

[0051] The training method can be (1 ) supervised learning or reinforcement learning. For example, in supervised learning, for RSRP prediction, the UE sends input of ML module and the measured RSRP corresponding to the prediction, and the NW central server uses this data to train / fine-tune an RSRP predictor. In reinforcement learning, the ML module outputs an indication of the configuration update triggering the UE and NN to update their configurations. The effect of the update (e.g. certain predefined performance metric, throughput, link outage time, etc.) is stored by UE. The input of the ML module, and the resulting metric is used to train the ML module.

[0052] Returning now to Fig. 5, the ML channel metric predictions are then input to an NWC recommendation module 530. The NWC recommendation module 530 (which, similar with the module 520, includes software and / or hardware and resides in the UE)may receive other local information (L1 , L2, etc.) and outputs a NWC recommendation identifier (RID) selected in view of the predictions from a plurality of RIDs corresponding to NWC candidates using predefined conditions (e.g., as in Table 1 ).

[0053] Figs. 6A and 6B illustrate timelines of UEs (according to two different embodiments) triggering an NWC update based on ML channel quality predictions at the UE. In Figs. 6A and 6B, based on measuring one or more reference signals (RS1 , RS2, etc.), the UE generates 305 channel metric predictions (R1 , P2, etc.) that (potentially together with other local information, L1 , L2, etc.) enable the UE to indicate a recommended NWC.

[0054] In the embodiment illustrated in Fig. 6A, the UE receives 312 a response to the NWC recommendation indicating whether (Y / N) the NW has adopted the recommended NWC. In view of the response, the UE updates 314 the NWC. The time interval (Xms) after receiving the response to the updating of the NWC may be predefined (e.g., signaled by the UE to the NW along with the RID, configured by the NW in the Y / N response, or specified in technical specifications).

[0055] In the embodiment illustrated in Fig. 6B, the UE optionally receives 312 a response to the NWC recommendation only to indicate that the NW has rejected the recommended NWC. Absent receiving any such response for a predetermined time interval (Delta T) after sending the NWC recommendation (310), the UE updates 314 the NWC. The time interval (Delta T, Yms) between receiving the sending the NWC recommendation and updating the NWC (absent receiving a contrary indication) may also be predefined (e.g., signaled by the UE to the NW along with the RID, configured by the NW in a separate message (not shown), or specified in technical specifications).

[0056] The time intervals X and Y may also depend on the NWC recommendation. For example, the UE may estimate when it will need to hand over to a future predicted cell (based on the prediction of its mobility) and indicate the time to NW such that the NW may prepare for the handover.

[0057] The NW response (e.g., an N response) to the NWC recommendation may indicate another NWC than the recommended NWC sent by the UE.

[0058] In some embodiments, updating the NWC may trigger the UE and / or the NN to measure some other RSs prior to the updating. Thus, the NWC recommendationmay also trigger / activate certain RSs (either in DL or UL) for measurements and optionally corresponding resources. For example, when the UE recommends turning on a currently inactive cell / beam, after turning on the beam / cell the NW sends certain RSs via the beam / cell in a preconfigured resource for the UE to measure. The activation of the RSs may be autonomously triggered by the recommendation without additional signaling for configuring these RSs.

[0059] The UE may indicate a time for the NWC to be enacted (e.g., an absolute or relative value or slots). Alternatively, the moment when the NWC is enacted may be determined based on predetermined rules.

[0060] Reference throughout this section to “one embodiment” or “an embodiment” means that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment. Thus, the appearances of the phrases “in one embodiment” or “in an embodiment” in various places throughout the specification are not necessarily all referring to the same embodiment. Further, the particular features, structures or characteristics may be combined in any suitable manner in one or more embodiments.

[0061] Numerical adjectives “first”, “second”, and “third” do not imply any order (are not ordinals) but are markers to distinguish separate instances of similar elements. References to the singular (e.g., “a” or “an”, “the”) should include the plural unless clearly indicated otherwise.

[0062] As used herein, a phrase referring to “at least one of” or “one or more of” a list of items refers to any combination of those items, including single members. For example, “at least one of: a, b, or c” is intended to cover the possibilities of: a only, b only, c only, a combination of a and b, a combination of a and c, a combination of b and c, and a combination of a and b and c.

[0063] Although the features and elements of the present embodiments are described in the embodiments in particular combinations, each feature or element can be used alone without the other features and elements of the embodiments or in various combinations with or without other features and elements disclosed herein. The methods or flowcharts may be implemented in a computer program, software, orfirmware tangibly embodied in a computer-readable storage medium for execution by a specifically programmed computer or processor.

Claims

WHAT IS CLAIMED IS:1 . A wireless communication method (300) performed by a user equipment, UE, (102), the method comprising: obtaining (304) a machine learning, ML, configuration for an ML module to predict channel quality; generating (305) channel quality predictions by inputting channel quality measurements to the ML module; and transmitting (310), to a network node, NN, a recommendation, based on the channel quality predictions, to update a network configuration, NWC, by at least one of: switching on or off a network entity, a cell, or a channel, adjusting a transmission power of NN-to-UE communications, or changing parameters of a discontinuous transmission and reception operation mode.

2. The wireless communication method of claim 1 , wherein the recommendation is further based on local information that is not concurrently shared with the NN, the local information including at least one of:ML predicted values,UE future application-related traffic information,UE power level information,UE mobility information, orUE location information.

3. The wireless communication method of claim 1 or 2, further comprising: receiving, from the NN, NWC candidates, wherein the recommendation indicates one of the NWC candidates.

4. The wireless communication method of claim 3, further comprising: receiving at least one trigger condition to include a specific one of the NWC candidates in the recommendation when the at least one trigger condition is fulfilled.

5. The wireless communication method of any of claims 1 to 4, further comprising: receiving, from the NN, a response to the recommendation, wherein the response indicates whether the NN adopts the NWC included in the recommendation.

6. The wireless communication method of any of claims 1 to 5, wherein the recommendation is to use a specific cell to serve the UE when a predicted reference signal received power, RSRP, value of a serving beam is a predetermined amount lower than an RSRP value associated with the specific cell.

7. The wireless communication method of any of claims 1 to 5, wherein the recommendation is for the NN to decrease a power level of a current serving beam or to use fewer array elements than currently used for beamforming.

8. The wireless communication method of any of claims 1 to 7, wherein the transmitting of the recommendation includes one of: transmitting a random access channel, RACH, message including the recommendation, transmitting a medium access control, MAC, control element, CE, associated with the recommendation, or transmitting uplink control information, UCI, including the recommendation via a preconfigured uplink resource.

9. The wireless communication method of any of claims 1 to 8, further comprising: communicating with the NN using a UE configuration updated according to the recommendation upon receiving a confirmation of the recommendation or when an indication that the recommendation was rejected is not received for a predetermined time interval from transmitting the recommendation.

10. The wireless communication method of claim 9, wherein the predetermined time interval depends on the recommendation.11 . The wireless communication method of any of claims 1 to 10, further comprising: transmitting, to the NN, UE capability information indicating support for providing an NWC-related recommendation.

12. A wireless communication method (400) performed by a network node, NN, (104) the method comprising: receiving (410), from a user equipment, UE, (102) a recommendation for updating a network configuration, NWC, based on ML channel quality predictions performed by the UE, the recommendation specifying at least one of: switching on or off a network entity, a cell, or a channel, adjusting a transmission power for NN-to-UE communications, or changing parameters of a discontinuous transmission and reception operation mode; and transmitting (412), to the UE (102), a response to the recommendation or updating (414) the NWC based on the recommendation.

13. The wireless communication method of claim 12, wherein the updating the NWC is performed at a time indicated in the recommendation.

14. The wireless communication method of any of claims 12 or 13, further comprising at least one of: transmitting, to the UE, a machine learning, ML, configuration for an ML module of the UE to generate the ML channel quality predictions, transmitting, to the UE, NWC candidates, the recommendation indicating one of the NWC candidates,transmitting, to the UE, at least one trigger condition to include a specific one of the NWC candidates in the recommendation when the at least one trigger condition is fulfilled, or receiving, from the UE, UE capability information indicating support for providing an NWC-related recommendation.

15. A wireless communication device (102, 104) comprising a processor (132, 142), a transceiver (134, 136, 144, 146), and computer readable recording medium (138, 148) storing executable codes that, when executed by the processor in collaboration with the transceiver, make the wireless communication device perform any of the methods recited in claims 1 to 14.

Images