Model-assisted beam management procedure

Abstract

Various examples of the disclosure pertain to a beam-management procedure, e.g., for beam tracking, beam refinement, initial acquisition of a beam, etc. According to examples, the beam-management procedure is assisted by a model, e.g., a machine-learning model such as a deep neural network. A gain prediction and / or a beam prediction is possible. For instance, the best beam of the best gain (699) can be predicted.

Description

[0001] D E S C R I P T I O N

[0002] MODEL-ASSISTED BEAM MANAGEMENT PROCEDURE

[0003] TECHNICAL FIELD

[0004] Various examples of the disclosure generally pertain to beam management. Various examples specifically pertain to beam-management procedures that are assisted by models for inferring one or more channel characteristics.

[0005] BACKGROUND

[0006] In a wireless communication system, a signal can travel from the transmit node to the receive node over multiple paths. This is referred to as multipath propagation where signal attenuation (gain) varies on different paths. The propagation over different spatial paths is caused by scattering, reflection, diffraction, and refraction of the radio waves by static and moving objects as well as the transmission medium.

[0007] Different spatial paths can be accessed by using different transmit (TX) beams at the TX node of the wireless communication system and / or by using different receive (RX) beams at the RX node of the wireless communication system.

[0008] To be able to select the best TX beam, knowledge of the radio channel is required. To obtain knowledge of the radio channel, so called beam management is employed. Here, multiple TX beams are tested. For this, the TX node transmits reference signals on multiple TX beams of a TX beam sweep. The RX node attempts to receive (monitors for) the reference signals and then provides a report. For instance, the highest reference signal received power (RSRP) and the associated beam index of the TX beam of the multiple TX beams providing that highest RSRP can be reported.

[0009] Such a technique face certain restrictions and drawbacks. In particular, upon obtaining the report from the UE the BS can only reliably select that particular TX beam from the multiple TX beams indicated by the beam index. Oftentimes, the multiple TX beams included in the initial TX beam sweep do not include the best TX beam, e.g., offering highest gain. Thus, a suboptimal TX beam is selected.

[0010] To mitigate such issues, it has been proposed to identify a channel model based on an output of a machine-learning model. See US 2023 / 0327788 A1 ; 3GPP document R1-2306743 as well as 3GPP TR 38.843 V18.0.0 (2023-12).

[0011] However, also such techniques face certain requirements. It has been found that such channel models have limited accuracy. The predictions provided by such machine-learning models can be inaccurate.

[0012] SUMMARY

[0013] Accordingly, a need exists for advanced be management procedures. This need is met by the features of the independent claims. The features of the dependent claims define embodiments.

[0014] Various disclosed techniques generally pertain to inferring a model to make a prediction on one or more channel characteristics such as a beam prediction and / or a gain prediction for a radio channel. For instance, the prediction may be limited to a gain prediction.

[0015] A method for use in a wireless communication device is disclosed. The wireless communication device is connectable to a cellular network. The method includes monitoring for one more reference signals using at least one receive beam. The one or more reference signals are transmitted by a base station of the cellular network using multiple transmit beams. The method also includes establishing one or more spatial characteristics of the multiple transmit beams. The method further includes communicating with the cellular network in accordance with a prediction of one or more channel characteristics of a radio channel from the base station to the wireless communication device. The prediction of the one or channel characteristics is obtained from a pre-defined model. The prediction of the one or more channel characteristics is based on the one or more spatial characteristics. The prediction of the one or more channel characteristics is further based on said monitoring.

[0016] A wireless communication device including a compute circuitry that is configured to execute such method is disclosed.

[0017] A method for use in a base station of a cellular network is disclosed. The method includes obtaining, from a wireless communication device that is connectable to the cellular network, an indication of one or more channel characteristics of a radio channel from the base station to the wireless communication device. The prediction of the one or channel characteristics is obtained from a pre-defined model. The prediction of the one or more channel characteristics is based on one or more spatial characteristics of multiple transmit beams used by the base station or a further base station for transmitting one or more reference signals. The prediction is further based on the wireless communication device monitoring of the one or reference signals.

[0018] A base station including a compute circuitry that is configured to execute such method is disclosed.

[0019] A system comprising such base station and wireless communication device is disclosed. It is to be understood that the features mentioned above and those yet to be explained below may be used not only in the respective combinations indicated, but also in other combinations or in isolation without departing from the scope of the disclosure.

[0020] BRIEF DESCRIPTION OF THE DRAWINGS

[0021] FIG. 1 schematically illustrates a wireless communication system according to various examples.

[0022] FIG. 2 is a polar plot of various example TX beams.

[0023] FIG. 3 is a polar plot of various example TX beams.

[0024] FIG. 4 is a polar plot of various example TX beams.

[0025] FIG. 5 is a flowchart of a method according to various examples.

[0026] FIG. 6 is a polar plot of various example TX beams, wherein FIG. 6 furthermore illustrates multiple spatial characteristics of the TX beams.

[0027] FIG. 7 schematically illustrates a gain of multiple spatial paths accessed by TX beams having different directions according to various examples.

[0028] FIG. 8 schematically illustrates a gain of multiple spatial paths accessed by TX beams having different directions according to various examples.

[0029] FIG. 9 is a signaling diagram of a beam-refinement-type beam-management procedure according to various examples. FIG. 10 is a signaling diagram of an initial-acquisition beam-management procedure for a handover according to various examples.

[0030] FIG. 11 schematically illustrates a processing pipeline for inferring a pre-defined model according to various examples.

[0031] FIG. 12 is a flowchart of a method according to various examples. DETAILED DESCRIPTION

[0032] Some examples of the present disclosure generally provide for a plurality of circuits or other electrical devices. All references to the circuits and other electrical devices and the functionality provided by each are not intended to be limited to encompassing only what is illustrated and described herein. While particular labels may be assigned to the various circuits or other electrical devices disclosed, such labels are not intended to limit the scope of operation for the circuits and the other electrical devices. Such circuits and other electrical devices may be combined with each other and / or separated in any manner based on the particular type of electrical implementation that is desired. It is recognized that any circuit or other electrical device disclosed herein may include any number of microcontrollers, a graphics processor unit (GPU), integrated circuits, memory devices (e.g., FLASH, random access memory (RAM), read only memory (ROM), electrically programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM), or other suitable variants thereof), and software which co-act with one another to perform operation(s) disclosed herein. In addition, any one or more of the electrical devices may be configured to execute a program code that is embodied in a non-transitory computer readable medium programmed to perform any number of the functions as disclosed.

[0033] In the following, embodiments of the disclosure will be described in detail with reference to the accompanying drawings. It is to be understood that the following description of embodiments is not to be taken in a limiting sense. The scope of the disclosure is not intended to be limited by the embodiments described hereinafter or by the drawings, which are taken to be illustrative only.

[0034] The drawings are to be regarded as being schematic representations and elements illustrated in the drawings are not necessarily shown to scale. Rather, the various elements are represented such that their function and general purpose become apparent to a person skilled in the art. Any connection or coupling between functional blocks, devices, components, or other physical or functional units shown in the drawings or described herein may also be implemented by an indirect connection or coupling. A coupling between components may also be established over a wireless connection. Functional blocks may be implemented in hardware, firmware, software, or a combination thereof.

[0035] Hereinafter, techniques of wirelessly communicating between nodes of a wireless communication system are disclosed. More specifically, techniques are disclosed for wireless communication systems that include at least one node with beamforming capability. A phased- array of antenna elements can be interfaced to direct energy into certain selected spatial directions (beamforming). Techniques associated with such wireless communication system are shown in FIG. 1 for an example implementation in which one node is a base station (BS) 101 of a cellular network and the other node is a UE 102 that is connected or that can connect to the cellular network via the BS 101 or via another BS of the cellular network.

[0036] FIG. 1 illustrates details with respect to the BS 101. The BS 101 implements an access node of a cellular network, e.g., a 3GPP-specified cellular network. The BS 101 includes control circuitry that is implemented by a processor 1011 and a non-volatile memory 1015. The processor 1011 can load program code that is stored in the memory 1015. The processor 1011 can then execute the program code. Executing the program code causes the processor to perform techniques as described herein, e.g.: participating in a beam-management procedure, e.g., by transmitting reference signals, establishing and maintaining a data connection with the UE 102; selecting a serving TX beam for a beamformed transmission to the UE 102, etc.

[0037] FIG. 1 also illustrates details with respect to the UE 102. The UE 102 includes control circuitry that is implemented by a processor 1021 and a non-volatile memory 1025. The processor 1021 can load program code that is stored in the memory 1025. The processor can execute the program code. Executing the program code causes the processor to perform techniques as described herein, e.g.: participating in a beam-management procedure such as a beam refinement procedure or a beam discovery procedure, e.g., for initial access or handover; monitoring for reference signals transmitted by the base station; providing reports on measurements are executed on reference signals received from the BS 101 ; establishing and maintaining a data connection with the BS 101; inferring a pre-defined model such as a machine-learning model, e.g., a deep neural network, to obtain one or more characteristic of the radio channel - e.g., in a predefined frequency band or multiple predefined frequency bands - between the BS 101 and the UE 102.

[0038] FIG. 1 also illustrates details with respect to communication between the BS 101 and the UE 102. The BS 101 includes an interface 1012 that can access and control multiple antennas 1014 of an antenna array. Likewise, the UE 102 includes an interface 1022 that can access and control at least one antenna 1024. In the scenario illustrated in FIG. 1 , the UE 102 has multiple antennas 1024. The UE 102 may also include only a single antenna in which case it cannot beamform.

[0039] While the scenario of FIG. 1 illustrates the antennas 1014 being coupled to the BS 101 , as a general rule, it would be possible to employ transmit-receive points (TRPs) that are spaced apart from the BS.

[0040] The interfaces 1012, 1022 may each include one or more TX chains and one or more RX chains. For instance, such RX chains can include low noise amplifiers, analogue to digital converters, mixers, etc. Analog and / or digital beamforming would be possible. Thereby, phase- coherent transmitting and / or receiving (communicating) can be implemented across the multiple antennas 1014, 1024. Multi-antenna techniques can be implemented. Specifically, beamforming can be employed.

[0041] By using a TX beam at the BS 101 , the direction of signals - e.g., reference signals and data-encoding signals - transmitted to the UE 102 is controlled. Energy is focused into a respective direction or even multiple directions, by phase-coherent superposition of the individual signals originating from each antenna 1014. Thereby, a specific spatial path of the radio channel can be selected (beamforming). Different TX beams select different spatial paths and different spatial paths are associated with different gains and / or delays.

[0042] As a general rule, alternatively or additionally to such TX beams, it is possible to employ RX beams at the UE 102.

[0043] Hereinafter, techniques of beam management are disclosed. Beam management pertains to the selection of the particular beam or beams employed at the BS 101 and / or the UE 102. TX beams and / or RX beams may be selected. Techniques will be specifically disclosed for beam management for selecting of a TX beam at the BS 101 used for transmitting signals to the UE 102. This may also be labeled a downlink (DL) TX beam. However, similar techniques as disclosed hereinafter may be readily employed for selecting RX beams at the UE 102 or selecting RX beams at the BS 101 or selecting TX beams at the UE 102. For instance, upon finding a certain TX beam at the BS 101 , under the assumption of channel reciprocity, also an uplink (UL) RX beam can be selected at the BS 101 , that RX beam being directed similar as the TX beam.

[0044] As a general rule, various types of beam-management procedures may be implemented according to the disclosure. Some possible options are described in TAB. 1.

[0045] TAB. 1 : Two types of beam-management procedures.

[0046] Both types of beam-management procedures as summarized in TAB. 1 , according to the disclosed techniques, employ a prediction I estimate of one or more channel characteristics of the radio channel. “Prediction” means that an initially hidden observable (not detected directly through measurement) is revealed by inferring a pre-defined model. In other words, the one or more channel characteristics are not directly measured, but estimated based on the model. As a general rule, the radio channel for which the prediction is made may be the same as the communication radio channel on which the BS and the UE communicate. It would also be possible that the prediction is made for a larger or smaller radio channel if compared to the communication radio channel. For instance, the prediction may be made for a sub-band of the communication radio channel. In some scenarios, multiple channel characteristics may be predicted, for multiple sub-bands of the communication radio channel. The channel characteristics may be provided frequency-resolved. I.e., it would be possible to determine and / or indicate a frequency band for which a given channel characteristic is valid. That frequency band may be a sub-band of the overall communication radio channel or may even span the entire communication radio channel.

[0047] At least some of the disclosed techniques are based on the finding that it is sometimes possible to reliably predict the one or more channel characteristics based on limited knowledge of the radio channel. A measurement of one or more properties of the radio channel can be executed and it is then possible to determine input data for the pre-defined model based on the measurement. The model can be inferred and this yields the one or more channel characteristics. Such inference of the one or more channel characteristics has the advantage that fewer measurements are required. This tends to reduce control signaling overhead and channel occupancy for channel-sounding purposes. According to various examples, measurements of one or more properties of the radio channel are implemented based on the UE monitoring, using at least one RX beam, for one or more reference signals that are transmitted by the BS of the cellular network using multiple TX beams. I.e., the BS can employ a TX beam sweep to access different spatial paths of the radio channel. Only a fraction of all available spatial paths of the radio channel is probed. Nonetheless, based on these samples of the radio channel, it is possible to make a prediction of the one or more channel characteristics.

[0048] As a general rule, various types of input-output data pairs associated with the predefined model are conceivable. I.e., different measurements are possible as well as different predicted one or more channel characteristics. Some example channel characteristics that may be predicted according to examples disclosed herein are listed in TAB. 2.

[0049] TAB. 2: Various channel characteristics that may be predicted using the techniques disclosed herein. Typically, fewer samples of the radio channel are provided for prediction tasks 1 and 2 if compared to prediction tasks 3 and 4. In other words, it is typically easier to predict the maximum achievable gain if compared to predicting the particular TX beam accessing that spatial path offering the optimal gain.

[0050] For instance, as shown in TAB. 1 , example 1 , sometimes it may be desirable to make a prediction of the globally optimal gain of the radio channel. This corresponds to predicting the path loss of the best spatial path potentially accessible by a BS to serve a UE. In this scenario, it may not be desirable to reveal the exact properties - such as direction or beamwidth - of the TX beam that enables access to that best spatial path (this may be subject to a subsequent beam-management procedure); rather, it may suffice to conclude on the gain of the best spatial path. For instance, such channel knowledge may be helpful t to decide whether to connect to a certain cell, e.g., during initial access or handover, cf. TAB. 1 , TAB. 1. Such knowledge may be helpful in an initial access beam-management procedure. Better decisions for initial access can be taken if knowledge of the globally optimal gain is available.

[0051] Various techniques are based on the finding that for a prediction of the globally optimal gain it is helpful to globally sample (yet typically sparsely) the directional space accessible by TX beamforming at the BS. Such a scenario is illustrated in FIG. 2. In FIG. 2, the entire 360 degrees (e.g., azimuthal) directional space that is accessible by beamforming at the BS 101 is sampled using a total of four TX beams 701-704 having relatively wide beamwidth. The beams 701-704 overlap. The BS 101 may transmit one or more reference signals on each of the beams 701-704. Another variant is shown in FIG. 3; FIG. 3 also illustrates a scenario in which the entire 360 degrees directional space that is accessible by beamforming at the BS 101 is sampled, here using for TX beams 711-714 having a relatively narrow beamwidth. The beams 711-714 do not overlap. The BS 101 may transmit one or more reference signals on each of the beams 711- 714.

[0052] It is noted that not all BSs necessarily need to have a 360 degrees directional space accessible; for instance, some BSs may only have a directional space of 180 degrees that may be accessed. As such, FIG. 2 and FIG. 3 are only illustrative examples. While FIG. 2 and FIG. 3 schematically illustrates the azimuthal directional space, as a general rule, elevation directional space and / or azimuthal directional space may be accessible by beamforming. As such, FIG. 2 and FIG. 3 are to be considered illustrative examples using a 2-D plot; similar techniques may be readily applied for a 3-D directional space. Hereinafter, for sake of simplicity, reference will be primarily made to an azimuthal directional space. Based on such sparse and spread-out sampling of the directional space, it is possible to make a reliable prediction of the globally optimal gain.

[0053] Another example - cf. TAB. 1 , examples 2 and 3 - is local predictions, e.g., locally optimal gain or even locally optimal TX beam that accesses the spatial path offering the locally optimal gain. In this scenario, it is helpful to sample at a relatively high density (e.g., if compared to the scenario discussed above in connection with FIG. 2 and FIG. 3 / TAB. 1, example 1) the directional space in the subarea in which that local channel characteristic is to be predicted. Such example is shown in FIG. 4. In FIG. 4, the TX beams 721, 722, 723 locally sample a roughly + / - 20 degrees subarea of the 360 degrees directional space accessible by the BS 101.

[0054] Thus, as will be appreciated from the above discussion of TAB. 1 and FIG. 2, FIG. 3, and FIG. 4, it is possible to predict different types of channel characteristics - e.g., global or local channel characteristic, gain or beam characteristic. Depending on the desired channel characteristic to be predicted, a different spatial characteristic of the multiple TX beams used for channel sounding can be selected, e.g., wide vs. narrow beams, densely-packed or even overlapping TX beams sampling a subarea of the directional space, or TX beams sampling the entire directional space of the BS. Various techniques are based on the finding that for inference of the pre-defined model that accomplishes such prediction, it is thus helpful to establish, at the UE, such one or more spatial characteristics of the TX beams used for channel sounding - e.g., their directions, beam widths, and / or relative arrangement. This enables an accurate prediction of the one or more channel characteristics.

[0055] Details with respect to such techniques according to which the UE establishes the one or more spatial characteristics of those TX beams used to probe the radio channel are disclosed below in connection with FIG. 5.

[0056] FIG. 5 is a flowchart of a method according to various examples. The method of FIG. 5 is for use in a UE, e.g., the UE 102 is illustrated in FIG. 1. The method of FIG. 5 can be executed by a processor, upon loading program code from a memory and upon executing the program code. For instance, the method of FIG. 2 may be executed by the processor 1021 upon loading program code from the memory 1025 and upon executing the program code. The method of FIG. 5 generally pertains to the UE participating in a beam-management procedure, e.g., one of the beam-management procedures as outlined in TAB. 1.

[0057] When executing FIG. 5, the UE may operate in a connected state. I.e., data connection may be active between the BS and the UE. This is, in particular, the case for a refinement-type beam-management procedure, cf. TAB. 1: example 2. The data connection may also be active for a handover-triggered initial-acquisition-type beam-management procedure.

[0058] In other scenarios, when executing FIG. 5, the UE may operate in a disconnected state. I.e., data connection may not be active between the BS and the UE. This is, in particular, true for an initial-access triggered initial-acquisition beam-management procedure, cf. TAB. 1, example 1.

[0059] The beam-management procedure includes a UE-side inference of a pre-defined model. The pre-defined model, when inferred, provides a prediction / estimation of one or more channel characteristics of the radio channel from the BS to the UE. While the particular type of predefined model is not germane for the techniques disclosed herein, it would be, in particular, possible to employ a pre-trained machine-learning model. Options include deep neural networks (DNNs).

[0060] At box 3005, it is optionally possible to provide, to the cellular network, an indication of an ability of the UE to infer the pre-defined model. More generally, at box 3005, it would be possible that the UE indicate, to the cellular network, that it is capable of participating in the beam-management procedure that employs UE-side inference of the pre-defined model. Such indication may be provided to the serving BS.

[0061] Box 3005 can include capability signaling. Box 3005 can include Layer 3 signaling, e.g., Radio Resource Control (RRC) signaling. For instance, as part of a procedure to establish a data connection between the UE and the cellular network, the UE may indicate that it is capable to infer the pre-defined model. For instance, the UE may indicate a type of the pre-defined model or multiple types of pre-defined models available as candidates. The UE may indicate for which specific types of beam-management procedures (cf. TAB. 1) it can infer a suitable predefined model. The types of pre-defined models available to the UE may be indicated. At box 3010, it is optionally possible that the UE provides a request for executing a beammanagement procedure to the cellular network. Alternatively or additionally, it would also be possible that the cellular network provides such request to the UE and the UE obtains, at box 3010, such request.

[0062] At box 3015, the UE may optionally obtain, from the cellular network, a configuration of one or more reference signals to be transmitted by the BS using multiple TX beams. For instance, signal properties of the one or more reference signals may be indicated. Different reference signals may have different signal shapes or sequence designs. For instance, timefrequency resources for the transmission of the one or more reference signals may be indicated. For instance, it may be indicated whether the multiple TX beams are activated sequentially or at the same time; this may depend on an analog versus digital beamforming capability of the BS. A bandwidth or frequency-hopping pattern used for the transmission of the one or more reference signals may be indicated.

[0063] As a general rule, it would be possible that one or more reference signals are transmitted spread out across a channel bandwidth. I.e., a frequency-dependent general characteristic may be accessible through such one or more reference signals that are transmitted across the channel bandwidth. It would also be possible that the one or more reference signals are transmitted on a subset of all available subcarriers or bandwidth parts.

[0064] As a general rule, the one or more reference signals used for channel sounding as part of the beam-management procedure may be UE-specific (e.g., when the UE has an established data connection with the cellular network) or may be cell-specific. As a general rule, where a configuration for the one or more reference signals is obtained by the UE, such configuration may be UE-specific or maybe cell specific. For instance, such configuration may be obtained from a broadcast message - e.g., using a system information block (SIB) - or may be obtained in a direct communication message targeting the UE - e.g., using an RRC control message communicated on a physical downlink shared channel (PDSCH).

[0065] Box 3015 is optional, because sometimes such properties of the one or more reference signals may be fixedly predefined.

[0066] At box 3016, it is optionally possible to obtain, from the cellular network, a configuration of a reporting scheme for reporting on the one or more channel characteristics of the radio channel that is predicted using the pre-defined model. For instance, the type of information to be included in a respective report message may be specified. A timing of the reporting may be configured. It may be specified which one or more channel characteristics are to be predicted. Box 3016 is optional. Sometimes, such configuration may be predefined.

[0067] For instance, the frequency bandwidth of the radio channel for which the one or channel characteristics are to be predicted may be configured at box 3016. For instance, it would be possible that the one or channel characteristics are to be predicted for one or more sub-bands of the communication radio channel employed for communicating between the UE and the cellular network. It would also be possible that the prediction is to be determined for the entire communication radio channel. In some scenarios, it would even be possible that multiple predictions of the one or more channel characteristics are to be determined, different predictions being associated with different frequency bands.

[0068] The configuration of the reporting scheme may also request that the UE reports on one or more confidence intervals of the prediction. The reporting scheme may request that the UE reports on an uncertainty and / or probability of the prediction. Respective techniques will be later on described in further detail in connection with box 3045.

[0069] At box 3020, the UE monitors for the one or more reference signals. Box 3020 may include attempting to receive or even receiving one or more reference signals.

[0070] Box 3020 may be in accordance with a configuration is obtained at box 3015. For instance, the UE may attempt to receive the one or more reference signals at one or more resource elements indicated in the configuration. A channel measurement is implemented.

[0071] For instance, the UE may use a single RX beam for monitoring for the one or more reference signals at box 3020. This may, in particular, be the case if the UE does not have beamforming capability, e.g., because it only has a single antenna element. In another scenario, it would be possible that the UE employs multiple RX beams to monitor for the one or more reference signals. In RX beam sweep can be implemented at box 3020. the UE may have analog or digital beamforming capability. Accordingly, multiple RX beams may be activated contemporaneously or sequentially. Using multiple RX beams has the advantage of obtaining additional channel knowledge, enabling more accurate predictions of the one or more channel characteristics.

[0072] For instance, it would be possible that the reference signals are Channel State Information (CSI) reference signals as specified by the 3GPP TS. Alternatively or additionally, it would be possible that the reference signals are included in a SSB, e.g., primary synchronization signals (PSS) or secondary synchronization signals (SSS) as specified by the 3GPP TS .

[0073] The one or more reference signals are transmitted by the BS using multiple TX beams. These multiple transmit beams have different directions, i.e., they select different spatial paths of the radio channel from the BS to the UE. The radio channel is thereby sounded. Multiple spatial paths are sampled. The radio channel is not completely sounded, but sufficiently sampled to reveal one or more channel characteristics by using a pre-defined model. Such prediction of the one or more channel characteristics can be facilitated by providing knowledge of one or more spatial characteristics of the multiple TX beams used for transmitting the one or more reference signals at box 3020.

[0074] At box 3025, the one or more spatial characteristics of the multiple TX beams are established. For instance, it would be possible to establish a direction of each of the multiple TX beams. Such direction indication could be based on directional cosines. Alternatively or additionally, the direction indication could include 3-D volumes in near-field beamforming. Alternatively or additionally, the direction indication could be indicative of the TX beams being uniformly spread around, and in the vicinity of, the current serving TX beam. To give a concrete, non-limiting example: The direction information may be given as an entry of a pre-defined table (e.g., n TX beams lying on a circle (defined in a directional cosine plane)around the current serving TX beam with radius r, n and rwould be a table entries).

[0075] Alternatively or additionally, it would be possible that the one or more spatial characteristics of the multiple TX beams include a beamwidth of each of the multiple TX beams. Alternatively or additionally, the one or more spatial characteristics of the multiple TX beams may include a relative arrangement of the multiple TX beams with respect to each other, e.g., an angular offset between the beam directions of the multiple TX beams, etc.

[0076] FIG. 6 illustrates spatial characteristics of multiple TX beams 181, 182, 186, 190, 199. For instance, the TX beams 181 , 182, 190 all have the same beamwidths 281 , 282, 290. These are narrow pencil beams, specifically suitable for densely sampling a sub-area of the directional space that is accessible by the BS 101, e.g., as part of a beam-refinement-type beammanagement procedure. Differently, the TX beam 199 is a broad beam having a large beamwidth 299. Furthermore, illustrated are the directions 381, 386, 390 of the TX beams 181, 186, 190. The directions 381 , 390 enclose an angle 300 (angular offset) that is smaller than the beamwidths 281 , 290 so that the TX beams 181 , 190 overlap. This is different for the TX beam 190 and the TX beam 186; the respective angle 301 is significantly larger than the beamwidths 290, 286 so that these TX beams 190, 186 do not overlap.

[0077] FIG. 6 schematically illustrates some example spatial characteristics of TX beams. Other spatial characteristics are conceivable. For instance, FIG. 6 is a 2-D polar plot. However, beams may be oriented in 3-D space. Here, directions may be defined with respect to azimuth and elevation angle in 3-D directional space.

[0078] Referring again to FIG. 5: As a general rule, there are various options available for implementing box 3025. For instance, as illustrated in box 3026, it would be possible that said establishing of the one or more spatial characteristics includes obtaining, from the cellular network, an indication of the one or more spatial characteristics. For instance, this may be responsive to providing a request at box 3010. For instance, an RRC control message may be obtained, e.g., on PDSCH. It would also be possible that the cellular network broadcasts the one or more spatial characteristics, e.g., in an SIB, on a broadcast channel. Such scenario of obtaining the indication of the one or more spatial characteristics from the cellular network has the advantage that the cellular network can dynamically reconfigure the one or more spatial characteristics of the multiple TX beams used as part of the beam-management procedure. For instance, more or fewer TX beams may be used at different occasions. TX beams using different directions may be used. For instance, TX beams of varying beamwidth may be used, to give just a few examples. Thereby, the number of TX beams used by the BS for channel sounding can be dynamically adapted to the coverage scenario of a particular UE, improving the quality of the channel sounding. Specific subregions of the directional space accessibly by TX beamforming at the BS may be sampled depending on where a specific UE is located. Accordingly, box 3026 may have particular advantages for a beam-refinement-type beammanagement procedure, cf. TAB. 1, example 2.

[0079] In another scenario, the one or more spatial characteristics may be predefined, e.g., in a communication protocol. For example, as illustrated in FIG. 2, box 3027, it is possible that the one or more spatial characteristics are established in accordance with a predefined rule set of the radio access communication protocol of the cellular network. Such scenario has the advantage of reduced communication overhead. The UE may not need to communicate with the cellular network to establish the one or more spatial characteristics based on the local rule set. Such a scenario may be particularly helpful if a global characteristic of the radio channel is to be predicted, e.g., as part of an initial acquisition type beam-management procedure, cf. TAB. 1: example 1. Here, a fixed set of TX beams - having predefined spatial characteristics - may be used irrespective of the particular position of the UE.

[0080] In some scenarios, it is optionally possible, at box 3030, to select the pre-defined model to-be-inferred at box 3035. The pre-defined model can be selected from multiple candidate predefined models, e.g., as indicated to the cellular network at box 3005. For instance, such selection may be based on the one or more spatial characteristics as established at box 3025. The selection may be based on the type of the prediction. The selection may be based on the particular type of one or more channel characteristics to-be-predicted.

[0081] For instance, depending on the particular spatial orientation of the TX beams, different pre-defined models may be more suited for inference of the one or more channel characteristics than others. For instance, different predefined models may be used for making different predictions in accordance with TAB. 2. For instance, different predefined models may be used for different types of beam-management procedures according to TAB. 1. Different pre-defined models may be defined or trained in a domain-specific to solving a specific task, e.g., making different predictions as shown in TAB. 2. In some scenarios, it would be possible that a foundation model is used that is able to solve multiple tasks, e.g., make multiple predictions as listed in TAB. 2; in such scenario, a selection may not be required. At box 3035, the model - e.g., a DNN - is inferred. Box 3035 is based on, both, box 3020, as well as box 3025. If box 3030 is executed, the model inferred at box 3035 is the model selected at box 3030.

[0082] Box 3035 yields one or more channel characteristics of the radio channel, cf. TAB. 2. This means that one or more input data is provided to the model and the output data of the model, defining the prediction, is derived by executing the algorithm defined by the model. An inference forward-pass of a DNN and can be performed. If box 3030 is executed, then the particular pre-defined model selected at box 3030 is inferred at box 3035.

[0083] At box 3035, input data is provided to the predefined model. The input data is determined based on monitoring the reference signals at box 3020. The input data is based on the channel measurements. The input data is based on the channel sounding. The input data can be based on respective channel measurements.

[0084] For instance, it would be possible that the input data includes received amplitude values and / or received phase values of the one or more reference signals of box 3020, for each of the multiple TX beams employed by the BS. For instance, the input data may include the received amplitude values and / or the received phase values of the one or more reference signals for each of multiple subcarriers of a multicarrier transmission of the BS. In other words, if the one or more reference signals are transmitted on multiple subcarriers, then received amplitude values and / or received phase values can be processed for each of those multiple subcarriers. Alternatively or additionally, the input data may include the received amplitude values and / or received phase values for each RX beam employed by the UE. Considering the UE employs m= 3 RX beams and the BS employs n=6 TX beams. There are s=1024 subcarriers. Both received amplitude value as well as received phase values are processed by the pre-defined model. Then, an array of size mxnxsx2=3x6x1024x2=36864 is processed. As will be appreciated, this corresponds to big-data processing. In particular, by implementing UE-side inference of the predefined model, communicating such large amounts of data can be dispensed with, reducing control signaling overhead.

[0085] Box 3035 may indirectly depend on box 3025. In other words, inference of the predefined model may indirectly depend on the one or more spatial characteristics as established at box 3025. This is the case if the one or more spatial characteristics established at box 3025 are used to select the predefined model at box 3030 that is then subsequently inferred at box 3035. In other examples, the inferring of the predefined model may obtain, as input data, the one or more spatial characteristics are established at box 3025. in such a scenario, box 3035 directly depends on the one or more spatial characteristics as established at box 3025. For instance, such scenario may be particularly applicable if TX beams having spatial characteristics that are dynamically set by the BS is used. Such scenario may also be applicable if a foundation model is used that is able to handle all sorts of different values of the spatial characteristics, e.g., in a large input parameter space.

[0086] At box 3040, the UE communicates with the cellular network in accordance with the prediction of the one or more channel characteristics as provided by box 3035. Box 3040 may include deciding on whether to connect to or not connect to the BS that transmitted the reference signals at box 3020, cf. TAB. 1 : example 1 ; shown in Box 3041. Such connection may be part of initial access or may be part of a handover, to give just two examples. Such decision may be additionally based on the uncertainty and / or probability associated with the prediction. For instance, if the uncertainty of a certain gain prediction used as a decision-basis for determining whether to connect to a certain cell is relatively high, the UE may favor to proceed measuring other cells. Similarly, alternatively or additionally, if the probability of a certain gain prediction used as a decision-basis for determining whether to connect to a certain cell is relatively low, the UE may likewise favor to proceed measuring other cells.

[0087] Alternatively, box 3040 may include providing, to the cellular network, an indication of the one or more channel characteristics, box 3042. Such reporting may be in accordance with the reporting configuration of box 3016. The BS may then refine the serving TX beam used for transmitting data associated with an established data connection - e.g., on the PDSCH - based on such report. For instance, the UE may provide an indication of a direction of a further TX beam, e.g., the globally best beam. As a general rule, such direction indication could be implemented in a 3- D volume for near-field beamforming or could be represented in some other way. For example, an entry of a pre-defined table may be provided.

[0088] It would be optionally possible - box 3045 - to provide, to the cellular network, an uncertainty of the prediction. The uncertainty of the prediction can enable the cellular network to make better decisions in the beam management procedure. The uncertainty can be implemented by at least one of the aleatoric uncertainty or the epistemic uncertainty. As a general rule, aleatoric uncertainty (also referred to as statistical uncertainty) refers to the notion of randomness, that is, the variability in the outcome of the prediction which is due to inherently random effects in the input data, here specifically the channel measurements. Example sources of aleatoric uncertainty are measurement or data noise, drifts the channel measurement, etc. Aleatoric uncertainty is different than epistemic uncertainty (also referred to systematic uncertainty). Epistemic uncertainty refers to uncertainty caused by limitations of the pre-defined model, e.g., limited model accuracy, etc.. Accordingly, epistemic uncertainty can stem from incomplete sampling of the input space with training data during the training of a DNN. For example, considering a DNN that is trained to discriminate cats from dogs; however, the training data only captures certain races of dogs (e.g., Swiss Saint Bernard dogs and German Shepard dogs), while other races of dogs (e.g., Mexican Chihuahua) are not captured. Then, the epistemic uncertainty will increase if a discrimination between cat and dog is to be made based on a picture for a race of a dog that has not been captured by the training model (e.g., Chihuahua). As opposed to aleatoric uncertainty, epistemic uncertainty can be reduced based on re-training the DNN. It is possible to determine the uncertainty based on stochastic dropout sampling, e.g., Monte-Carlo dropout sampling. Such stochastic dropout sampling includes multiple forward passes through the DNN, wherein for each of the multiple forward passes a random subset of weights or interconnections (neurons in a deep neural network) are disabled I set to zero, i.e. , temporarily removed from the DNN. By performing multiple forward passes, multiple predictions are obtained. The variation of the predictions is a measure of the uncertainty. Further, it would be possible to determine the uncertainty (more specifically the epistemic uncertainty) based on ensemble sampling of the distribution of the respective prediction. For this, multiple channel measurements may be repetitively executed, i.e., the one or more reference signals may be repetitively received for a given frequency and TX beam and RX beam.

[0089] It would be optionally possible- box 3045 - to provide, to the cellular network, a probability of the prediction. Probability of the prediction, e.g., a categorial probability associated with the prediction if the model is a classifier (e.g., optimal global gain higher than a threshold); or a Gaussian probability distribution associated with the prediction, if the model is a regression model (here specifying the direction of the best beam in a continuous directional space. The probability of a prediction is different than the uncertainty of the prediction. The probability measures the likelihood or belief in an inferred outcome, while the uncertainty characterizes the range of possible outcomes due to imperfect information. For instance, categorial probabilities of multiple classes amongst which a DNN discriminates can be obtained from a softmax activation function at an output layer of the DNN. Softmax is only one example of an activation function. Other activation functions can be used at the output layer of the DNN and it is also possible to obtain categorial probabilities of multiple classes from other types of activation functions. To give a concrete example of the difference between probability and uncertainty: a deep neural network can be trained to classify images as depicting a "dog" or a "cat". When the deep neural network processes an image that depicts only a dog, the corresponding categorical probability of the "dog" class will be high, while the categorical probability of the "cat" class will be low. The uncertainty is also likely to be relatively low, unless the image is noisy or includes an odd perspective. On the other hand, if the deep neural network processes an image that shows both a dog and a cat, then the categorical probability of the class "dog" will be about 50% and the categorical probability of the class "cat" will be about 50%. The uncertainty would remain unaffected by the picture depicting both the cat and the dog; i.e., is also likely to be relatively low, unless the image is noisy or includes an odd perspective. For instance, if multiple TX beams are predicted to have a relatively high gain, then it may be of interest to provide their probabilities. If the probabilities of these predictions are highly asymmetric (e.g., one predicted beam direction has a high probability to be the best beam, while others have a low probability), then the choice for beam refinement may be unambiguous; else, the BS may decide to perform further channel sounding.

[0090] Next, some concrete examples are provided. For instance, a gain prediction may be executed, cf. TAB. 2: example 1 and example 2. The gain prediction could be executed as a classification task, to predict whether the gain, e.g., the globally optimal gain, is above or below a certain predefined threshold. It would then be possible to provide the uncertainty associated with this classification-type prediction. For instance, a higher uncertainty may be rooted in a limited availability of radio channel measurements. A higher uncertainty may be rooted in a certain unfavorable arrangement of the TX beams used by the BS to sample the radio channel. It would also be possible to provide the probability of that prediction, e.g., whether the probability of the gain being above the predefined threshold has a certain value, e.g., defined with respect to a further threshold. For instance, it could be signaled whether the probability that the gain is above the predefined gain threshold is at least 80%. For instance, a lower probability may be rooted in conflicting features or patterns seen in the radio-channel measurements. Some received reference signals may be indicative of a relatively low gain, while other received reference signals may be indicative of a relatively high gain. Such indication of the uncertainty and / or the probability is not only feasible for classification-type predictions. Such indication of the uncertainty and / or the probability is also feasible for regression-type predictions, e.g., to determine a concrete value of the gain order to determine a certain specific direction of a TX beam to be used by the BS for a beam-prediction task.

[0091] FIG. 7 schematically illustrates the gain 692 of the radio channel from the BS 101 to the UE 102 for multiple spatial paths that are accessible by TX beams having various directions 691 , e.g., defined in a 2-D or 3-D directional space. The UE 102 is connected to the BS 101. The BS 101 currently uses a serving TX beam 170 for transmitting data-encoding signals associated with the data connection between the BS 101 and the UE 102. The serving TX beam 170 has a decent gain 692; however, the serving TX beam 170 does not access the globally optimal spatial path. The globally optimal spatial path is accessed by a further TX beam 190.

[0092] FIG. 7 illustrates aspects with respect to multiple TX beams 181, 182, 183, 184 of a TX beam sweep at the BS 101. The BS 101 transmits one or more reference signals using the TX beams 181-184. As illustrated in FIG. 7, the TX beams 181-184 locally sample a subarea 610 of the entire directional space 609 that is accessible by TX beamforming at the BS 101. For instance, the TX beams 181-184 may be overlapping. They may all have similar beamwidths. They may be pencil beams. The beamwidth of the TX beams 181-184 may be the same or somewhat the same as the beamwidth of the serving TX beam 170. That subarea 610 is roughly centered around the serving TX beam 170. More generally, the beam widths of the TX beams 181-184 may be in the range of 50%-150% of the beam width of the serving TX beam 170. The multiple TX beams 181-184 are used as part of a beam refinement type beammanagement procedure. The goal of the beam-refinement-type beam-management procedure is to reveal the globally optimal TX beam 190.

[0093] Using a predefined model that is trained to predict the direction of the globally optimal TX beam 190 (dashed-dotted arrow in FIG. 7) is enabled by the received amplitude values and / or the received phase values of the reference signals transmitted by the BS 101 on each of the multiple TX beams 181-184. Optionally, the uncertainty or probability of that prediction can be determined.

[0094] FIG. 8 schematically illustrates the gain 692 of the radio channel from the BS 101 to the UE 1024 multiple spatial paths that can be accessed by TX beams having various direction 691. The UE 102, in the scenario FIG. 8, is not connected to the BS 101. Thus, there is no serving TX beam (as in FIG. 7).

[0095] FIG. 8 illustrates aspects with respect to multiple TX beams 186, 187, 188 of a TX beam sweep at the BS 101. The BS 101 transmits one or more reference signals using the TX beams 186, 187, 188. As illustrated in FIG. 8, the TX beams 186-188 globally sample the directional space 609 that is accessible by TX beamforming at the BS 101. The multiple TX beams 186- 188 are used as part of an initial-acquisition-type beam-management procedure, cf. TAB. 1: example 1. The goal of the initial-acquisition-type beam-management procedure is to reveal the globally optimal gain 699. Using a predefined model that is trying to determine a prediction of the globally optimal gain 699 (circle in FIG. 8) is enabled by the received amplitude values and / or the received phase values of the one or more reference signals transmitted by the BS 101 on each of the multiple TX beams 186-188.

[0096] In some scenarios, the output of the model may be limited to a gain prediction, i.e., may not include any other predictions such as a beam prediction.

[0097] The UE 102 may decide whether to connect or not connect to the BS 101 based on the globally optimal gain 699. Optionally, the UE may take into account an uncertainty 699-1 of this prediction of the globally optimal gain 699.

[0098] FIG. 9 is a signaling diagram of communication between the UE 102 and the BS 101. For instance, the signaling of FIG. 9 may implement the method of FIG. 5.

[0099] At 5005, capability signaling 4005 is provided by the UE 102 and obtained by the BS 101. 5005 can implement box 3005 of the method of FIG. 5.

[0100] At 5010, data encoding signals 4099 are transmitted by the BS 101 to the UE 102. The data-encoding signals 4099 are associated with the data connection. The UE 102 operates in the connected mode. The BS 11 transmits the data-encoding signals 4099 using a serving TX beam. For instance, the serving TX beam 170 as shown in FIG. 7 may be used.

[0101] At 5015, the UE 102 provides a request 4010 to the BS 101. The request 4010 is for a beam-refinement-type beam-management procedure refining the serving TX beam. For instance, the UE 102 may provide the request 4010 responses to increase the bit error rate. This may implement box 3010.

[0102] The BS 101 provides one or more configuration messages 4012, at 5017. The one or more configuration messages 4012 may implement configurations explained previously in connection with box 3015 and box 3016.

[0103] The BS 101 transmits CSI reference signals 4015 at 5020, using multiple TX beams. A TX beam sweep is implemented. For instance, the TX beams 181-184 as shown in FIG. 7 may be used. Analog and / or digital beamforming may be implemented. This implements box 3020.

[0104] At 5025, the BS 101 provides an indication 4020 of one or more spatial characteristics of the multiple TX beams used at 5020. This implements box 3026.

[0105] The UE 102 provides, at 5030, an indication of the one or more channel characteristics that is obtained from a predefined model based on the one or more spatial characteristics of 5025 as well as based on monitoring the CSI reference signals at 5020. More specifically, the UE 102 provides an indication of the locally best TX beam, i.e., its direction and optionally its beam width. The respective report message may also include an uncertainty of that prediction. The respective report message may also include a probability of multiple predicted locally best TX beams.

[0106] The BS 101 then refines the serving TX beam and using the refined serving TX beam, at 5035, transmits further data-encoding signals 4099. For instance, the TX beam 190 as shown in FIG. 7 may be used.

[0107] As will be appreciated, FIG. 9 corresponds to the example 2 of TAB. 1 in combination with example 3 of TAB. 2. FIG. 10 is a signaling diagram of communication between the UE 102 and the BS 101. For instance, the signaling of FIG. 10 may implement the method of FIG. 5.

[0108] 5105 corresponds to 5005 is illustrated in FIG. 9. 5110 corresponds to 5010 as illustrated in FIG. 9.

[0109] At 5115, the BS 101 provides a request 4010 to the UE 102. The request 4010 is in preparation of a handover. The request 4010 requests the UE to assess the radio channel between the further BS 101 and the UE 102.

[0110] The UE 102, accordingly, at 5120, monitors for PSS and SSS reference signals included in the SSBs 4115 transmitted by the BS 103 using multiple TX beams.

[0111] At 5130, the UE 102 provides the report message 40252 the BS 101. The report message is indicative of the globally optimum gain of the radio channel from the BS 103 to the UE 102. For instance, a numerical value indicative of the globally optimum gain can be provided. It would also be possible to provide a 1 -bit Boolean flag that is indicative of whether the globally optimum gain is above or below a predefined threshold. Optionally, an uncertainty of such prediction may be provided. Depending on such reporting, the BS 101 then selectively triggers a handover at box 5135 using a handover control message 4150.

[0112] An access procedure 4155 is executed at 5140. This can include a further beammanagement procedure to identify the TX beam to be used at the BS 1034 subsequently transmitting data-encoding signals 4099 at 5145. For instance, a TX beam sweep employed at 5140 may include significantly more TX beams if compared to a TX beam sweep employed at 5120. This is because at 5140 it is required to determine the particular TX beam, i.e., the direction and / or beamwidth of the TX beam, used for serving the UE 102. Differently, at box 5120, it is only required to provide a prediction of the maximum achievable gain; the latter is typically a simpler task that can be executed successfully on relatively limited measurements of the radio channel.

[0113] FIG. 11 schematically illustrates a data processing pipeline 900. A predefined model 920 - e.g., a DNN - is provided. The predefined model 920 obtains input data 911. For instance, the input data 911 may be an array including received amplitude values and received phase values of one or more reference signals for each of multiple TX beams and for each of multiple frequency ranges (e.g., subcarriers, bandwidth parts, etc.). Optionally, further input data 912 may be provided. The further input data 912 may be one or more spatial characteristics of the multiple TX beams, e.g., their directions, the angular offset between those TX beams, and / or their beamwidth, etc.

[0114] The predefined model 920 provides output data 930. The output data 930 corresponds to one or more channel characteristics of a radio channel that is sounded by means of the one or more reference signals. Example one or more channel characteristics have been previously discussed in connection with TAB. 2.

[0115] FIG. 12 is a flowchart of a method according to various examples. The method of FIG. 12 pertains to set up, deployment and inference of a predefined model, such as the predefined model 920 is illustrated in FIG. 11.

[0116] At box 3105, the predefined model 920 is parametrized. For instance, for a machine learning model, a machine learning algorithm is executed, e.g., back propagation. This is based on a training data set that links input data to output data (cf. FIG. 11). The output data as ground truth. According to various examples, such pairs of input data and output data can be obtained from channel simulations.

[0117] At box 3110, the predefined model 920 is deployed. The predefined model may be provided to multiple UEs. And over-the-air update may be executed.

[0118] At box 3115, the predefined model is inferred. Details have been disclosed above in connection with FIG. 5: box 3035.

[0119] The particular type of pre-defined model is not germane for the techniques disclosed herein. However, according to various examples, the pre-defined model may be a DNN. A DNN is a specific type of an artificial neural network that contains multiple layers of interconnected nodes or neurons, designed to model complex patterns in data through a hierarchical feature extraction process. In the context of a Convolutional Neural Network (CNN), which is a specialized kind of DNN. A Convolutional Neural Network (CNN) - a special type of DNN - comprises several layers that automatically and adaptively learn spatial hierarchies of features. A CNN typically comprises of an input layer, multiple hidden layers, and an output layer. The hidden layers include convolutional layers, activation functions, pooling layers, and fully connected layers. Convolutional layers apply a convolution operation to the input, passing the result to the next layer. This operation involves sliding a set of trainable filters or kernels over the input image to produce feature maps, capturing local dependencies and spatial hierarchies of features. Activation functions, such as the Rectified Linear Unit (ReLU), introduce nonlinearities into the network, allowing it to learn complex patterns. Pooling layers, typically max pooling, reduce the spatial dimensions of the input by downsampling, reducing the number of parameters and computation in the network, and helping to make the representation invariant to small translations of the input. Fully connected layers, which come after several convolutional and pooling layers, flatten the high-level features learned and combine them to perform classification. The output layer typically uses a softmax function for multi-class classification tasks, providing a probability distribution over different classes. The training of a CNN involves backpropagation and optimization algorithms, such as stochastic gradient descent, to adjust the weights of the filters to minimize a loss function. This process requires a large amount of labeled data provided in the training dataset.

[0120] Summarizing, techniques have been disclosed that enable to predict, based on a small number of swept TX beams at the BS, whether it is worthwhile for the UE to connect to that BS and / or whether it is worthwhile for the BS to continue to continue to transmit one or more reference signals on or more TX beams. That is, if the UE detects, from the small number of swept TX beams, that the globally best gain is very low, then the BS can abandon any further beam sweeps, and serve the UE by other means (for example from another TRP). Furthermore, techniques have been disclosed to determine a better serving TX beam when a UE is already served by another TX beam.

[0121] Further summarizing, at least the following EXAMPLES have been disclosed:

[0122] EXAMPLE 1. A method for use in a wireless communication device (102) connectable to a cellular network, the method comprising: - monitoring (3020), using at least one receive beam, for one or more reference signals transmitted by a base station (101) of the cellular network using multiple transmit beams (181 , 182, 183, 184, 186, 187, 188, 190, 199, 701, 702, 703, 704, 711 , 712, 713, 714),

[0123] - establishing (3025, 3026, 3027) one or more spatial characteristics (281, 282, 286, 290, 299, 300, 381 , 386, 390) of the multiple transmit beams,

[0124] - communicating (3040, 3041 , 3042) with the cellular network in accordance with a prediction of one or more channel characteristics of a radio channel from the base station (101) to the wireless communication device (102), the prediction of the one or more channel characteristics being obtained from a pre-defined model (920), the prediction of the one or more channel characteristics being based on the one or more spatial characteristics and further being based on said monitoring (3020), wherein the one or more channel characteristics comprises a gain of a path of the radio channel accessed by a further transmit beam of the base station (101).

[0125] EXAMPLE 2. The method of example 1 , wherein the gain of the path of the radio channel is a globally optimal gain of the radio channel.

[0126] EXAMPLE 3. The method of example 1 , wherein the gain of the path of the radio channel is a locally optimal gain of the radio channel.

[0127] EXAMPLE 4. The method of any one of the preceding examples, wherein the one or more channel characteristics do not comprise one or more spatial characteristics of the further transmit beam.

[0128] EXAMPLE 5. The method of any one of the preceding examples, wherein the multiple transmit beams globally sample a directional space accessible by beamforming at the base station (101).

[0129] EXAMPLE 6. The method of any one of the preceding examples, further comprising:

[0130] - providing (3045), to the cellular network, an indication of at least one of an uncertainty (699-1) or a probability of the prediction.

[0131] EXAMPLE 7. The method of any one of the preceding examples, wherein the wireless communication device monitors for the one or more reference signals as part of an initial access beam management procedure for connecting to the base station.

[0132] EXAMPLE 8. The method of any one of the preceding examples, further comprising:

[0133] - inferring the pre-defined model, wherein the pre-defined model is a deep neural network wherein, for said inferring, the pre-defined model is provided with input data that is determined based on said monitoring, wherein the input data comprises at least one of received amplitude values or received phase values of the one or more reference signals for each of the multiple transmit beams. wherein the input data comprises the at least one of the received amplitude values or the received phase values of the one or more reference signals for each of multiple subcarriers of a multicarrier transmission of the base station. EXAMPLE 9. The method of any one of the preceding examples, further comprising:

[0134] - selecting the pre-defined model from multiple candidate pre-defined models based on at least one of the one or more spatial characteristics of the multiple transmit beams or a type of the prediction.

[0135] EXAMPLE 10. The method of any one of the preceding examples, wherein the one or more channel characteristics are limited to the gain of the path of the radio channel accessed by the further transmit beam of the base station (101).

[0136] Although the disclosure has been shown and described with respect to certain preferred embodiments, equivalents and modifications will occur to others skilled in the art upon the reading and understanding of the specification. The present disclosure includes all such equivalents and modifications and is limited only by the scope of the appended claims. For illustration, various scenarios have been disclosed in which a predefined model - e.g., a DNN - is inferred at the UE. However, similar techniques as disclosed herein may also be applicable to scenarios in which the predefined model is inferred at the cellular network, e.g., at the BS, based on a measurement report on one or more properties of the DL reference signals transmitted by the BS. For instance, the UE may provide a measurement report comprising an encoded representation of received amplitude values and / or received phase values. Such encoded representation may be in a latent space accessed by an encoder-type DNN. The BS may then execute a decoder-type DNN, wherein the encoder-type DNN and the decoder-type DNN are trained end-to-end. Such scenario corresponds to a distributed inference, both at the UE as well as the BS. By transmitting an encoded representation in the latent space (feature embedding), the control-signalling overhead can be reduced. Alternatively or additionally, the UE can also offload inference of the predefined model to a cloud computing node, e.g., situated in the Internet.

[0137] For further illustration, various scenarios have been disclosed for the management of TX beams at the BS. Similar techniques may also be applicable for be management for RX beams at the BS. In such a scenario, it would be possible to select an RX beam based on the techniques disclosed herein assuming channel reciprocity. Alternatively or additionally, it would also be possible to employ UL reference signals transmitted by the UE. Then, a predefined model may be inferred at the BS. More generally, the techniques disclosed herein may be applicable to various kinds and types of communication nodes of a wireless communication system. For instance, device-to-device or sidelink radio channels may be sounded for respective beam-management procedures.

[0138] For still further illustration, techniques have been disclosed in which one or more channel characteristics of a radio channel are predicted. It would be possible that such prediction is made per sub-band of the radio channel. I.e., it would be possible to make multiple predictions of one or channel characteristics, one for each sub-band of the radio channel.

Claims

C L A I M S1. A method for use in a wireless communication device (102) connectable to a cellular network, the method comprising:- monitoring (3020), using at least one receive beam, for one or more reference signals transmitted by a base station 8101) of the cellular network using multiple transmit beams (181, 182, 183, 184, 186, 187, 188, 190, 199, 701, 702, 703, 704, 711, 712, 713, 714),- establishing (3025, 3026, 3027) one or more spatial characteristics (281, 282, 286, 290, 299, 300, 381 , 386, 390) of the multiple transmit beams,- communicating (3040, 3041 , 3042) with the cellular network in accordance with a prediction of one or more channel characteristics of a radio channel from the base station (101) to the wireless communication device (102), the prediction of the one or more channel characteristics being obtained from a pre-defined model (920), the prediction of the one or more channel characteristics being based on the one or more spatial characteristics and further being based on said monitoring (3020).

2. The method of claim 1 , wherein said communicating (3040) with the cellular network comprises providing to the cellular network an indication of the one or more channel characteristics.

3. The method of claim 2, further comprising:- obtaining (3015), from the cellular network, a configuration of a reporting scheme for providing the indication of the one or more channel characteristics.

4. The method of claim 1, wherein said communicating with the cellular network comprises selectively connecting (3041) to the base station (101) depending on the one or more channel characteristics.

5. The method of any one of claims 1 to 4, wherein said establishing of the one or more spatial characteristics comprises:- obtaining, from the cellular network, an indication of the one or more spatial characteristics.

6. The method of any one of claims 1 to 4, wherein the one or more spatial characteristics are established in accordance with a predefined ruleset of a radio-access communication protocol of the cellular network.

7. The method of any one of the preceding claims, wherein the one or more channel characteristics comprises a gain of a path of the radio channel accessed by a further transmit beam.

8. The method of claim 7, wherein the gain of the path of the radio channel is a globally optimal gain of the radio channel.

9. The method of claim 7, wherein the gain of the path of the radio channel is a locally optimal gain of the radio channel.

10. The method of any one of claims 7 to 9,wherein the one or more channel characteristics does not comprise one or more spatial characteristics of the further transmit beam, and wherein the one or more channel characteristics are optionally limited to the gain of the path of the radio channel accessed by the further transmit beam of the base station (101).

11. The method of any one of the preceding claims, wherein the one or more channel characteristics comprises a direction of a further transmit beam.

12. The method of any one of claims 1 to 11 , wherein the multiple transmit beams locally sample a subarea of a directional space accessible by beamforming at the base station.

13. The method of any one of claims 1 to 11 , wherein the multiple transmit beams globally sample a directional space accessible by beamforming at the base station (101).

14. The method of any one of claims 1 to 13, wherein a data connection is active between the base station (101) and the wireless communication device (102) while monitoring for the one or more reference signals.

15. The method of claim 14, wherein the base station (101) transmits signals encoding data associated with the data connection to the wireless communication device (102) using a serving transmit beam (170), wherein multiple transmit beams have beam widths in the range of 50% - 150% of the beam width of the serving transmit beam (170).

16. The method of any one of the preceding claims, further comprising:- providing (3045), to the cellular network, an indication of at least one of an uncertainty (699-1) or a probability of the prediction.

17. The method of any one of the preceding claims, wherein the wireless communication device (102) monitors for the one or more reference signals (4015, 4115) as part of a beam-refinement procedure for refining a serving transmit beam (170) used by the base station (101) for serving the wireless communication device (102).

18. The method of any one of claims 1 to 17, wherein a data connection is not active between the base station (101) and the wireless communication device (102) while monitoring for the one or more reference signals (4015, 4115).

19. The method of any one of the preceding claims, wherein the wireless communication device monitors for the one or more reference signals as part of an initial access beam management procedure for connecting to the base station.

20. The method of any one of the preceding claims, further comprising:- inferring the pre-defined model, wherein the pre-defined model is a deep neural network.

21. The method of claim 20, wherein, for said inferring, the pre-defined model is provided with input data that is determined based on said monitoring.

22. The method of claim 21 , wherein the input data comprises at least one of received amplitude values or received phase values of the one or more reference signals for each of the multiple transmit beams.

23. The method of claim 22, wherein the input data comprises the at least one of the received amplitude values or the received phase values of the one or more reference signals for each of multiple subcarriers of a multicarrier transmission of the base station.

24. The method of claim 22 or 23, wherein the input data comprises at least one of received amplitude values or received phase values for each of the at least one receive beam.

25. The method of any one of claims 20 to 24, wherein, for said inferring, the pre-defined model is provided with input data that comprises an indication of the one or more spatial characteristics.

26. The method of any one of the preceding claims, further comprising:- selecting the pre-defined model from multiple candidate pre-defined models based on at least one of the one or more spatial characteristics of the multiple transmit beams or a type of the prediction.

27. The method of any one of the preceding claims, further comprising:- providing (3010), to the cellular network, a request for a beam refinement procedure, wherein said monitoring is executed upon said providing of the request.

28. The method of any one of the preceding claims, further comprising:- providing (3005), to the cellular network, an indication of an ability of the wireless communication device (102) to infer the pre-defined model (920), wherein said monitoring is executed upon said providing 83005) of the indication of the ability of the wireless communication device (102) to infer the pre-defined model.

29. The method of any one of the preceding claims, wherein the at least one receive beam comprises multiple receive beams of a receive beam sweep.

30. The method of any one of the preceding claims, wherein the one or more spatial characteristics of the multiple transmit beams comprise spatial information for each of the multiple transmit beams.

31. The method of any one of the preceding claims, wherein the one or more spatial characteristics of the multiple transmit beams comprises directions of each of the multiple transmit beams.

32. The method of any one of the preceding claims, wherein the one or more spatial characteristics of the multiple transmit beams comprises beam widths of each of the multiple transmit beams.

33. The method of any one of the preceding claims, wherein the one or more spatial characteristics of the multiple transmit beams comprises a relative arrangement of the multiple transmit beams with respect to each other.

34. The method of any one of the preceding claims,wherein the prediction comprises multiple channel characteristics associated with different sub-bands of a communication radio channel used for communication between the base station and the wireless communication device.

35. The method of any one of the preceding claims, wherein the one or more channel characteristics are frequency-resolved.

36. A method for use in a base station (101) of a cellular network, the method comprising:- obtaining, from a wireless communication device (102) connectable to the cellular network, an indication of a prediction of one or more channel characteristics of a radio channel from the base station (101) to the wireless communication device (102), the prediction of the one or more channel characteristics being obtained from a pre-defined model (920), the prediction of the one or more channel characteristics being based on a one or more spatial characteristics of multiple transmit beams used by the base station or a further base station for transmitting one or more reference signals and further being based on monitoring, at the wireless communication device, of the one or more reference signals.

37. The method of claim 36, further comprising:- executing a beam refinement based on the prediction of the one or more channel characteristics.

38. The method of claim 36, further comprising:- triggering a handover to the further base station based on the prediction of the one or more channel characteristics.

39. The method of any one of claims 36 to 38, further comprising:- obtaining, from the wireless communication device, an indication of at least one of an uncertainty (699-1) or a probability of the prediction.

40. The method of claim 39, further comprising:- executing a beam refinement based on the prediction of the one or more channel characteristics and further based on the at least one of the uncertainty or the probability of the prediction.

41. The method of claim 40, further comprising:- triggering a handover to the further base station based on the prediction of the one or more channel characteristics and further based on the at least one of the uncertainty or the probability of the prediction.

42. A deep neural network for use in a beam management procedure, the deep neural network obtaining, as an input, one or more receive properties of one or more reference signals transmitted by a transmitter node using multiple transmit beams, and further obtaining, as a further input, a one or more spatial characteristics of the multiple transmit beams.

43. A method, comprising:- obtaining a prediction of a one or more channel characteristics of a radio channel,- obtaining an uncertainty of the prediction, and- executing a beam management based on the prediction and the uncertainty of the prediction.

44. A wireless communication device comprising compute circuitry configured to perform the method of any one of claims 1 to 35.

45. A base station comprising compute circuitry configured to perform the method of any one of claims 36 to 41.

46. A system comprising the wireless communication device of claim 44 and the base station of claim 45.

47. A method for use in a wireless communication device (102) connectable to a cellular network, the method comprising:- receiving (3020), using at least one receive beam, one or more reference signals transmitted by a base station 8101) of the cellular network using multiple transmit beams (181, 182, 183, 184, 186, 187, 188, 190, 199, 701, 702, 703, 704, 711 , 712, 713, 714), - establishing (3025, 3026, 3027) one or more spatial characteristics (281, 282, 286,290, 299, 300, 381 , 386, 390) of the multiple transmit beams,- communicating (3040, 3041 , 3042) with the cellular network in accordance with a prediction of one or more channel characteristics of a radio channel from the base station (101) to the wireless communication device (102), the prediction of the one or more channel characteristics being obtained from a pre-defined model (920), the prediction of the one or more channel characteristics being based on the one or more spatial characteristics and further being based on said monitoring (3020).

Images