WO2023198265A1

WO2023198265A1 - Traffic load distribution based on predicted client device trajectories

Info

Publication number: WO2023198265A1
Application number: PCT/EP2022/059554
Authority: WO
Inventors: Stephen MWANJE; Borislava GAJIC; Márton KAJÓ
Original assignee: Nokia Solutions And Networks Oy
Priority date: 2022-04-11
Filing date: 2022-04-11
Publication date: 2023-10-19

Abstract

Traffic load distribution based on predicted client device trajectories is disclosed. A first network node device obtains current traffic information related to one or more client devices in one or more cells in a radio access network. The current traffic information comprises information on current trajectories of the one or more client devices in the one or more cells. The first network node device utilizes the obtained current traffic information in traffic load distribution by applying a machine learning model to the obtained information on the current trajectories to predict future trajectories of the one or more client devices in the one or more cells.

Description

TRAFFIC LOAD DISTRIBUTION BASED ON PREDICTED CLIENT DEVICE

TRAJECTORIES

TECHNICAL FIELD

The disclosure relates generally to communications and, more particularly but not exclusively, to traf fic load distribution based on predicted client device traj ectories .

BACKGROUND

Cognitive autonomous networks ( CAN) , for example , may provide intelligence and autonomy in network operations , administration and management ( CAM) to support the increasing flexibi lity and complexity of radio access networks (RANs ) . One use case for such cognitive automation or automated decision making in the RAN is traf fic steering ( TS ) or distribution of load among cells .

At least some of the exi sting TS solutions may select values for TS control parameters ( typically handover settings ) for a given pair of cells by considering load and handover characteristics of the cells . For example , handover triggers may be controlled by selecting appropriate parameters . Selection of such parameters needs to be done carefully in order to improve network performance .

However, at least in some situations the existing TS solutions may not take into account user tra ectories , i . e . , the selected handover trigger settings may not be speci fic to a given cell-pair boundary and the traj ectories the users take across the cell boundaries .

SUMMARY

The scope of protection sought for various example embodiments of the invention is set out by the independent claims . The example embodiments and features , if any, described in this speci fication that do not fall under the scope of the independent claims are to be interpreted as examples useful for understanding various example embodiments of the invention . An example embodiment of a first network node device comprises at least one processor, and at least one memory including computer program code . The at least one memory and the computer program code are configured to , with the at least one processor, cause the first network node device at least to perform obtaining current traf fic information related to one or more client devices in one or more cells in a radio access network . The current traf fic information comprises information on current traj ectories of the one or more client devices in the one or more cells . The at least one memory and the computer program code are further configured to , with the at least one processor, cause the first network node device at least to perform utili zing the obtained current traf fic information in traf fic load distribution . The utili zing of the obtained current traf fic information in the traf fic load distribution comprises applying a machine learning model to the obtained information on the current traj ectories to predict future traj ectories o f the one or more client devices in the one or more cells .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to predict how the traf fic load in the one or more cells i s likely to change in view of the predicted future traj ectories of the one or more client devices in the one or more cells .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to identi fy cells likely to experience congestion .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to determine one or more handover options resolving the congestion identi fied likely to be experienced . In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to select the client devices of the one or more cl ient devices whose serving cells need to be adj usted in order to minimi ze the likelihood of the congestion resulting from the predicted future traj ectories of the one or more client devices and the predicted likely traf fic load change .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to select one or more of the one or more client devices for handover .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to determine at least one of : mappings between the predicted future traj ectories and cells available along the predicted future traj ectories , mappings between the predicted future traj ectories and the traf fic load distribution in the one or more cells , or mappings between the predicted future traj ectories and coverages of the one or more cells or corresponding cell layers .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in traf fic load distribution comprises at least one of : utili zing the obtained current traf fic information in traf fic load distribution for the one or more client devices in the one or more cells , or uti li zing the obtained current traffic information in traf fic load distribution towards a cell of the one or more cells for the one or more client devices and their supported services . In an example embodiment , alternatively or in addition to the above-described example embodiments , the at least one memory and the computer program code are further configured to , with the at least one processor, cause the first network node device to perform receiving past traffic information related to the one or more client devices in the one or more cells . The past traf fic information comprises information on past traj ectories of the one or more client devices in the one or more cells .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the at least one memory and the computer program code are further configured to , with the at least one processor, cause the first network node device to perform transmitting the received past traf fic information to a second network node device for use as training data .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the at least one memory and the computer program code are further configured to , with the at least one processor, cause the first network node device to perform receiving the machine learning model from the second network node device after the machine learning model has been trained by the second network node device .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the at least one memory and the computer program code are further configured to , with the at least one processor, cause the first network node device to perform training the machine learning model to identi fy li kely traj ectories of the one or more cl ient devices in the one or more cells by feeding the received past traf fic information to the machine learning model .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the information on the current traj ectories comprises at least one of a sequence of positioning coordinates of the one or more client devices measured at a fixed frequency, or a sequence of traf fic measurement signals generated by the one or more client devices .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the traf fic measurement signals comprise at least one of reference signal received power signals or signal-to-interf erence-plus-noise ratio signals .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the first network node device comprises a base station .

An example embodiment of a first network node device comprises means for performing obtaining current traf fic information related to one or more client devices in one or more cells in a radio access network . The current traf fic information comprises information on current traj ectories of the one or more cl ient devices in the one or more cells . The means are further configured to perform utili zing the obtained current traf fic information in traf fic load distribution . The utili zing of the obtained current traf fic information in the traffic load distribution comprises applying a machine learning model to the obtained information on the current traj ectories to predict future traj ectories of the one or more client devices in the one or more cells .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traf fic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to determine one or more handover options resolving the congestion identi fied likely to be experienced .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to select the client devices of the one or more cl ient devices whose serving cells need to be adj usted in order to minimi ze the likelihood of the congestion resulting from the predicted future traj ectories of the one or more client devices and the predicted likely traf fic load change .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in traf fic load distribution comprises at least one of : utili zing the obtained current traf fic information in traf fic load distribution for the one or more client devices in the one or more cells , or uti li zing the obtained current traffic information in traf fic load distribution towards a cell of the one or more cells for the one or more client devices and their supported services .

In an example embodiment, alternatively or in addition to the above-described example embodiments , the means are further configured to perform causing receiving of past traf fic information related to the one or more client devices in the one or more cells . The past traf fic information comprises information on past traj ectories of the one or more client devices in the one or more cells .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the means are further configured to perform causing transmitting of the received past traf fic information to a second network node device for use as training data .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the means are further configured to perform causing receiving of the machine learning model from the second network node device after the machine learning model has been trained by the second network node device .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the means are further configured to perform training the machine learning model to identi fy likely traj ectories of the one or more client devices in the one or more cells by feeding the received past traf fic information to the machine learning model .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the traf fic measurement signals comprise at least one of reference signal received power signals or signal-to-interf erence-plus-noise ratio signals . In an example embodiment , alternatively or in addition to the above-described example embodiments , the first network node device comprises a base station .

An example embodiment of a method comprises obtaining, by a first network node device , current traf fic information related to one or more client devices in one or more cells in a radio access network . The current traf fic information comprises information on current traj ectories of the one or more client devices in the one or more cell s . The method further comprises utili zing, by the first network node device , the obtained current traf fic information in traf fic load distribution . The utili zing of the obtained current traf fic information in the traf fic load distribution comprises applying a machine learning model to the obtained information on the current traj ectories to predict future traj ectories of the one or more client devices in the one or more cells .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current traj ectories to determine one or more handover options resolving the congestion identi fied likely to be experienced .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the method further comprises receiving, at the first network node device , past traf fic information related to the one or more client devices in the one or more cells . The past traf fic information comprises information on past traj ectories of the one or more client devices in the one or more cells .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the method further comprises transmitting the received past traf fic information from the first network node device to a second network node device for use as training data .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the method further comprises receiving, at the first network node device , the machine learning model from the second network node device after the machine learning model has been trained by the second network node device .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the method further comprises training, by the first network node device , the machine learning model to identi fy likely traj ectories of the one or more client devices in the one or more cells by feeding the received past traf fic information to the machine learning model .

An example embodiment of a computer program comprises instructions for causing a first network node device to perform at least the following : obtaining current traf fic information related to one or more client devices in one or more cells in a radio access network, the current traf fic information compri sing information on current traj ectories of the one or more client devices in the one or more cells ; and utili zing the obtained current traf fic information in traf fic load distribution . The utili zing of the obtained current traf fic information in the traf fic load distribution comprises applying a machine learning model to the obtained information on the current traj ectories to predict future traj ectories of the one or more client devices in the one or more cells .

An example embodiment of a second network node device comprises at least one processor, and at least one memory including computer program code . The at least one memory and the computer program code are configured to , with the at least one processor, cause the second network node device at least to perform receiving past traf fic information related to one or more client devices in one or more cells in a radio access network . The past traf fic information comprises information on past traj ectories of the one or more client devices in the one or more cells . The at least one memory and the computer program code are further configured to , with the at least one proces sor, cause the second network node device at least to perform training a machine learning model to identi fy likely traj ectories of the one or more cl ient devices in the one or more cell s by feeding the received past traf fic information to the machine learning model .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the at least one memory and the computer program code are further configured to , with the at least one processor, cause the second network node device at least to perform transmiting the trained machine learning model to a first network node device .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the information on the past traj ectories comprises at least one of a sequence of positioning coordinates of the one or more client devices measured at a fixed frequency, or a sequence of traf fic measurement signals generated by the one or more client devices .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the past traf fic information comprises traf fic information at least one of before or after at least one past handover event .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the second network node device comprises an operations , administration and management unit or a base station .

An example embodiment of a second network node device comprises means for performing causing receiving of past traf fic information related to one or more client devices in one or more cells in a radio access network . The past traffic information comprises information on past traj ectories of the one or more cl ient devices in the one or more cells . The means are further configured to perform training a machine learning model to identi fy li kely traj ectories of the one or more cl ient devices in the one or more cells by feeding the received past traf fic information to the machine learning model .

In an example embodiment , alternatively or in addition to the above-described example embodiments , the means are further configured to perform caus ing transmiting of the trained machine learning model to a first network node device .

An example embodiment of a method comprises receiving, at a second network node device , past traf fic information related to one or more client devices in one or more cells in a radio access network . The past traf fic information comprises information on past traj ectories of the one or more client devices in the one or more cell s . The method further comprises training, by the second network node device , a machine learning model to identi fy likely traj ectories of the one or more client devices in the one or more cells by feeding the received past traf fic information to the machine learning model .

In an example embodiment , alternatively or in addition to the above-described example embodiments , method further comprises transmitting the trained machine learning model from the second network node device to a first network node device .

An example embodiment of a computer program comprises instructions for causing a second network node device to perform at least the following : receiving past traf fic information related to one or more client devices in one or more cells in a radio access network, the past traf fic information compri sing information on past traj ectories of the one or more client devices in the one or more cells ; and training a machine learning model to identi fy likely traj ectories of the one or more client devices in the one or more cells by feeding the received past traf fic information to the machine learning model .

DESCRIPTION OF THE DRAWINGS

The accompanying drawings , which are included to provide a further understanding of the embodiments and constitute a part of this speci fication, illustrate embodiments and together with the description help to explain the principles of the embodiments . In the drawings :

FIG . 1 shows an example embodiment of the subj ect matter described herein illustrating an example system, where various embodiments of the present disclosure may be implemented;

FIG . 2A shows an example embodiment of the subj ect matter described herein illustrating a first network node device ;

FIG . 2B shows an example embodiment of the subj ect matter described herein illustrating a second network node device ;

FIG . 3 shows an example embodiment of the subj ect matter described herein illustrating traj ectory aware cell traf fic steering through handovers ;

FIG . 4 shows an example embodiment of the subj ect matter described herein illustrating generating labelled data and loss functions from traj ectory and signal profiles ;

FIG . 5 shows an example embodiment of the subj ect matter described herein illustrating machine learning model training;

FIG . 6 shows another example embodiment of the subj ect matter described herein illustrating machine learning model training;

FIG . 7 shows an example embodiment of the subj ect matter described herein illustrating an example topology o f a sequence-processing convolutional neural network;

FIG . 8A shows an example embodiment of the subj ect matter described herein illustrating inference using a traj ectory of a single client device ;

FIG . 8B shows an example embodiment of the subj ect matter described herein illustrating inference using RSRP or S INR sequences of a single client device measured on multiple cells ;

FIG . 80 shows an example embodiment of the subj ect matter described herein illustrating inference using traj ectories and RSRP or S INR sequences of a single client device ;

FIG . 9 shows an example embodiment of the subj ect matter described herein illustrating a method; and FIG. 10 shows an example embodiment of the subject matter described herein illustrating another method.

Like reference numerals are used to designate like parts in the accompanying drawings.

DETAILED DESCRIPTION

Reference will now be made in detail to embodiments, examples of which are illustrated in the accompanying drawings. The detailed description provided below in connection with the appended drawings is intended as a description of the present examples and is not intended to represent the only forms in which the present example may be constructed or utilized. The description sets forth the functions of the example and the sequence of steps for constructing and operating the example. However, the same or equivalent functions and sequences may be accomplished by different examples.

Fig. 1 illustrates an example system 100, where various embodiments of the present disclosure may be implemented. The system 100 may comprise a fifth generation (5G) new radio (NR) network that may include one or more radio access networks 110, such as one or more open radio access networks (O-RANs) . An example representation of the system 100 is shown depicting a client device 120 and network node devices 200i, 2002, 200₃ in cells 111, 112, 113, respectively. Any or all of the network node devices 200i, 2OO2, 200₃ may correspond with a first network node device 200 of Fig. 2A. At least in some embodiments, the 5G NR network may comprise one or more massive machine-to-ma- chine (M2M) network (s) , massive machine type communications (mMTC) network(s) , internet of things (ToT) network(s) , industrial internet-of-things (IIoT) network(s) , enhanced mobile broadband (eMBB) network (s) , ultra-reliable low-latency communication (URLLC) network(s) , and/or the like. In other words, the 5G NR network may be configured to serve diverse service types and/or use cases, and it may logically be seen as comprising one or more networks . The system 100 may further comprise a second network node device 210 described in more detail below. The second network node device 210 may comprise an operations, administration and management (OAM) entity or a base station.

At least some of the disclosed embodiments may be implemented in an 0-RAN architecture. The 0-RAN aims for interoperability and standardization of RAN elements including a unified interconnection standard for network functions from different vendors. The 0-RAN architecture provides a foundation for building a virtualized RAN on open hardware with an embedded artificial intelligence (Al) -powered radio control.

The one or more client devices 120 may include, e.g., a mobile phone, a smartphone, a tablet computer, a smart watch, or any hand-held, portable and/or wearable device. The client device (s) 120 may also be referred to as a user equipment (UE) . The network node devices 200i, 2OO2, 200₃ may comprise a base station. The base station may include, e.g., a fifth-generation base station (gNB) or any such device suitable for providing an air interface for client devices to connect to a wireless network via wireless transmissions.

In the following, various example embodiments will be discussed. At least some of these example embodiments may allow traffic load distribution based on predicted client device tra- j ectories .

As discussed above, the distribution of load among cells may also be referred to as traffic steering (TS) . Radio access technology (RAT) refers to an underlying physical connection method for a radio-based communication network. RATs include, e.g., bluetooth, Wi-Fi, long-term evolution (LTE) , second-generation cellular network (2G) , third-generation cellular network (3G) , fourth-generation cellular network (4G) , and 5G NR.

In the following, the terms RAT and layer are used interchangeably. Furthermore, as used herein, the term 'cell' may also include 'cell layer' .

Traffic steering may be applied both to intra-RAT and inter-RAT scenarios. For the inter-RAT scenario, inter-RAT handovers (HOs) may be optimized to manage the traffic distribution. For 5G NR, an inter-RAT event Bl has been defined which is an event in which an inter-RAT neighbor becomes better than a given threshold. A client device may enter the event Bl once the following condition is satisfied:

Mn + Ofn + Ocn - Hys > Thresh in which:

- Mn is a measurement result of an inter-RAT neighbor cell, not taking into account any offsets;

- Ofn is a measurement object specific offset of a frequency of the inter-RAT neighbor cell (i.e., eutra-Q-Off- setRange as defined within measObj ectEUTRA corresponding to the frequency of the neighbor inter-RAT cell, utra-FDD-Q-Off setRange as defined within measObj ectUTRA-FDD corresponding to the frequency of the neighbor inter-RAT cell) ;

- Ocn is a cell specific offset of the inter-RAT neighbor cell (i.e., celllndividualOffset as defined within the mea- sObj ectEUTRA corresponding to the neighbor inter-RAT cell) , and set to zero if not configured for the neighbor cell;

- Hys is a hysteresis parameter for this event; and

- Thresh is a threshold parameter for this event.

In an example scenario in the system 100 of Fig. 1, there may be overload in the cell 111 with the client device 120 moving on route A near the cell border. In this example, the client device 120 is most likely to stay in cell 112 for a very limited time, as most of the client devices moving on route A continue to route B and not route C. At least some of the following example embodiments may allow identifying the trajectories on which the client devices are likely to move along and choose (e.g., adjust handover settings for) the cells which the client devices are more likely to end up in. In the case of the example of Fig. 1, at least some of the following example embodiments may determine that the client device 120 is more likely to move to the cell 113 and choose handover settings that move the client device 120 to the cell 113 and thus shift the traffic load to the cell 113 instead of shifting the traffic load to the cell 112.

In other words, at least some of the following example embodiments may allow traj ectory- aware traffic steering among the cells. Furthermore, at least some of the following example embodiments may allow machine learning (ML) -based load distribution optimization that learns to distribute load in cells depending on the predicted trajectory of client devices in one or more cells.

For example, at least some of the following example embodiments may allow learning the trajectories of client devices in a network or a set of cells and the related probability of a client device moving along a particular trajectory given a previous trajectory over a short time period. Thereby, at least some of the following example embodiments may allow training a machine learning model that learns the client device trajectories and client device -mobility profiles based on, e.g., global positioning system (GPS) locations recorded and submitted by the client devices as well as historical radio signal receive power (RSRP) or signal-to-interf erence-plus-noise ratio (SINR) measurements. Herein, the term mobility profile here refers to a combination of speed and general or specific direction, e.g., 30 kilometers per hour (km/h) and straight through a junction x, or 20 km/h and turning to the right at the junction x.

For example, at least some of the following example embodiments may allow training a machine learning model that learns the network load distribution towards any cell or cell layer (RAT) for given client devices and their supported services.

For example, at least some of the following example embodiments may allow training a machine learning model that learns the mapping between the client device tra j ectory/learned trajectories and learned load distribution in different cells and cell layers. Based on the learned trajectories, the trained ML model may be used to predict the trajectories of a set of client devices given past trajectories, past RSRP or SINR measurement sequences, and a starting point. The predicted trajectories may be expressed in the form of a probability of a client device following certain trajectory. The trajectory may be expressed, e.g., in terms of a sequence of GPS points or sequence of RSRPs in particular cells / cell layers (RATs) . The predicted trajectories may be mapped to the cell / cell layer coverage in order to assess the availability of a certain cell/cell layer over the entire trajectory.

For example, at least some of the following example embodiments may allow predicting how the load in set of cells or cell layers is likely to change given the predicted trajectories of the client devices in those cells and their neighbors.

For example, at least some of the following example embodiments may allow predicting which cells / cell layers are likely to experience congestion if the client device follows the predicted trajectories.

For example, at least some of the following example embodiments may allow selecting the set of client devices whose serving cells needs to be adjusted in order to minimize the likelihood of congestion as a result of the predicted trajectories of the client devices and predicted load. The selection of the client devices and their cells may (besides radio conditions and trajectory) also depend on the type of the application and its expected Quality of Experience (QoE) as well as on the RAT.

For example, at least some of the following example embodiments may allow controlling the congestions by deciding when and by how much to adjust the handover trigger points (e.g., frequency offsets and cell individual offsets) among a set of cells / cell layers (RATs) and client devices to effect congestion control related to the predicted trajectories of the client devices. For example, at least some of the following example embodiments may allow controlling single layer intra handover settings as well as controlling an inter-RAT HO among the cells and influencing the load distribution among the inter-RAT layers .

Diagram 300 of Fig. 3 shows an example embodiment of the subject matter described herein illustrating trajectory aware cell traffic steering through intra-RAT (via cell-individual offset (CIO) and time-to- trigger (TTT) ) and inter-RAT (via Ofn) handovers. Diagram 300 includes an input data set 301 (including, e.g., trajectories and/or signal sequences) , a trajectory ML model 302, an intermediate data set 303, a load predictor 304, a load distribution agent 305, and an output data set 306. In practice, the load predictor 304 and the load distribution agent 305 may be combined into a single ML model, such as the load prediction and distribution ML model 610 of diagram 600 of Fig. 6.

For training purposes, the model of Fig. 3 may take anonymized data 301 on location -specific mobility and HO events of the client device (s) 120, including inter-RAT handovers. To achieve this, the client device (s) 120 may be configured to log location tagged handover data to be used for training the ML model. In addition, the client device (s) 120 may be configured to log the information on RSRP or SINR measurements for available cell layers / RATs. The client device (s) 120 may log and anonymize the location-tagged data before forwarding the data to the first network node device 200 so that the first network node device 200 (or the second network node device 210, depending on the embodiment) may undertake the training. The anonymization may not only hide the identity of the client device and its user but may also ensure that it is impossible to track the client device (s) , e.g., by sending the location-identified reports at a time that is randomly different from the time at which the report was compiled.

At least in some embodiments, the generation of the labelled data to be used for training the ML model may be automated. For a given sequence of locations and RSRP or SINR measurements (including measurements on different frequency layers) , the labeling may split the sequence into an input portion and an output portion to be used as a label during the training. The length of the label portion may be selected in consideration of the expected lengths for which the prediction is to be done.

The labeled data may be used to train a trajectory ML model 302 to concurrently:

- learn possible trajectories / trajectories across the handover boundaries of a plurality of cells,

- learn a model of the radio conditions to capture the likely RSRPs/SINRs 308 for any available cell layer / RAT at any one location in the radio access network 110,

- learn a load distribution for specific cells and cell layers / RATs,

- learn a mapping between locations or trajectories / trajectories and the load for specific cells and cell layers / RATs, and - learn a mapping between the trajectories and the coverage of each of the different cell layers/RATs.

The machine learning model may be split into multiple models, one for each learning challenge. The outcome of the training may be a cell-specific, cell-boundary-specific or cellcluster-specific, or cell layer / RAT specific machine learning model, e.g., a neural network model.

For inference, if the TS is evaluated concurrently for multiple cells, a trajectory (e.g., a sequence of GPS coordinates measured at a fixed frequency) may be received from each of a number of client devices 120 in the plurality of cells 111, 112, 113 for which the TS is to be evaluated. In a case in which the TS is evaluated in a single cell, the data may be received from the client devices within the respective cell. Then, the predicted future locations, the respective load on available cell / cell layers and their neighbors, the probability of congestion in different cells / cell layers, the probability of a particular RAT / layer not being accessible to the client device on the predicted trajectory, as well as appropriate CIO values, TTT values, and updates to cell layer / RAT preference configurations (e.g., selecting the layers with high predicted availability on the predicted trajectory) to be applied in the cell (s) may be computed, while also minimizing the number of handovers needed to diminish the congestion.

The inference may also be based on RSRP or SINR measurements related to available frequency layers.

For example, the client device (s) 120 may be signaled to log the required data, i.e., the mapping (e.g., via time stamps) between location, RSRP or SINR measurements, QCI and the cell layer / RAT, and/or to send the collected training data to the network for training the model.

Fig. 2A is a block diagram of the first network node device 200, in accordance with an example embodiment.

The first network node device 200 comprises at least one processor 202 and at least one memory 204 including computer program code. The first network node device 200 may also include other elements, such as a transceiver configured to enable the first network node device 200 to transmit and/or receive information to/from other devices, as well as other elements not shown in Fig . 2A. In one example , the first network node device 200 may use the transceiver to transmit or receive signaling information and data in accordance with at least one cellular communication protocol . The transceiver may be configured to provide at least one wireless radio connection, such as for example a 3GPP mobile broadband connection ( e . g . , 5G) . The transceiver may be configured to be coupled to at least one antenna to transmit and/or receive radio frequency signals .

Although the first network node device 200 is depicted to include only one processor 202 , the first network node device 200 may include more processors . In an embodiment , the memory 204 is capable of storing instructions , such as an operating system and/or various applications . Furthermore , the memory 204 may include a storage that may be used to store , e . g . , at least some of the information and data used in the disclosed embodiments .

Furthermore , the processor 202 is capable of executing the stored instructions . In an embodiment , the processor 202 may be embodied as a multi-core processor, a single core processor, or a combination of one or more multi-core processors and one or more single core proces sors . For example , the processor 202 may be embodied as one or more of various processing devices , such as a coprocessor, a microprocessor, a controller, a digital signal processor ( DSP ) , a processing circuitry with or without an accompanying DSP, or various other processing devices including integrated circuits such as , for example , an application speci fic integrated circuit (AS IC ) , a field programmable gate array ( FPGA) , a microcontroller unit (MCU) , a hardware accelerator, a special-purpose computer chip, or the like . In an embodiment , the processor 202 may be configured to execute hard- coded functionality . In an embodiment , the processor 202 is embodied as an executor of software instructions , wherein the instructions may speci fically configure the processor 202 to perform the algorithms and/or operations described herein when the instructions are executed .

The memory 204 may be embodied as one or more volatile memory devices , one or more non-volatile memory devices , and/or a combination of one or more volatile memory devices and nonvolatile memory devices . For example , the memory 204 may be embodied as semiconductor memories (such as mask ROM, PROM (programmable ROM) , EPROM (erasable PROM) , flash ROM, RAM (random access memory) , etc.) .

The first network node device 200 may comprise a base station. The base station may include, e.g., a fifth-generation base station (gNB) or any such device providing an air interface for client devices to connect to the wireless network via wireless transmissions. At least in some embodiments, the first network node device 200 may comprise a multiple-input and multiple-output (MIMO) capable network node device.

The at least one memory 204 and the computer program code are configured to, with the at least one processor 202, cause the first network node device 200 at least to perform obtaining current traffic information related to the one or more client devices 120 in the one or more cells 111, 112, 113 in the radio access network 110.

The current traffic information comprises information on current trajectories of the one or more client devices 120 in the one or more cells 111, 112, 113. For example, the information on the current trajectories may comprise a sequence of positioning (e.g., global positioning system (GPS) ) coordinates of the one or more client devices 120 measured at a fixed frequency, and/or a sequence of traffic measurement signals generated by the one or more client devices 120. For example, the traffic measurement signals may comprise reference signal received power (RSRP) signals or signal-to-interf erence- plus-noise ratio (SINR) signals.

The at least one memory 204 and the computer program code are further configured to, with the at least one processor 202, cause the first network node device 200 at least to perform utilizing the obtained current traffic information in traffic load distribution.

The utilizing of the obtained current traffic information in the traffic load distribution comprises applying a machine learning model to the obtained information on the current trajectories to predict future trajectories of the one or more client devices 120 in the one or more cells 111, 112, 113. At least in some embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution may further comprise applying the machine learning model to the obtained information on the current traj ectories to predict how the traf fic load in the one or more cells 111 , 112 , 113 is likely to change in view of the predicted future traj ectories of the one or more client devices 120 in the one or more cells 111 , 112 , 113 .

At least in some embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution may further comprise applying the machine learning model to the obtained information on the current traj ectories to identi fy cells likely to experience congestion .

At least in some embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution may further comprise applying the machine learning model to the obtained information on the current traj ectories to determine one or more handover options resolving the congestion identi fied likely to be experienced . For example , the handover options may include controlling single layer intrahandover settings , controlling inter-RAT handover settings among the cells , and influencing load distribution among the inter- RAT layers .

At least in some embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution may further comprise applying the machine learning model to the obtained information on the current traj ectories to select the client devices of the one or more cl ient devices 120 whose serving cel ls need to be adj usted in order to minimi ze the likelihood of the congestion resulting from the predicted future traj ectories of the one or more client devices 120 and the predicted likely traf fic load change .

At least in some embodiments , the utili zing of the obtained current traf fic information in the traffic load distribution may further comprise applying the machine learning model to the obtained information on the current traj ectories to select one or more of the one or more client devices 120 for handover . At least in some embodiments, the utilizing of the obtained current traffic information in the traffic load distribution may further comprise applying the machine learning model to the obtained information on the current trajectories to determine mappings between the predicted future trajectories and cells available along the predicted future trajectories, to determine mappings between the predicted future trajectories and the traffic load distribution in the one or more cells 111, 112, 113, and/or to determine mappings between the predicted future trajectories and coverages of the one or more cells 111, 112, 113 or corresponding cell layers.

At least in some embodiments, the utilizing of the obtained current traffic information in the traffic load distribution may further comprise utilizing the obtained current traffic information in traffic load distribution for the one or more client devices 120 in the one or more cells 111, 112, 113, and/or utilizing the obtained current traffic information in traffic load distribution towards a cell of the one or more cells 111, 112, 113 for the one or more client devices 120 and their supported services.

In other words, for inference, if TS is evaluated concurrently for multiple cells and cell layers, an ML model may receive a trajectory (e.g., a sequence of GPS coordinates or the like measured at a fixed frequency) from a single client device (as illustrated in diagram 800A of Fig. 8A) which may then be evaluated for connecting or reconnecting to one of a plurality of candidate cells/cell layers. Diagram 800A includes an input data set 801A (including, e.g., trajectories) , a trajectory ML model 802A, an intermediate data set 803A, a load predictor 804A, a load distribution agent 805A, and an output data set 806A. For a single client device, e.g., the following may be inferred :

- the trajectory that the client device may exhibit in the future, and

- the mapping between the inferred future trajectory and the cell layers/RATs that will be available along that tra- j ectory .

Given the inferred information for a single client device, the ML model 802A may further be applied to evaluate multiple client devices at the same time. For a plurality of client devices, the ML model 802A may further infer the following : a hot-spot identification: which cell and cell layer/RAT is likely to experience the congestion at which point in time (given the expected number of client devices and corresponding services that will camp in that cell/cell layer at a certain point in time while following the predicted trajectory) ,

- a hot-spot resolution: derivation of intra-RAT and/or inter-RAT related handover decisions and parameter configurations, in relation with a running service / expected QoS of the client device. For example, the following may be decided (non- exhaustive list) :

- preparing an inter-RAT handover for video, bursty, or mobile broadband (MBB) traffic towards higher frequency RATs by adjusting the offset of the frequency of the inter-RAT neighbor cell,

- avoiding/minimizing an inter-RAT handover for URLLC traffic, with the client device staying at a RAT which has the availability over entire predicted trajectory, by adjusting the offset of the frequency of the inter-RAT neighbor cells such that inter-RAT handover is avoided, or

- keeping voice traffic at a lower frequency RAT, performing an intra-RAT handover of client devices based on the predicted load, and adjusting the CIO and TTT parameters accordingly.

As illustrated by diagram 800B of Fig. 8B, the same outcomes may be derived when only the signal sequences are available. Diagram 800B includes an input data set 801B (including, e.g., signal sequences) , a trajectory ML model 802B, an intermediate data set 803B, a load predictor 804B, a load distribution agent 805B, and an output data set 806B. For example, for client devices unwilling or unable to provide the GPS coordinates, the ML model 802B may use measurements from these client devices to derive the right TS configurations. An even better prediction may be achieved when both the trajectory and the signal sequences are available, as illustrated by diagram 800C of Fig. 8C. Diagram 800C includes an input data set 801C (in- eluding, e.g., trajectories and/or signal sequences) , a trajectory ML model 802C, an intermediate data set 803C, a load predictor 804C, a load distribution agent 805C, and an output data set 806C.

At least in some embodiments, the at least one memory 204 and the computer program code may be further configured to, with the at least one processor 202, cause the first network node device 200 to perform receiving past traffic information related to the one or more client devices 120 in the one or more cells 111, 112, 113. The past traffic information may comprise information on past trajectories of the one or more client devices 120 in the one or more cells 111, 112, 113.

In other words, to collect input data for training, selected client devices may report trajectory and other traffic- related properties (such as speed, applicable service, quality of service (QoS) / 5G QoS identifier (5QI) characteristics, and the like) , as well as the observed RSRP or SINR of visible cells, and optionally HO events to, e.g., the first network node device 200.

The client devices required to report do not need to include every client device, and may be randomly selected, helping anonymization. The reporting may also be an opt-out, but by default may be enabled on capable devices.

The report may be sent, e.g., when the client device is not actively in use, and with a delay between the recording of the report and its upload. This may for example be implemented by selecting a random time for sending the report. This may further enhance client device privacy, as an out-of-date client device location or trajectory is not as sensitive information as an up-to-date location / trajectory data.

The reports may be encrypted. As the communication of the report (uploading) is not time-sensitive, the encryption and the reporting may be done, e.g., in off-hours, when the client device / network is not heavily in use.

The speed and direction of the client devices may be accounted for. The sequence of, e.g., GPS points taken at a fixed frequency may identify both the trajectory on which the client device was moving and the speed with which it was moving. As such, capturing the GPS coordinates or the like at a fixed frequency may provide the data for learning the client device tra j ectories .

The RSRP and/or SINR values of the client devices may also be considered as the client devices move along the measured tra j ectories .

Other traffic characteristics of the client device (e.g., a heavy / light traffic client device) as well as characteristics and priorities of the services supported by the client device (e.g., URLLC vs. eMBB services) may also be considered .

The frequency of taking these measurements may be configured by the network differently for different client device locations. For example, near a highway where client devices are likely to move at high speed, a higher frequency may be needed to ensure a good characterization of the trajectory. On the other hand, a city's business district may have many trajectories close to each other, so a high frequency may also be needed there to distinguish the different trajectories. Conversely, a mountain resort area where most users of the client devices walk on foot and on a few trajectories may require only intermittent GPS records to adequately identify the trajectories and the proper settings on those trajectories.

At least in some embodiments, the trajectories and location coordinates may not be quantized. Rather, the exact values read off the GPS may be used leaving the selection of granularity to the ML model.

The reports may comprise the input data for training the model, which may include time series of the locations visited (e.g., GPS coordinates) before and after a handover, and corresponding measured RSRP and/or SINR of visible cells (including different RATs / cell layers) . The lengths of the time series both before and after the handover may be configured by the network and may be equal or otherwise. The time series may be supplemented with load-related information and the RSRP or SINR of the client device. The RSRP or SINR history may be used to estimate the RSRP or SINR of the client device in the new location as well as the load that the client device will likely present to each of the concerned cells and cell layers. At least in some embodiments, labelled data may be generated for the trajectory and RSRP/SINR prediction model. The speed and direction of the client devices may be taken into consideration. These may be estimated by the client device and sent to the network. However, to ease the job at the client device, the raw values that are measured by the client devices towards the network may be sent since the sequence of GPS points or the like taken at a fixed frequency identifies both the trajectory and the speed on which the client device was moving. Therefore, only the GPS coordinates at a fixed frequency may be captured and sent to the network.

Alternatively, the client device may collect GPS measurements or the like at a non-constant frequency, e.g., to give priority to other events (for instance call processing) in the client device, but with the GPS measurement time-stamped so that the network may determine the time between successive GPS measurements .

The RSRP or SINR measurement sequences for the serving and candidate target cells may be used to identify the likely RSRP or SINR at different points in the network. The RSRP or SINR measurements may be associated with the locations at which they are taken and submitted to the ML model during training. The ML model then learns to associate past sequences of locations and RSRP or SINR to future sequences of locations and RSRP or SINR.

At least in some embodiments, labelled data may be generated for the load prediction and distribution model. Diagram 400 of Fig. 4 shows an example embodiment of the subject matter described herein illustrating generating labelled data and loss functions from trajectory and signal profiles. Diagram 400 includes an input data set 401 (including, e.g., path data, input signal sequences, output signal sequences, and/or load sequences) , an output data set 406, and a rule set 407. Sequences of past and future RSRP or SINR measurements may be used to predict the likely best serving cell at the future location and the load that the client device is likely to induce in that cell. Combined with the service load in the candidate future cell, the ML model may predict whether the future cell is likely to be overloaded if the client device gets / stays connected to that cell.

To generate the training data, a data labelling module may evaluate the loss associated with different candidate TTT- ClO-Ofn combinations. The label generation is given the set of rules 407 for evaluating how good or bad given changes in cell load and handover events are, the quality being measured, e.g., in terms of a loss for each trajectory signal and recommendation combination. The loss may be computed on, e.g., whether the proposed change leads to a reduction or increase in traffic in one or more cells and whether the changes lead to unwanted handover events (such as handover failures, radio link failures, and/or ping pongs) . Then, for each CIO, TTT and cell layer preference configuration setting applied to the trajectory (and related RSRP or SINR sequences) , the resulting changes in cell traffic and handover events may be graded to compute the loss function. The end result may comprise a hash function of {trajectory + load + handover and cell layer preference settings: resulting loss} .

Although signal and position sequences 401 may be used to generate the labels for training, availability of handover events may help to improve the labelling. For example, special rare scenarios leading to specific handover events (e.g., a sharp blockage that causes handover failures) may not be easy to identify through reverse engineering signal profiles. In such a case, the handover event may be used to identify the point at which the event occurred to train the model to derive the settings that would avoid such an event.

At least in some embodiments, the at least one memory 204 and the computer program code may be further configured to, with the at least one processor 202, cause the first network node device 200 to perform transmitting the received past traffic information to the second network node device 210 for use as training data.

At least in some embodiments, the at least one memory 204 and the computer program code may be further configured to, with the at least one processor 202, cause the first network node device 200 to perform receiving the machine learning model from the second network node device 210 after the machine learning model has been trained by the second network node device 210.

At least in some embodiments, the at least one memory 204 and the computer program code may be further configured to, with the at least one processor 202, cause the first network node device 200 to perform training the machine learning model to identify likely trajectories of the one or more client devices 120 in the one or more cells 111, 112, 113 by feeding the received past traffic information to the machine learning model.

For example, for training a trajectory ML model, the training may be done, e.g., in a supervised-learning-form using the trajectories and the sequences of RSRP or SINR values before and after a HO event. Diagram 500 of Fig. 5 shows an example embodiment of the subject matter described herein illustrating training a trajectory ML model 502. Diagram 500 includes an input data set 501 (including, e.g., trajectories and/or signal sequences) , the trajectory ML model 502, an intermediate data set 503, a loss agent 508, and a loss 509. The client devices may send the data to the network (e.g., the first network node device 200) and the network may aggregate the data to perform batched learning.

Training for the trajectory prediction may be accomplished separately from training for the RSRP or SINR estimation .

The trajectory estimation ML model 502 may be a fingerprinting type solution that learns the fingerprints of the specific trajectories in the network. Then, given a sequence of locations in the input data set 501 the ML model 502 predicts the point (s) that is/are most likely to succeed the given sequence. The fingerprint may be extended to include cell layers besides the cells.

For example, for training a load prediction and distribution model, the training may be done, e.g., in a supervised- learning-form using the trajectory, the past and predicted sequences of RSRP or SINR values, and the load sequences in the cell (s) /cell layers. Diagram 600 of Fig. 6 shows another example embodiment of the subject matter described herein illustrating training a load prediction and distribution model 610. Diagram 600 includes an input data set 601 (including, e.g., path data, input signal sequences, output signal sequences, and/or load sequences) , the load prediction and distribution ML model 610, an intermediate data set 603, a loss agent 608, and a loss 609. The data may be collected as described above. Since the collected data may not include losses for all the possible configurations proposed by the ML model 610, a loss agent 608 may be needed to derive the loss 609 for each proposed configuration in consideration of the known losses.

The training for trajectory prediction and for load prediction and distribution may all be combined into a single model as discussed next.

Diagram 700 of Fig. 7 shows an example embodiment of the subject matter described herein illustrating an example topology of a sequence-processing convolutional neural network. Diagram 700 includes an input data set 701 (including, e.g., GPS locations and/or RSRP values for various cells at various timesteps) , convolution layers 702, 706, batch normalization layers 703, 707, 710, 713, rectified linear unit layers 704, 708, 711, 714, a pooling layer 705, fully connected layers 709, 712, 715, and an output data set 716.

Convolutional neural networks (CNNs) are modelling tools which may use the layout of the data as additional information in their inference by learning filters that take into consideration neighbor relations between the input features.

1-dimensional CNNs may be used as sequence processing tools which can learn to infer various things from input sequences, such as predicting the next few steps in a sequence or predicting optimal CIO and TTT values. In this use, the filters formed by the convolutional layer may consider each input feature's location in a single dimension, i.e., the time dimension. With this information, the filters may internally learn to deduce, e.g., current and future client device locations, speed, inertia, and possible trajectories the client device will take, as well as the corresponding radio environment, all helping in the correct prediction of optimal handover parameters.

Fig. 2B is a block diagram of the second network node device 210, in accordance with an example embodiment.

The second network node device 210 comprises at least one processor 212 and at least one memory 214 including computer program code. The second network node device 210 may also include other elements, such as a transceiver configured to enable the second network node device 210 to transmit and/or receive information to/from other devices, as well as other elements not shown in Fig. 2B. In one example, the second network node device 210 may use the transceiver to transmit or receive signaling information and data in accordance with at least one cellular communication protocol. The transceiver may be configured to provide at least one wireless radio connection, such as for example a 3GPP mobile broadband connection (e.g., 5G) . The transceiver may be configured to be coupled to at least one antenna to transmit and/or receive radio frequency signals.

Although the second network node device 210 is depicted to include only one processor 212, the second network node device 210 may include more processors. In an embodiment, the memory 214 is capable of storing instructions, such as an operating system and/or various applications. Furthermore, the memory 214 may include a storage that may be used to store, e.g., at least some of the information and data used in the disclosed embodiments .

Furthermore, the processor 212 is capable of executing the stored instructions. In an embodiment, the processor 212 may be embodied as a multi-core processor, a single core processor, or a combination of one or more multi-core processors and one or more single core processors. For example, the processor 212 may be embodied as one or more of various processing devices, such as a coprocessor, a microprocessor, a controller, a digital signal processor (DSP) , a processing circuitry with or without an accompanying DSP, or various other processing devices including integrated circuits such as, for example, an application specific integrated circuit (ASIC) , a field programmable gate array (FPGA) , a microcontroller unit (MCU) , a hardware accelerator, a special-purpose computer chip, or the like. In an embodiment, the processor 212 may be configured to execute hard- coded functionality. In an embodiment, the processor 212 is embodied as an executor of software instructions, wherein the instructions may specifically configure the processor 212 to perform the algorithms and/or operations described herein when the instructions are executed. The memory 214 may be embodied as one or more volatile memory devices, one or more non-volatile memory devices, and/or a combination of one or more volatile memory devices and nonvolatile memory devices. For example, the memory 214 may be embodied as semiconductor memories (such as mask ROM, PROM (programmable ROM) , EPROM (erasable PROM) , flash ROM, RAM (random access memory) , etc.) .

The second network node device 210 may comprise an operations, administration and management (OAM) unit. Alternatively, the second network node device 210 may comprise a base station. The base station may include, e.g., a fifthgeneration base station (gNB) or any such device providing an air interface for client devices to connect to the wireless network via wireless transmissions. At least in some embodiments, the second network node device 210 may comprise a multiple-input and multiple-output (MIMO) capable network node device .

The at least one memory 214 and the computer program code are configured to, with the at least one processor 212, cause the second network node device 210 at least to perform receiving (e.g., from the first network node device 200) past traffic information related to the one or more client devices 120 in the one or more cells 111, 112, 113 in the radio access network 110.

As discussed above in more detail, the past traffic information comprises information on past trajectories of the one or more client devices 120 in the one or more cells 111, 112, 113. For example, the past traffic information may comprise traffic information before and/or after at least one past handover event. For example, the information on the past trajectories may comprise a sequence of positioning (e.g., GPS) coordinates of the one or more client devices 120 measured at a fixed frequency, and/or a sequence of traffic measurement signals generated by the one or more client devices 120.

The at least one memory 214 and the computer program code are further configured to, with the at least one processor 212, cause the second network node device 210 at least to perform training the machine learning model to identify likely trajectories of the one or more client devices 120 in the one or more cells 111 , 112 , 113 by feeding the received past traf fic information to the machine learning model .

At least in some embodiments , the at least one memory 214 and the computer program code may be further configured to , with the at least one processor 212 , cause the second network node device 210 at least to perform transmiting the trained machine learning model to the first network node device 200 .

Further features of the second network node device 210 directly result from the functionalities and parameters of the first network node device 200 and thus are not repeated here .

Fig . 9 illustrates an example flow chart of a method 900 , in accordance with an example embodiment .

At optional operation 901 , the past traf fic information related to the one or more client devices 120 in the one or more cells 111 , 112 , 113 may be received at the first network node device 200 . As discussed above in more detail , the past traf fic information may comprise information on past traj ectories of the one or more client devices 120 in the one or more cells 111 , 112 , 113 .

Next , either the sequence of optional operations 902A and 903A or the optional operation 902B may be performed . That is , the sequence of optional operations 902A and 903A is alternative to the optional operation 902B .

At optional operation 902A, the received past traf fic information may be transmitted from the first network node device 200 to the second network node device 210 for use as training data .

At optional operation 903A, the machine learning model may be received at the first network node device 200 from the second network node device 210 after the machine learning model has been trained by the second network node device 210 .

At optional operation 902B, the first network node device 200 may train the machine learning model to identi fy likely traj ectories of the one or more client devices 120 in the one or more cells 111 , 112 , 113 by feeding the received past traf fic information to the machine learning model .

Regardless of whether the sequence of optional operations 902A and 903A or the optional operation 902B is performed, the first network node device 200 now has a trained machine learning model.

At operation 904, the first network node device 200 obtains current traffic information related to the one or more client devices 120 in the one or more cells 111, 112, 113 in the radio access network 110. As discussed in more detail above, the current traffic information comprises information on the current trajectories of the one or more client devices 120 in the one or more cells 111, 112, 113.

At operation 905, the first network node device 200 utilizes the obtained current traffic information in traffic load distribution. As discussed in more detail above, the utilizing of the obtained current traffic information in the traffic load distribution comprises applying the trained machine learning model to the obtained information on the current trajectories to predict the future trajectories of the one or more client devices 120 in the one or more cells 111, 112, 113.

The method 900 may be performed by the first network node device 200 of Fig. 2A. The operations 901-905 can, for example, be performed by the at least one processor 202 and the at least one memory 204. Further features of the method 900 directly result from the functionalities and parameters of the first network node device 200, and thus are not repeated here. The method 900 can be performed by computer program (s) .

Fig. 10 illustrates an example flow chart of a method 1000, in accordance with an example embodiment.

At operation 1001, the past traffic information related to the one or more client devices 120 in the one or more cells

111, 112, 113 may be received at the second network node device 210. As discussed above in more detail, the past traffic information comprises information on past trajectories of the one or more client devices 120 in the one or more cells 111,

112, 113.

At operation 1002, the second network node device 210 trains the machine learning model to identify likely trajectories of the one or more client devices 120 in the one or more cells 111, 112, 113 by feeding the received past traffic information to the machine learning model. At optional operation 1003 , the second network node device 210 may transmit the trained machine learning model to the first network node device 200 .

The method 1000 may be performed by the second network node device 210 of Fig . 2B . The operations 1001- 1003 can, for example , be performed by the at least one processor 212 and the at least one memory 214 . Further features of the method 1000 directly result from the functionalities and parameters of the second network node device 210 , and thus are not repeated here . The method 1000 can be performed by computer program ( s ) .

At least some of the embodiments may ensure that optimal traf fic steering settings are learned for each traj ectory in the network, thereby improving the robustness of the handovers and the management of load distribution among cells and cell layers .

The first network node device 200 may comprise means for performing at least one method described herein . In an example , the means may comprise the at least one processor 202 , and the at least one memory 204 including program code configured to , when executed by the at least one processor 202 , cause the first network node device 200 to perform the method .

The second network node device 210 may compri se means for performing at least one method described herein . In an example , the means may comprise the at least one processor 212 , and the at least one memory 214 including program code configured to , when executed by the at least one proces sor 212 , cause the second network node device 210 to perform the method .

The functionality described herein can be performed, at least in part , by one or more computer program product components such as software components . According to an embodiment , the first network node device 200 and/or the second network node device 210 may comprise a processor or processor circuitry, such as for example a microcontroller, configured by the program code when executed to execute the embodiments of the operations and functionality described . Alternatively, or in addition, the functionality described herein can be performed, at least in part , by one or more hardware logic components . For example , and without limitation, illustrative types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs) , Program-specific Integrated Circuits (ASICs) , Programspecific Standard Products (ASSPs) , System-on-a-chip systems (SOCs) , Complex Programmable Logic Devices (CPLDs) , and Graphics Processing Units (GPUs) .

Any range or device value given herein may be extended or altered without losing the effect sought. Also, any embodiment may be combined with another embodiment unless explicitly disallowed .

Although the subject matter has been described in language specific to structural features and/or acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as examples of implementing the claims and other equivalent features and acts are intended to be within the scope of the claims.

It will be understood that the benefits and advantages described above may relate to one embodiment or may relate to several embodiments. The embodiments are not limited to those that solve any or all of the stated problems or those that have any or all of the stated benefits and advantages. It will further be understood that reference to 'an' item may refer to one or more of those items.

The steps of the methods described herein may be carried out in any suitable order, or simultaneously where appropriate. Additionally, individual blocks may be deleted from any of the methods without departing from the spirit and scope of the subject matter described herein. Aspects of any of the embodiments described above may be combined with aspects of any of the other embodiments described to form further embodiments without losing the effect sought.

The term 'comprising' is used herein to mean including the method, blocks or elements identified, but that such blocks or elements do not comprise an exclusive list and a method or apparatus may contain additional blocks or elements.

It will be understood that the above description is given by way of example only and that various modifications may be made by those skilled in the art. The above specification, examples and data provide a complete description of the structure and use of exemplary embodiments . Although various embodiments have been described above with a certain degree of particularity, or with reference to one or more individual embodi- ments , those skilled in the art could make numerous alterations to the disclosed embodiments without departing from the spirit or scope of this speci fication .

Claims

CLAIMS :

1. A first network node device (200) , comprising: at least one processor (202) ; and at least one memory (204) including computer program code ; the at least one memory (204) and the computer program code configured to, with the at least one processor (202) , cause the first network node device (200) at least to perform: obtaining current traffic information related to one or more client devices (120) in one or more cells (111, 112, 113) in a radio access network (110) , the current traffic information comprising information on current trajectories of the one or more client devices (120) in the one or more cells (111, 112, 113) ; and utilizing the obtained current traffic information in traffic load distribution, wherein the utilizing of the obtained current traffic information in the traffic load distribution comprises applying a machine learning model to the obtained information on the current trajectories to predict future trajectories of the one or more client devices (120) in the one or more cells (111, 112, 113) .

2. The first network node device (200) according to claim 1, wherein the utilizing of the obtained current traffic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current trajectories to predict how the traffic load in the one or more cells (111, 112, 113) is likely to change in view of the predicted future trajectories of the one or more client devices (120) in the one or more cells (111, 112, 113) .

3. The first network node device (200) according to claim 2, wherein the utilizing of the obtained current traffic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current trajectories to identify cells likely to experience congestion.

4. The first network node device (200) according to claim 3, wherein the utilizing of the obtained current traffic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current trajectories to determine one or more handover options resolving the congestion identified likely to be experienced .

5. The first network node device (200) according to claim 3 or 4, wherein the utilizing of the obtained current traffic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current trajectories to select the client devices of the one or more client devices (120) whose serving cells need to be adjusted in order to minimize the likelihood of the congestion resulting from the predicted future trajectories of the one or more client devices (120) and the predicted likely traffic load change.

6. The first network node device (200) according to any of claims 1 to 5, wherein the utilizing of the obtained current traffic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current trajectories to select one or more of the one or more client devices (120) for handover.

7. The first network node device (200) according to any of claims 1 to 6, wherein the utilizing of the obtained current traffic information in the traffic load distribution further comprises applying the machine learning model to the obtained information on the current trajectories to determine at least one of: mappings between the predicted future trajectories and cells available along the predicted future trajectories, mappings between the predicted future trajectories and the traffic load distribution in the one or more cells (111, 112, 113) , or mappings between the predicted future trajectories and coverages of the one or more cells (111, 112, 113) or corresponding cell layers.

8. The first network node device (200) according to any of claims 1 to 7, wherein the utilizing of the obtained current traffic information in traffic load distribution comprises at least one of: utilizing the obtained current traffic information in traffic load distribution for the one or more client devices (120) in the one or more cells (111, 112, 113) , or utilizing the obtained current traffic information in traffic load distribution towards a cell of the one or more cells (111, 112, 113) for the one or more client devices (120) and their supported services .

9. The first network node device (200) according to any of claims 1 to 8, wherein the at least one memory (204) and the computer program code are further configured to, with the at least one processor (202) , cause the first network node device (200) to perform receiving past traffic information related to the one or more client devices (120) in the one or more cells (111, 112, 113) , the past traffic information comprising information on past trajectories of the one or more client devices (120) in the one or more cells (111, 112, 113) .

10. The first network node device (200) according to claim 9, wherein the at least one memory (204) and the computer program code are further configured to, with the at least one processor (202) , cause the first network node device (200) to perform transmitting the received past traffic information to a second network node device (210) for use as training data.

11. The first network node device (200) according to claim 10, wherein the at least one memory (204) and the computer program code are further configured to, with the at least one processor (202) , cause the first network node device (200) to perform receiving the machine learning model from the second network node device (210) after the machine learning model has been trained by the second network node device (210) .

12. The first network node device (200) according to claim 11, wherein the at least one memory (204) and the computer program code are further configured to, with the at least one processor (202) , cause the first network node device (200) to perform training the machine learning model to identify likely trajectories of the one or more client devices (120) in the one or more cells (111, 112, 113) by feeding the received past traffic information to the machine learning model.

13. The first network node device (200) according to any of claims 1 to 12, wherein the information on the current trajectories comprises at least one of a sequence of positioning coordinates of the one or more client devices (120) measured at a fixed frequency, or a sequence of traffic measurement signals generated by the one or more client devices (120) .

14. The first network node device (200) according to claim 13, wherein the traffic measurement signals comprise at least one of reference signal received power signals or signal- to-interf erence-plus-noise ratio signals.

15. The first network node device (200) according to any of claims 1 to 14, wherein the first network node device (200) comprises a base station.

16. A method (900) , comprising: obtaining (904) , by a first network node device, current traffic information related to one or more client devices in one or more cells in a radio access network, the current traffic information comprising information on current trajectories of the one or more client devices in the one or more cells; and utilizing (905) , by the first network node device, the obtained current traffic information in traffic load distribution, wherein the utilizing (905) of the obtained current traffic information in the traffic load distribution comprises applying a machine learning model to the obtained information on the current trajectories to predict future trajectories of the one or more client devices in the one or more cells.

17. A computer program comprising instructions for causing a first network node device to perform at least the following : obtaining current traffic information related to one or more client devices in one or more cells in a radio access network, the current traffic information comprising information on current trajectories of the one or more client devices in the one or more cells; and utilizing the obtained current traffic information in traffic load distribution, wherein the utilizing of the obtained current traffic information in the traffic load distribution comprises applying a machine learning model to the obtained information on the current trajectories to predict future trajectories of the one or more client devices in the one or more cells.

18. A second network node device (210) , comprising: at least one processor (212) ; and at least one memory (214) including computer program code ; the at least one memory (214) and the computer program code configured to, with the at least one processor (212) , cause the second network node device (210) at least to perform: receiving past traffic information related to one or more client devices (120) in one or more cells (111, 112, 113) in a radio access network (110) , the past traffic information comprising information on past trajectories of the one or more client devices (120) in the one or more cells (111, 112, 113) ; and training a machine learning model to identify likely trajectories of the one or more client devices (120) in the one or more cells (111, 112, 113) by feeding the received past traffic information to the machine learning model.

19. The second network node device (210) according to claim 18, wherein the information on the past trajectories comprises at least one of a sequence of positioning coordinates of the one or more client devices (120) measured at a fixed frequency, or a sequence of traffic measurement signals generated by the one or more client devices (120) .

20. The second network node device (210) according to claim 18 or 19, wherein the past traffic information comprises traffic information at least one of before or after at least one past handover event.

21. The second network node device (210) according to any of claims 18 to 20, wherein the second network node device (210) comprises an operations, administration and management unit or a base station.

22. A method (1000) , comprising: receiving (1001) , at a second network node device, past traffic information related to one or more client devices in one or more cells in a radio access network, the past traffic information comprising information on past trajectories of the one or more client devices in the one or more cells; and training (1002) , by the second network node device, a machine learning model to identify likely trajectories of the one or more client devices in the one or more cells by feeding the received past traffic information to the machine learning model .

23. A computer program comprising instructions for causing a second network node device to perform at least the following : receiving past traffic information related to one or more client devices in one or more cells in a radio access network, the past traffic information comprising information on past trajectories of the one or more client devices in the one or more cells; and training a machine learning model to identify likely trajectories of the one or more client devices in the one or more cells by feeding the received past traffic information to the machine learning model.