Antenna tilting based on prediction of performance indicator values

Abstract

A computer-implemented method may comprise: determining a machine learning (ML) model configured to predict performance indicator values of a plurality of cells of a cellular communication network, wherein the plurality of cells comprises a first cell and other cell(s), and wherein the ML model is configured to receive input data comprising: network topology information of the plurality of cells, spatial traffic distribution information of at least the first cell, and antenna configuration information of the first cell, wherein the antenna configuration information is indicative of a tilt angle of an antenna of the first cell; predicting, by the ML model based on the input data, the performance indicator values for a plurality of antenna configurations of the first cell; selecting a new antenna configuration for the first cell based on the performance indicator values; and causing tilting of the antenna of the first cell based on the new antenna configuration.

Description

ANTENNA TILTING BASED ON PREDICTION OF PERFORMANCEINDICATOR VALUESTECHNICAL FIELD

[0001] Various example embodiments generally relate to the field of wireless communications. Some example embodiments relate to antenna tilting of cells based on performance indicator values predicted by a machine learning model.BACKGROUND

[0002] Wireless communication may be implemented with a cellular radio network comprising transmission sites that offer communication services via multiple cells. A cell may correspond to certain geographical coverage area and be operated on a particular frequency. Antenna tilt may refer to an angle at which a radiation pattern of an antenna is vertically inclined. Antenna tilt may be used in cellular communication systems to optimize signal coverage and thereby to improve performance of the network. Machine learning (ML) is a field of technology that may relate to development of statistical algorithms that learn from training data to generalize their performance to unseen data, and thus to perform tasks without explicit instructions, for example in context of cellular communication networks.SUMMARY

[0003] This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.

[0004] Example embodiments of the present disclosure enable to improve operation of a cellular communication network by adjusting antenna tilts. This benefit may be achieved by the features of the independent claims. Further example embodiments are provided in the dependent claims, the description, and the drawings.

[0005] According to a first aspect, a computer-implemented method is disclosed. The method may comprise: determining a machine learning model configured to predict performance indicator values of a plurality of cells of a cellular communication network, wherein the plurality of cells comprises a first cell and one or more other cells, and wherein the machine learning model is configured to receive input data comprising: network topology information of the plurality of cells, spatial traffic distribution information of at least the first cell, and antenna configuration information of the first cell, wherein the antenna configuration information is indicative of a tilt angle of an antenna of the first cell; predicting, by the machine learning model based on the input data, the performance indicator values of the plurality of cells for a plurality of antenna configurations of the first cell; selecting a new antenna configuration for the first cell based on the performance indicator values of the plurality of cells predicted for the plurality of antenna configurations of the first cell; and causing tilting of the antenna of the first cell based on the new antenna configuration.

[0006] According to an example embodiment of the first aspect, the antenna configuration information or the new antenna configuration comprises one or more of the following: the tilt angle of the antenna of the first cell, a half-power distance of a radiation pattern of the antenna of the first cell, a half-power beam width of the radiation pattern of the antenna of the first cell, a half-power angle of a vertical radiation pattern of the antenna of the first cell in a horizontal direction of a main lobe of the radiation pattern of the antenna of the first cell, or an attenuation level of a horizontal direction of the vertical radiation pattern in the horizontal direction of the main lobe of the radiation pattern of the antenna of the first cell.

[0007] According to an example embodiment of the first aspect, the performance indicator values comprise one or more of the following: a channel quality indicator, a multiple-input multiple output channel rank indicator, a spectral efficiency, or a data volume.

[0008] According to an example embodiment of the first aspect, the input data further comprises: a received signal strength of a signal of the first cell as measured by users of the one or more other cells, or an interference level of the signal of the first cell at the one or more other cells.

[0009] According to an example embodiment of the first aspect, the method further comprises: causing the tilting of the antenna of the first cell based on the new antenna configuration, in response to determining that a difference between at least one performance indicator value of the plurality of cells predicted for the new antenna configuration and at least one measured performance indicator value of the plurality of cells associated with a current antenna configuration of the first cell exceeds a first threshold.

[0010] According to an example embodiment of the first aspect, the method further comprises: predicting, by the machine learning model based on the input data, the performance indicator values of the plurality of cells for a current antenna configuration of the first cell; and determining not to use the machine learning model for selecting the new antenna configuration, in response to determining that a difference between at least one performance indicator value of the plurality of cells predicted for the current antenna configuration and at least one measured performance indicator of the plurality of cells associated with the current antenna configuration exceeds a second threshold.

[0011] According to an example embodiment of the first aspect, the method further comprises: updating the machine learning model based on input data comprising the new antenna configuration and performance indicator values measured with the new antenna configuration.

[0012] According to an example embodiment of the first aspect, the method further comprises: determining a second machine learning model configured to predict at least one timing advance distribution value of the first cell, wherein the second machine learning model is configured to receive second input data comprising: the traffic load level information of the one or more other cells, and the antenna configuration information of the first cell, wherein the second input data does not comprise the spatial traffic distribution information of the first cell; predicting, by the second machine learning model based on the second input data, the at least one timing advance distribution value of the first cell; providing the predicted timing advance value distribution value of the first cell to the machine learning model as the spatial traffic distribution information of the first cell for prediction of the performance indicator values of the plurality of cells.

[0013] According to an example embodiment of the first aspect, the input data further comprises traffic load level information of the one or more other cells, or the second input data further comprises network topology information.

[0014] According to an example embodiment of the first aspect, the network topology information comprises: a minimum distance between a transmission site of the first cell and any other transmission site of the cellular communication network, or a number of cells overlapping with the first cell.

[0015] According to an example embodiment of the first aspect, the antenna configuration information comprises: a height level of the antenna of the first cell.

[0016] According to an example embodiment of the first aspect, the traffic load level information of the one or more other cells comprises physical resource block utilization rate values of the one or more other cells.

[0017] According to an example embodiment of the first aspect, the input data to the machine learning model further comprises: a type of propagation environment of the plurality of cells, a frequency band of the plurality of cells, or an identity of a manufacturer of network equipment of the plurality of cells.

[0018] According to an example embodiment of the first aspect, the method further comprises: selecting the new antenna configuration based on determining that the new antenna configuration provides best predicted performance among the plurality of antenna configurations.

[0019] According to an example embodiment of the first aspect, the method further comprises: normalizing the performance indicator values of the plurality of cells predicted for the plurality of antenna configurations; summing normalized performance indicator values for each antenna configuration of the plurality of antenna configurations; and selecting the new antenna configuration based on comparing the summed normalized performance indicator values of each antenna configuration of the plurality of antenna configurations.

[0020] According to a second aspect, an apparatus may comprise means for performing the method of the first aspect, or any example embodiment thereof.

[0021] According to an example embodiment of the second aspect, the means comprises at least one processor; and at least one memory storing instructions that,when executed by the at least one processor, cause the apparatus to perform the method of the first aspect, or any example embodiment thereof.

[0022] According to a third aspect, computer program or a computer program product may comprise program code configured to, when executed by a processor, cause an apparatus at least to perform the method of the first aspect, or any example embodiment thereof.

[0023] According to a fourth aspect, an apparatus may comprise at least one processor; and at least one memory including computer program code; the at least one memory and the computer code configured to, with the at least one processor, cause the apparatus at least to: determine a machine learning model configured to predict performance indicator values of a plurality of cells of a cellular communication network, wherein the plurality of cells comprises a first cell and one or more other cells, and wherein the machine learning model is configured to receive input data comprising: network topology information of the plurality of cells, spatial traffic distribution information of at least the first cell, and antenna configuration information of the first cell, wherein the antenna configuration information is indicative of a tilt angle of an antenna of the first cell; predict, by the machine learning model based on the input data, the performance indicator values of the plurality of cells for a plurality of antenna configurations of the first cell; select a new antenna configuration for the first cell based on the performance indicator values of the plurality of cells predicted for the plurality of antenna configurations of the first cell; and cause tilting of the antenna of the first cell based on the new antenna configuration. The computer program may be configured to, with the at least one processor, cause the apparatus to perform the method of the first aspect, or any example embodiment thereof.

[0024] Any example embodiment may be combined with one or more other example embodiments. Many of the attendant features will be more readily appreciated as they become better understood by reference to the following detailed description considered in connection with the accompanying drawings.DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings, which are included to provide a further understanding of the example embodiments and constitute a part of this specification, illustrate example embodiments and together with the description help to understand the example embodiments. In the drawings:

[0026] FIG. 1 illustrates an example of a cellular communication network;

[0027] FIG. 2 illustrates an example of an apparatus configured to practise one or more example embodiments;

[0028] FIG. 3 illustrates an example of input feature engineering for a machine learning model configured to predict performance indicator values of a cellular communication network;

[0029] FIG. 4 illustrates an example of antenna configuration parameters;

[0030] FIG. 5 illustrates an example of a flow chart for antenna tilt adjustment;

[0031] FIG. 6 illustrates an example of determining a machine learning model for predicting performance indicator values;

[0032] FIG. 7 illustrates an example of cells affected by antenna tilt of one cell;

[0033] FIG. 8 illustrates an example of selecting a new antenna configuration based on predicted performance indicator values;

[0034] FIG. 9 illustrates an example of determining a machine learning model for predicting timing advance distribution value(s);

[0035] FIG. 10 illustrates an example of predicted and measured channel quality indicator values;

[0036] FIG. 11 illustrates an example of channel quality indicator values before and after adjustment of antenna tilt;

[0037] FIG. 12 illustrates an example of spectral efficiency values before and after adjustment of antenna tilt; and

[0038] FIG. 13 illustrates an example of a method for determining new antenna configuration for a cell in a communication network.

[0039] Like references are used to designate like parts in the accompanying drawings.DETAILED DESCRIPTION

[0040] Reference will now be made in detail to example 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.

[0041] FIG. 1 illustrates an example of a cellular communication network. Communication network 100 may comprise one or more devices, which may be also referred to as client nodes, user nodes, user equipment (UE), terminal devices, or simply users. An example of such a device is UE 110, which may communicate with one or more access nodes of radio access network (RAN) 120. An access node may be also referred to as an access point or a base station. Communication network 100 may be configured for example in accordance with the 4thor 5thgeneration (4G, 5G) digital cellular communication networks, or any future generation of digital cellular communication networks (e.g., 6G), as defined by the 3rdGeneration Partnership Project (3GPP). In one example, communication network 100 may operate according to 3GPP (4G) LTE (Long-Term Evolution) or 3GPP 5G NR (New Radio) standards. Access nodes 122, 124, 126 of RAN 120 may for example comprise 5thgeneration access nodes (gNB) or 4thgeneration access nodes (eNodeB). It is however appreciated that example embodiments presented herein are not limited to these example networks and may be applied in any present or future wireless communication networks, or combinations thereof, for example other type of cellular networks, short-range wireless networks, multicast networks, broadcast networks, or the like.

[0042] An access node may provide communication services within one or more cells, illustrated in FIG. 1 with dotted circles, which may correspond to geographical area(s) covered by signals transmitted by the access node. Communication network 100 may therefore comprise a cellular radio network. For example, access node 122 may be configured to serve cells 132-1, 132-2, and 132-3,for example at respective sectors of the transmission site at which access node 122 is deployed. A transmission site may comprise a geographical location comprising equipment for serving cell(s), for example access node circuitry and antenna(s) configured to enable communication with users. A sector may comprise a range of angles at the horizontal direction from an access node. A sector may comprise one or more cells on one or more frequencies. For example, access node 122, or in general the respective transmission site, might be configured with three 120-degree sectors comprising respective cells 132-1, 132-2, 132-3, and optionally one or more other cells. Similarly, access node 124, or in general the respective transmission site, might be configured with three 120-degree sectors comprising respective cells 134-1, 134-2, 134-3, and optionally one or more other cells. RAN 120 may comprise further access nodes, e.g., access node 126, with respective cell(s).

[0043] Communication network 100 may comprise a core network 130, which may comprise various network functions (NF) for establishing, configuring, and controlling data communication sessions of users, for example UE 110. Communication network 100 may comprise a network controller 140, which may be responsible of configuring various operations of RAN 120 and / or core network 130. Even though illustrated as a separate entity, network controller 140 may be alternatively embodied as part of core network 130. Even though some operations have been described as being performed by network controller 140, it is understood that similar functions may be performed alternatively by other network device(s) or network function(s) of communication network 100. One task of network controller 140 may be to optimize antenna tilts of cells, such as cell 132-1, within RAN 120.

[0044] Even though not illustrated in FIG. 1, communication network 100 may comprise a network management system (NMS), or another entity, which may be configured to process and store performance management data of communication network 100. The performance management data may comprise various types of information collected from different network elements, for example access nodes 122, 124, 126 of RAN 120. The NMS may be configured to operate as a centralized data management system (e.g., a server), which processes the collected data and provides it for network management functions, for example for antenna tiltingapplications. Alternatively, similar functionality may be provided at an operations support system (OSS) of communication network 100, or another network device.

[0045] Examples of the performance indicators include channel quality indicator (CQI), channel rank, spectral efficiency, received signal strength (e.g., reference signal received power, RSRP), and data volume. Performance indicators may be provided for individual users or for different cells, for example as average values of users of the cell.

[0046] CQI may indicate the most spectrally efficient modulation and coding scheme (MCS) applicable for achieving a certain error rate for given channel conditions. CQI may for example comprise an integer number (index), for example between 0 and 15. Low CQI may indicate worse radio performance and higher CQI may indicate better radio performance, e.g., in terms of spectral efficiency. CQI may therefore increase with increasing spectral efficiency of the associated MCS. UE 110 may be configured to estimate the CQI, for example based on reference signals received from an access node, and to report the CQI to the access node and / or core network 130.

[0047] Channel rank may indicate the number of independent data streams that can be transmitted simultaneously over a multiple input multiple output (MIMO) system in the currently prevailing radio conditions. For example, channel rank may be configured to indicate the number of spatial layers that can be used for data transmission. In a 4x4 MIMO system, the maximum channel rank may be equal to four, meaning that up to four independent data streams may be transmitted simultaneously. However, the actual channel rank in real-world environments may be often lower due to non-optimal radio conditions, for example due to correlation between different radio paths. UE 110 may be configured to determine the channel rank based on radio channel measurements. UE 110 may be configured to report the channel rank to the access node, for example by transmitting a rank indicator (RI), which may be part of channel state information (CSI), e.g., along with the CQI.

[0048] Data volume may indicate an amount of data (e.g., number of bit or bytes) communicated with a UE, or via a cell, for example during a certain time period.Using the data volume as one output parameter provides the benefit of enabling to ensure that coverage is not lost in case of downtilt of cell 134-1.

[0049] Timing advance may be used in communication network 100 to synchronize transmission and reception between an individual user and the serving access node such that the propagation time of the signal over a particular distance between the user and the access node is compensated. Timing advance values may therefore correlate with distances of users from the access node and the same applies also to received signal strength due to the propagation loss.

[0050] Network controller 140, or another network entity such as the NMS, may collect information about any of the above parameters from different UEs, for example to determine statistical information (e.g., average) about the different parameters. The statistical information may be determined for example at cell level such that a statistical value is provided for each parameter and for each cell.

[0051] A dominance area of a cell may comprise a geographical area in which the cell has the strongest signal level. Handover between cells may be performed when UE 110 is at or near the border of the dominance area. Coverage areas cells may overlap to some extent, for example to facilitate smooth handover for mobile users. The serving cell of a user may be changed when another cell has the strongest signal level. Even though some overlapping may be useful for handover purposes, it may be generally desired to minimize the signal level outside the dominance area. Overshooting is one example of a phenomenon that may occur due to wrong antenna tilting. Another example is the emergence of coverage holes between cells.

[0052] Dominance and coverage areas of cells may be adjusted by a remote antenna tilt (RET) mechanism, which may be configured to adjust antenna tilt of a cell, for example upon a request received from network controller 140, or another control entity of communication network 100. In general, a RET mechanism may comprise any solution for remotely adjusting antenna tilt of a cell, for example in contrast to manually redirecting the antenna at the transmission site. For example, the RET mechanism may comprise a motor coupled to a phase shifter and be configured to adjust the phase shift generated by the phase shifter, in order to cause the vertical power radiation pattern of the antenna to change. Adjusting the power radiation pattern by such a mechanism may be called electrical tilting (E-tilting).Mechanical antenna tilting may refer to mechanically adjusting the antenna tilt by rotating the antenna itself. RET may be used to optimize network performance by improving coverage of the network or by reducing interference between cells. Antenna tilt of a cell may be adjusted in order to affect the received signal strength at different locations, for example to improve coverage or to reduce interference between cells.

[0053] Interference between cells may occur due to non-optimal antenna tilting. For example, insufficient antenna downtilt may be observed in communication network 100 as overshooting, e.g., directing the power radiation pattern of the antenna unnecessarily high, thereby extending the coverage area of the associated cell. Also, network controller 140 may be configured to detect undershooting cells, e.g., cells, where the power radiation pattern of the antenna is directed unnecessarily steep towards the ground. Network controller 140 may be configured to detect overshooting and undershooting cells and to perform corresponding counteraction(s), e.g., antenna downtilting or uptilting, by RET, or otherwise adjust the antenna tilts to optimize operation of RAN 120. Uptilting / downtilting a cell may comprise uptilting / downtilting the vertical power radiation pattern of an antenna of the cell. Tilting an antenna may comprise electrically adjusting the direction of the power radiation pattern, for example by applying differently delayed and / or weighted versions of transmitted / received signals.

[0054] One approach to determine optimal antenna tilts in the network is to use optimization algorithms. However, setting the optimization target may be challenging. This may be also reflected in the parametrization of input values and challenges in setting proper input parameters for the algorithms. Furthermore, some tilt adjustment method may require setting limits for the antenna tilt changes and because of that some cells may remain outside of the scope of tilt change.

[0055] Example embodiments of the present disclosure provide methods for optimizing antenna tilts in cellular communication network 100 by using a machine learning model configured to predict performance indicator values of different cells for different antenna configurations (e.g., tilt angles). This provides the benefit of enabling to optimize performance (e.g., overall data volume) in cellular communication network 100 by selecting antenna configurations that improvequality of the network in terms of the performance indicators, while maintaining coverage. Example embodiments enable predicting optimized antenna tilt for each cell in network. In addition, threshold(s) may be set to prevent unwanted tilt changes, for example to prevent overtilting.

[0056] In accordance with the example embodiments, a computer-implemented method may comprise: determining a machine learning (ML) model configured to predict performance indicator values of a plurality of cells of a cellular communication network, wherein the plurality of cells comprises a first cell and other cell(s), and wherein the ML model is configured to receive input data comprising: network topology information of the plurality of cells, spatial traffic distribution information of at least the first cell, and antenna configuration information of the first cell, wherein the antenna configuration information is indicative of a tilt angle of an antenna of the first cell; predicting, by the ML model based on the input data, the performance indicator values for a plurality of antenna configurations of the first cell; selecting a new antenna configuration for the first cell based on the performance indicator values; and causing tilting of the antenna of the first cell based on the new antenna configuration.

[0057] FIG. 2 illustrates an example embodiment of an apparatus 200 configured to perform one or more example embodiments. Apparatus 200 may be for example used to implement network controller 140 or in general a device configured to at least one of determine or implement antenna tilting of one or more cells. Apparatus 200 may comprise at least one processor 202. The at least one processor 202 may comprise, for example, one or more of various processing devices or processor circuitry, such as for example a co-processor, 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.

[0058] Apparatus 200 may further comprise at least one memory 204. The at least one memory 204 may be configured to store, for example, computer program code or the like, for example operating system software and application software.The at least one memory 204 may comprise one or more volatile memory devices, one or more non-volatile memory devices, and / or a combination thereof. For example, the at least one memory 204 may be embodied as magnetic storage devices (such as hard disk drives, magnetic tapes, etc.), optical magnetic storage devices, or semiconductor memories (such as mask ROM, PROM (programmable ROM), EPROM (erasable PROM), flash ROM, RAM (random access memory), etc.).

[0059] Apparatus 200 may further comprise a communication interface 208 configured to enable apparatus 200 to transmit and / or receive information to / from other devices, functions, or entities. In one example, apparatus 200 may use communication interface 208 to output indication(s) of inoperative RET mechanisms to an automated service ticket system. Apparatus 200 may further comprise a user interface 210, for example for configuring apparatus 200 or for providing user output by the apparatus, such as for example visual and / or audible signal(s), for example by speaker(s), display(s), light(s), or the like. User interface 210 may be for example configured to output indication(s) of optimized antenna tilting values to a human user.

[0060] When apparatus 200 is configured to implement some functionality, some component and / or components of apparatus 200, such as for example the at least one processor 202 and / or the at least one memory 204, may be configured to implement this functionality. Furthermore, when the at least one processor 202 is configured to implement some functionality, this functionality may be implemented using program code 206 comprised, for example, in the at least one memory 204.

[0061] The functionality described herein may be performed, at least in part, by one or more computer program product components such as for example software components. According to an embodiment, the apparatus comprises 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. A computer program or a computer program product may therefore comprise instructions for causing, when executed, apparatus 200 to perform the method(s) described herein. Alternatively, or in addition, the functionality described herein can be performed, at least in part, by one or morehardware logic components. For example, and without limitation, illustrative types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs), application-specific Integrated Circuits (ASICs), applicationspecific Standard Products (ASSPs), System-on-a-chip systems (SOCs), Complex Programmable Logic Devices (CPLDs), Graphics Processing Units (GPUs).

[0062] Apparatus 200 comprises means for performing at least one method described herein. In one example, the means comprises the at least one processor 202, the at least one memory 204 including program code 206 configured to, when executed by the at least one processor, cause the apparatus 200 to perform the method.

[0063] Apparatus 200 may comprise a computing device such as for example an access point, a base station, a server, a network device, a network function device, or the like. Although apparatus 200 is illustrated as a single device it is appreciated that, wherever applicable, functions of apparatus 200 may be distributed to a plurality of devices, for example to implement example embodiments as a cloud computing service.

[0064] FIG. 3 illustrates an example of input feature engineering for a machine learning model configured to predict performance indicator values of a cellular communication network. Input values of the ML model may be based on existing configurations of cellular communication network 100. The selected input values may be used to determine a ML model that is configured to predict performance indicators of cells. The ML model which is configured to predict performance indicator values may be also referred to as a first ML model. The predicted performance indicators may be then used to determine new antenna configurations for cells such that quality of the network is improved. The quality of the network may be indicated by the performance indicators, which may be also called key performance indicators (KPI). Performing prediction by a ML model may comprise evaluating or executing the ML model or performing inference with the ML model.

[0065] Data on performance of cellular communication network 100 may be modelled against the listed input values. The modelled data provides output parameters including performance indicators such as, CQI, channel rank, received signal strength (e.g., RSRP), spectral efficiency, and / or data volume. The outputparameters may comprise values of both the cell subject to antenna tilt adjustment, also referred to as cell to be tilted or tilted cell, and other cell(s) surrounding the cell to be tilted. The cell to be tilted and the surrounding cells may be collectively referred to as affected cells, indicating that performance indicators of these cells are affected by the adjustment of the antenna tilt. For the purposes of this specification, cell 134-1 may be considered as the cell to be tilted. Cell 134-1 may be also referred to as a first cell. The other cell(s) may include neighbouring or surrounding cells such as cell 132-1, 132-2, 132-3, 136-1, 136-2, or 136-3. The other cells may include cells that network controller 140 estimates to be affected by antenna tilting of cell 134-1, as will be further described with reference to FIG.7.

[0066] Several feature engineering rounds may be performed before selecting the input values and output values of the ML model. An example of feature engineering is provided in FIG. 3, which illustrates correlations between different configuration parameters. A darker shade of grey indicates higher correlation. It may be beneficial not to select correlated parameters as input values for the ML model. For example, correlated parameters may bias the output of the ML model towards certain features. In this example, D50C, D60C, and D70C indicate different distance percentiles calculated from the TA distribution, where D50C represents the median distance, D60C indicates the 60% percentile distance, and D70C indicates the 70% percentile distance. Because these distance percentile parameters are highly correlated, only one of the parameters may be selected as an input feature for determining the ML model.

[0067] Input data of the ML model may comprise traffic load level information, e.g., PRB utilization, of the other cells. This provides the benefit of enabling the ML model to take into account the load level of both the cell 134-1 and the other cell(s), when predicting the performance indicator values. In some example embodiments, the input data may comprise traffic load level information of cell 134-1. Load level may indicate the share of used (e.g., occupied) transmission resources, e.g., PRBs, with respect to available transmission resources of the cell.

[0068] The input data of the ML model may comprise spatial traffic distribution data of cell 134-1. The spatial traffic distribution data may for example comprise TA value distribution value(s), for example distance(s) corresponding to particularpercentile of TA values of cell 134-1. An / / -th percentile value (e.g., TA value in time or the corresponding distance) of the TA distribution may comprise a value which is higher than n % of the values. This provides the benefit of enabling the ML model to take into account from where the data traffic of cell 134-1 comes from. In some example embodiments, the input data may comprise spatial traffic distribution information of the other cell(s). This provides the benefit of enabling the ML model to consider locations of users of the other cell(s) when estimating how antenna tilt of cell 134-1 affects performance indicators in RAN 120, in particular the other cell(s).

[0069] The input data of the ML model may comprise antenna configuration information of cell 134-1, for example one or more of the following parameters: tilt angle of an antenna of cell 134-1, half-power (-3 dB) distance of a radiation pattern of the antenna, vertical half-power beam width of the antenna, half-power tilt angle (e.g., upper / lower), height of the antenna (e.g., height level with reference to ground or some reference level such as sea level), type of installation of the antenna (e.g., building or pole), an attenuation level of a horizontal direction of the vertical radiation pattern of the antenna.

[0070] The input data of the ML model may comprise network topology information of cells (e.g., cell 134-1 and one or more other cells) such as for example nearest cell distance, type of propagation environment (e.g., urban or rural), or the number of interfering / overlapping cells for at least cell 134-1. This provides the benefit of enabling the ML model to take into account the network topology, e.g., in terms of density of cells, when predicting the performance indicator values for different antenna configurations. Nearest cell distance may comprise the distance between the transmission site of cell 134-1 and the closest neighbouring transmission site, in other words a minimum distance between the transmission site of cell 134-1 and any other transmission site of cellular communication network 100.

[0071] An example of the above antenna configuration parameters is illustrated in FIG. 4. The tilt angle of the antenna may indicate the angle of the maximum gain of the vertical antenna radiation pattern, for example with respect the horizontal (0°) direction from the antenna. In the example of FIG.4 the direction of themaximum gain of the vertical radiation pattern is towards the 0 dB point. Upper and lower half-power (-3 dB) slopes may be defined by directions of the vertical power radiation patters, for which the gain is half of the maximum gain. The half-power distance may comprise the distance between the transmission site and a point where the upper half-power slope intersects with ground level. The half-power (-3 dB) angle may comprise the angle between the horizontal direction and the upper halfpower slope. It is however noted that the angles might be defined in any suitable way, for example with respect to a vertical line. The vertical beamwidth may comprise the angular range between the upper and lower half-power slopes of the vertical radiation pattern. The attenuation level of the horizontal direction of the vertical radiation pattern may comprise the difference in gain between the direction of the maximum gain (pointing towards the 0 dB point in FIG. 4) and the horizontal (0°) direction. The antenna configuration information may be therefore indicative of the tilt angle of the antenna of cell 134-1, for example by including one or more of the above parameters. Any of the above parameters may be defined considering the horizontal direction of the main lobe of the radiation pattern. Using different antenna configuration data as inputs to the ML model provides the benefit of obtaining a ML model that is able to predict the performance indicator values for different antenna configurations. Suitable antenna configurations may be then determined for cell 134-1 by performing inference with the ML model for different antenna configurations and selecting the antenna configuration that provides best, or sufficient, performance.

[0072] Further input data of the ML model may comprise the frequency band (e.g., LTE800), the vendor of network equipment (e.g., radio equipment), tilted cell measured RSRP in affected cells (Layer 3 data), or tilted cell measured signal-to- interference-plus-noise ratio (SINR) in affected cells (Layer 3 data). Providing the vendor of network equipment of the cell as input data to the ML model provides the benefit of enabling to cover different implementations by different vendors, for example with respect to reporting of the other parameters of the input data. The vendor information may comprise an identity of the manufacturer of network equipment of cell 134-1 and / or the other cells. Tilted cell measured RSRP may comprise the RSRP of a signal of cell 134-1 as measured by users of the othercell(s). For example, network controller 140 may determine affected cells, for which cell 134-1 is the second best cell in terms of RSRP and use these value(s) as the tilted cell measured RSRP. Tilted cell measured SINR in affected cells may comprise an interference level of the signal of cell 134-1 at the other cell(s). For example, network controller 140 may determine the tilted cell measured SINR by subtracting the RSRP of cell 134-1 from the RSRP of the other cell(s), when cell 134-1 has been the second best cell. In other words, this parameter may indicate how much cell 134-1 causes interference in the other cell(s). A signal of cell 134-1 may comprise a signal transmitted by an access node of cell 134-1. Providing RSRP and / or SINR as input data to the ML model improves the prediction of the performance indicator values, because the ML model is enabled to consider the current signal strength and / or interference level in the network in the prediction for different antenna configurations. However, any suitable measure of the received signal strength may be used instead of RSRP.

[0073] The input parameters may be used to determine, e.g., train or obtain, a ML model predicting the performance indicator values. The ML model may be then used to select new antenna configuration(s) for cells such that the performance indicator values are maximized or improved, thereby improving performance of the network as well as user experience. Note that the disclosed methods enable a low- complexity implementation, where input data may contain input data related to data traffic only. The nature of the traffic may be reflected in two input values, the spatial traffic distribution (e.g., timing advance data) of cell 134-1 and optionally other cell(s), and the load of the surrounding network indicated for example by PRB utilization. Additional input parameters may be selected such that they reflect network topology, antenna configurations or properties, or derivatives of those.

[0074] It is however noted that antenna type might not be included in the input data. Including antenna type might come with the risk of the sample data being split into too small quantities. This may be the case for example when there is a large number of different antenna types in the network. Instead of the antenna type, antenna configuration may be taken into account by antenna parameter(s) such as vertical and / or horizontal 3dB beam width, for example together with the 3dB distance.

[0075] FIG. 5 illustrates an example of a flow chart for antenna tilt adjustment. Even though operations of FIG. 5 have been described to be performed by network controller 140, it is understood that similar operations may be performed by any suitable computing device, such as apparatus 200, for example to execute a computer-implemented method for antenna tilting.

[0076] At operation 501, network controller 140 may obtain original input data, for example one or more of the input parameters described above. As illustrated in FIG. 6, the input data may comprise antenna configuration parameters of cell 134-1, such as attenuation level to horizontal direction, 3 dB angle, 3 dB distance, antenna tilt angle, antenna height, or 3 dB beam width. The input data may comprise network topology information such as the nearest cell distance or number of overlapping cells. An overlapping cell may be a cell whose coverage area overlaps with cell 134-1. The input data may comprise spatial traffic distribution information such as TA distribution data (e.g., TA percentile value(s)) of cell 134-land / or the other cell(s). The input data may comprise traffic load level information such as PRB utilization of cell 134-1 and / or other cell(s). The input data may comprise other parameters such as type of propagation environment, frequency band, and / or vendor of network equipment. The frequency band and the vendor may be configuration management parameters. Network controller 140 may be configured to use particular input parameters, for example based on the feature engineering described with reference to FIG. 3.

[0077] The ML model may be configured (e.g., by training) to provide performance indicator values as its output. The output performance indicator values may comprise one or more of the following: CQI, channel rank, spectral efficiency, received signal strength (e.g., RSRP), or data volume. Using received signal strength as an output parameter provides the benefit of enabling signal coverage to be considered when determining the new antenna configuration. For example, the coverage can be estimated to be maintained when a number of poor RSRP samples, or a share of poor RSRP samples, does not increase. An RSRP sample may be considered to be poor when the RSRP value indicates that received signal strength is below a certain threshold. Using data volume as one output enables to ensure no coverage is lost in case of downtilt. Using CQI, spectral efficiency, or channel rankenables to ensure that the quality of communication is considered when determining the new antenna configuration.

[0078] At operation 502, network controller 140 may determine the ML model. Network controller 140 may be configured to apply existing libraries such as scikit- learn to for both building ML models and using them for predicting the performance indicator values. The scikit-learn library, also known as skleam, is a free and open- source machine learning library for the Python programming language, featuring various classification, regression, and clustering algorithms including supportvector machines, random forests, gradient boosting, k-means and DBSCAN (density-based spatial clustering of applications with noise), and it is designed to interoperate with the Python numerical and scientific libraries such as NumPy and SciPy.

[0079] The following pseudocode may be used for determining the ML model. Lines beginning with ‘#’ are comments.1 : # inputs and outputs to own variables2: X, Y = network datafinputs], network data [outputs]3:4: # create polynomial model5: poly = PolynomialFeatures(polymian degree)6:7 : # train model with input- and output-values 8: poly reg model = polynomial_regression() 9: poly_reg_model.fit(X, Y)

[0080] On line 2, the input data and output data are stored in variables X and Y. On line 5, a polynomial model is created. On lines 8 and 9 the ML model is trained with the input and output data. Network controller 140 may be however configured to determine the ML model by obtaining a pre-trained ML model, for example from a memory of network controller 140 or by receiving it from another device, which may have trained the ML model, for example based on the above pseudocode.

[0081] The input data, corresponding to variable X in the pseudocode, may comprise, for example, the input data described with reference to operation 501. The output data (Y) may comprise performance indicator values (e.g., CQI, channelrank, spectral efficiency, received signal strength such as RSRP, or data volume) for cell 134-1 and the other cell(s). The ML model may be therefore configured, e.g., by training, to predict performance indicator values of cells, including cell 134-land the other cell(s). The ML model may be therefore created based on the input parameters such that it is designed to output one or more KPIs indicative of the quality of RAN 120.

[0082] At operation 503, network controller 140 may perform a validity check for the ML model, for example on a cell level. Network controller 140 may for example compare output values of the ML model against real performance of the network. For example, network controller 140 may be configured to determine that a cell is not eligible for new tilt definition based on the ML model, if measured CQI values of the network differ too much from the CQI predicted by the ML model The validity check may be based on thresholds set for each output value comparison.

[0083] For example, network controller 140 may be configured to predict, by evaluation of the ML model with the input data, the performance indicator values of the cells for a current antenna configuration of cell 134-1. The input data may comprise current parameters of the network. Network controller 140 may be configured to determine not to use the ML model for selecting a new antenna configuration, in response to determining that a difference between performance indicator value(s) of the cells predicted for the current antenna configuration and measured performance indicator(s) of the cell(s) associated with the current antenna configuration exceeds a threshold. This threshold may be also referred to as a second threshold and it may be for example set to indicate a 10 % difference between the measured and predicted values. The measured performance indicators may be associated with the current antenna configuration of cell 134-1, e.g., by being measured when the current antenna configuration.

[0084] At operation 504, network controller 140 may check whether the validity check was successful or not. If the threshold was not exceeded, network controller 140 may determine the validity check to be successful. After a successful validity check, network controller 140 may proceed to execution of operation 505. If the threshold is exceeded, network controller 140 may determine the validity check tobe unsuccessful. After an unsuccessful validity check, network controller 140 may end the process or perform operations the flow chart considering another cell to be tilted.

[0085] At operation 505, network controller 140 may select a new antenna configuration for cell 134-1. For example, a new tilt angle be defined for cell 134-1. The new antenna configuration may be defined, based on evaluation of the ML model for different antenna configurations, such that the quality of RAN 120 in terms of the performance indicators is maximized or improved.

[0086] Network controller 140 may be configured to estimate cells affected by adjustment the antenna tilt of cell 134-1. The estimation may be based on location(s) of the transmission site(s) of the other cell(s), e.g., distance from transmission site of cell 134-1, and / or direction of radiation of the cells, as illustrated in FIG. 7. Network controller 140 may be configured to determine the affected cells based on topology of cellular communication network 100, for example by finding cells located within a particular angular range (a) from the transmission site of cell 134-1 (Site A) and / or cells whose transmission site is located within a first predetermined distance (t / i) from Site A. Therefore, in the example of FIG.7 all three cells of Site B served by access node 126 may be considered as affected cells. Network controller 140 may be configured to determine a cell to be an affected cell, if its transmission site is located within a second predetermined distance (tfe) from Site A and if its radiation pattern points towards Site A, e.g., such that the coverage area of the other cell intersects with coverage area of cell 134-1 between their transmission sites. Therefore, in the example of FIG.7 cells 132-1 and 702 of Sites C and D may be considered to be affected cells. Network controller 140 may select the affected cells as the other cells whose traffic load level information and / or spatial traffic distribution information is included in the input and / or output data of the ML model. The set of affected cells may further comprise cell 134-1, because it is also affected by adjustment of its antenna tilt.

[0087] The process of selecting the new antenna configuration at operation 505 is further illustrated in FIG. 8, where operation 801 may be iterated over different antenna configurations and the resulting outputs of the ML model may be compared to select the new antenna configurations. Hence, network controller 140 mayevaluate the ML model with different antenna configurations and select the antenna configuration that maximises, or at least improves, the predicted performance values.

[0088] At each iteration, the input data may comprise one or more of the parameters of FIG. 6, or in general the traffic load level information of the other cell(s) and the spatial traffic distribution information of at least cell 134-1. However, the antenna configuration information of cell 134-1 may be different for different iterations. The ML model may be identical at each iteration, e.g., as determined at operation 502 based on the original input (training) data at operation 501. Network controller 140 may therefore predict, by the ML model based on the input data, the performance indicator values of cell 134-1 and the other cells for different antenna configurations of cell 134-1. Subsequently, network controller 140 may select the new antenna configuration for cell 134-1 based on the performance indicator values predicted for the different antenna configurations of cell 134-1. Network controller 140 may for example select the new antenna configuration based on determining that the new antenna configuration provides best predicted performance among the different antenna configurations.

[0089] Outputs of the ML model may be normalized. This provides the benefit of simplifying use of multiple ML model outputs simultaneously. If there are multiple outputs, network controller 140 may select the antenna configuration with the highest sum of ML model outputs. Network controller 140 may normalize the performance indicator values of the cells predicted for the different antenna configurations. Normalization may comprise dividing a performance indicator value with a respective reference value, for example the highest value of corresponding performance indicator values predicted for different antenna configurations. Different reference values may be therefore used for different output parameters (e.g., CQI, channel rank, spectral efficiency, received signal strength, data volume). Network controller 140 may then sum the normalized performance indicator values for each antenna configuration and select the new antenna configuration for cell 134-1 based on comparing the summed normalized performance indicator values of each antenna configuration, e.g. by selecting the antenna configuration associated with the highest sum.

[0090] Continuing the pseudocode provided above for training the ML model, the ML model may be evaluated for different antenna configurations using the following pseudocode:10: # create input_values for different tilts11 : # network data with different tilt includes network data and differences tilt- 12: # variations for cell.13: input values = network data with different tilt values [inputs]14:15: # predict kpi-values for different tilts16: input_values_poly = poly, transform (input values)17: predicted kpis = poly_reg_model.predict(input_values_poly)

[0091] On line 13, input values are obtained for different antenna configurations and on lines 16 and 17 the performance indicator values (KPI) are predicted based on the input values of different antenna configurations.

[0092] Referring back to FIG. 5, at operation 506, network controller 140 may cause tilting of the antenna of cell 134-1 based on, or according to, the new antenna configuration. Tilting of the antenna may comprise adjusting the antenna tilt according to the new antenna configuration. For example, network controller 140 may transmit instructions to the RET mechanism associated with cell 134-1 to cause the RET mechanism to adjust the antenna tilt of cell 134-1 accordingly. Alternatively, network controller 140 may output an automated service ticket to cause a serviceman to adjust the antenna tilt. The instructions or the automated service ticket may comprise an indication of the new antenna configuration, e.g., by an indication of any suitable parameter of the antenna configuration information, such as for example a new tilt angle, a new -3 dB angle, a new half-power distance, or an indication of the amount of adjustment, for example to uptilt or downtilt the antenna by a particular amount of degrees.

[0093] Causing the antenna tilt may be however conditioned on the expected performance improvement. Accordingly, a threshold may be defined for the predicted improvement. This threshold may be referred to as a first threshold. Network controller 140 may cause the tilting of the antenna of cell 134-1 based on the new antenna configuration, in response to determining that a difference betweenperformance indicator value(s) the cells, including cell 134-1 and the other cells, that have been predicted for the new antenna configuration and measured performance indicator value(s) of the cells exceeds a threshold. In other words, network controller 140 may determine to use the new antenna configuration if it is expected to provide meaningful performance improvement. Otherwise network controller 140 may determine not to use the new antenna configuration. This provides the benefit of avoiding unnecessary adjustment of antenna tilts in RAN 120.

[0094] The measured performance indicator values may be associated with the current antenna configuration of cell 134-1, e.g., by having been measured with the current antenna configuration. Network controller 140 may obtain the measured performance indicator value(s) by receiving them from access nodes of RAN 120. For example, CQI may be required to improve over 5% and if the improvement in predicted CQI values does not satisfy this condition, even for the best antenna configuration, network controller 140 may determine not to cause tilting of the antenna of cell 134-1. The threshold may be however applicable to more than one output parameter of the ML model, for example such that all or a subset of the predicted performance indicators are required to meet the threshold for improvement. Separate thresholds for the improvement may be for example set for each output performance indicator value. The thresholds may vary depending on the communication network being modelled. Alternatively, a common threshold may be set for the sum of the improvements in individual performance indicator values.

[0095] Furthermore, preconfigured limits for antenna configuration parameters of the antenna configuration information may be set to ensure that too radical changes will not be made. For example, upper and / or lower limits, or an acceptable range may be defined for any of the antenna configuration parameters such as the tilt angle of the antenna. Network controller 140 may be configured to cause the tilting of the antenna based on the new antenna configuration, in response to determining that parameter(s) of the new antenna configuration comply with the limit(s).

[0096] At operation 507, network controller 140 may determine whether performance of the network, again in terms of the performance indicator value(s), is acceptable after tilting of cell 134-1. Network controller 140 may for example obtain measured performance indicator values associated with the new antenna configuration of cell 134-1, e.g., performance indicator values measured with the new antenna configuration. Network controller 140 may compare the performance indicator values predicted for the new antenna configuration to the measured performance indicator values. Network controller 140 may determine that performance of the network is acceptable (ok), if the measured performance indicator values are close enough to the predicted performance indicator values, e.g., within a predetermined margin from the predicted performance indicator values. Network controller 140 may determine that performance of the network is not acceptable (not ok), if the measured performance indicator values are not close enough to the predicted performance indicator values, e.g., not within the predetermined margin.

[0097] If the performance is acceptable, network controller 140 may move to execution of operation 509 to obtain adjusted input data for re-training the ML model. If not, network controller 140 may move to operation 508 to cause rollback of the antenna of cell 134-1. This provides the benefit of enabling to ensure that desired performance improvement is achieved, or at least that the antenna tilting of cell 134-1 does not degrade performance of RAN 120. Alternatively, the method may be ended or initiated for another cell without moving to operation 509.

[0098] At operation 508, network controller 140 may cause rollback of the antenna of cell 134-1. This may be in response to determining, at operation 507, that at least one or a group of measured performance indicator value(s) of the network does not meet a condition with the new antenna configuration. Network controller 140 may transmit instructions to the RET mechanism associated with cell 134-1 to cause the RET mechanism to adjust the antenna tilt of cell 134-1 accordingly, e.g., back to the tilt angle of the previous antenna configuration. Alternatively, network controller 140 may output another automated service ticket to cause a serviceman to adjust the antenna tilt accordingly.

[0099] At operation 509, network controller 140 may obtain adjusted input data. Network controller 140 may receive, e.g., from access nodes of the network, updated values of the input data resulting from the new antenna configuration. Subsequently, at operation 502, network controller 140 may update the ML model (e.g., by determining a new ML model) based on the input data comprising the new antenna configuration and performance indicator values measured with the new antenna configuration. Hence, performance of the ML model may be improved by training it with further data after deploying the new antenna configuration in the network. In case the performance was not acceptable at operation 507, e.g., the quality of the network in terms of performance indicator values having not improved, the actual measured input values resulting from the antenna tilt change of operation 506 may be used as adjusted inputs to improve the ML model at operation 502. The process may be repeated with new, adjusted, inputs, for example until a desired improvement in the measured performance indicator values is achieved.

[0100] FIG. 9 illustrates an example of determining a machine learning model for predicting timing advance distribution value(s). Network controller 140 may be configured to use a second ML model for predicting TA distribution value(s) of cell 134-1, for example as part of operation 505. Network controller 140 may provide the predicted TA distribution value(s) as the spatial traffic distribution information to the (first) ML model, which is configured to predict the performance indicator values.

[0101] The second ML model may be determined (e.g., trained) based on input data similar to FIG. 6. However, in this case the input data may not comprise the spatial traffic distribution of cell 134-1 (e.g., TA distribution value(s)), because this information is to be predicted by the second ML model. The input data of the second ML model may be referred to as second input data. The second ML model may be configured to provide as its outputs the TA distribution values of cell 134-1, for example particular TA percentile value(s) (e.g., 50thand 90thpercentile values as in the figure). The second ML model may be trained using the pseudocode described above, but with appropriate input and output data. Note that when a ML model istrained, it is aware of both the inputs and the outputs and therefore the ML model may be fitted for providing desired outputs even for unseen input data.

[0102] When predicting performance indicator values for a new antenna configuration in a current network implementation, varying parameters of the antenna configuration used as input values to the first ML mode may include the tilt angle of the antenna of cell 134-1, the half-power distance of the radiation pattern of the antenna, and / or the half-power angle of the vertical radiation pattern of the antenna. Values of these varying configuration parameters may be dependent on the antenna configuration. In addition, the second ML model may be used to predict the change in the TA distribution associated with the change of antenna tilt. The predicted change in the TA distribution may be provided as a varying input value to the first ML model, for example if it is determined that the prediction reliability of the second ML model is good enough.

[0103] Network controller 140 may predict, by the second ML model, the TA value distribution value(s), for the current antenna configuration. Network controller 140 may predict the TA distribution value(s) with the second ML model for new antenna configuration(s) and use the predicted TA distribution value(s) as input for the first ML model, in response determining that a difference between the performance indicator values predicted for the current antenna configuration and measured performance indicator values associated with the current antenna configuration is below a threshold (e.g., < 10 %). This threshold may be referred to as a third threshold.

[0104] For example, reliability of TA distribution prediction may be determined by calculating the difference between particular value(s) of the predicted TA distribution (e.g., 50thpercentile and / or 90thpercentile values) and corresponding TA distribution values observed in cell 134-1 with the current antenna configuration. Predicting the TA distribution provides the benefit of enabling to estimate the effect of antenna tilting for different antenna configurations, which improves prediction accuracy of the first ML model, because the effect of antenna tilting is taken into account in the spatial traffic distribution information provided as part of the input data to the first ML model.

[0105] A numerical example is provided below for describing some example embodiments disclosed herein:

[0106] In this example, the original antenna tilt of a cell is 6 degrees. The predicted and measured performance indicator values may be first compared for the original antenna configuration, in order to determine whether the prediction of the first ML model is good enough for using it in determination of the new antenna configuration. In this example, measured CQI = 9 and predicted CQI = 8,9 for the original antenna tilt. Based on the difference between the predicted and measured CQI values, network controller 140 may determine that the prediction is reliable enough in terms of quality. Further, measured data volume is 0.96 GB and predicted data volume is 0.98 GB for the original antenna tilt. Based on the difference between the measured and predicted values of the data volume, network controller 140 may determine that the prediction is reliable enough in terms of data volume. Further, the measured 50thTA percentile is 0.5 km and the predicted 50thTA percentile is 0.52 for the original antenna tilt. Based on the difference between the measured and predicted TA distributions, network controller 140 may determine that TA distribution prediction by the second ML model is reliable enough. Hence, the predicted TA distribution values can be used as inputs for the first ML model.

[0107] Network controller 140 may then predict performance indicator values a candidate antenna configuration defined by 5-degree tilt angle, with the following parameters:- varying input values:3dB angle: 2.5 deg,3dB distance: 0.7 km, antenna tilt: 5 deg, predicted TA 50th percentile: 0.6 km other input values: nearest cell distance: 0.4 km, antenna height: 30 m, 3dB beam width: 6 deg, PRB utilization: 30%, overlapping cells: 4.With these input values, the ML model may output a predicted new CQI = 8,7 and predicted data volume = 1 GB for the 5 degree tilt.

[0108] The new CQI may be then predicted also for one or more other candidate antenna configurations with different tilt values. For example, the ML model may be evaluated in order to predict the CQI for 7-degree tilt angle with the following parameters:- varying input values:3dB angle: 4.5 deg,3dB distance: 0.4 km, antenna tilt: 7 deg, predicted TA 50th percentile: 0.42 km other input values: nearest cell distance: 0.4 km, antenna height: 30m,3dB beamwidth 6 deg,PRB utilization 30%, overlapping cells: 4.With these input values, the ML model may a predicted new CQI = 10 and predicted data volume = 0.97 GB for the 7-degree tilt angle.

[0109] The predicted performance indicator values may be then normalized such that the normalized values of the outputs become 0.87 for CQI of the 5-degree tilt angle, 0.89 for the 6-degree tilt angle, 1 for the CQI of the 7-degree tilt angle, 1 for the data volume of the 5-degree tilt angle, 0.98 for the data volume of the 6-degree tilt angle and 0.97 for the data volume of the 7-degree tilt angle.

[0110] The normalized values may be then summed separately for each antenna configuration. In this example, the sum is 1.87 for the 5-degree tilt angle, 1.87 for the 6-degree tilt angle, and 1.97 for the 7-degree tilt angle. Based on comparison of the summed values, network controller 140 may determined that the best predicted performance is achieved with the tilt angle of 7 degrees as 1.97 > 1.87. Further, the improvement compared to the original antenna tilt is > 5% based on the summed values. Hence, network controller 140 may determine to select the antenna configuration with the 7-degree tilt angle as the new antenna configuration.[01 1 1 ] Antenna configurations may be similarly determined also for new transmission sites. For new transmission sites the input parameters may be known apart from the spatial traffic distribution. This information may be however estimated by network controller 140 utilizing the second ML model. The predicted TA distribution, combined with the rest of the network configuration features may be then applied for the new transmission site and used in the tilt definition for cells.

[0112] FIG. 10 illustrates an example of predicted and measured channel quality indicator values. Each point of the scatter plot indicates the measured CQI value and the corresponding predicted CQI value. For example, R2, values of over 0.6 are obtained for the CQI value using the current input features that use almost only topology. The R2-value may indicate how well the statistical model explains the predicted data. Furthermore, the prediction accuracy has been observed to improve when using the predicted TA distribution as inputs for the first ML model.[01 1 3] FIG. 11 and FIG. 12 illustrate CQI and spectral efficiency in cell 134-1 (black dots) and affected cells (white dots) before and after adjusting antenna tilt of cell 134-1 from four to seven degrees. It is observed that performance is improved after the tilt change on Thursday 06.06. both in terms of CQI and spectral efficiency.

[0114] FIG. 13 illustrates an example of a method for determining new antenna configuration for a cell in a communication network[01 1 5] At operation 1301, the method may comprise determining a machine learning model configured to predict performance indicator values of a plurality of cells of a cellular communication network, wherein the plurality of cells comprises a first cell and one or more other cells, and wherein the machine learning model is configured to receive input data comprising: network topology information of the plurality of cells, spatial traffic distribution information of at least the first cell, and antenna configuration information of the first cell, wherein the antenna configuration information is indicative of a tilt angle of an antenna of the first cell.

[0116] At operation 1302, the method may comprise predicting, by the machine learning model based on the input data, the performance indicator values of the plurality of cells for a plurality of antenna configurations of the first cell.[01 1 7] At operation 1303, the method may comprise selecting a new antenna configuration for the first cell based on the performance indicator values of theplurality of cells predicted for the plurality of antenna configurations of the first cell.

[0118] At operation 1304, the method may comprise causing tilting of the antenna of the first cell based on the new antenna.

[0119] Further features of the method directly result for example from the functionalities of network controller 140, or in general apparatus 200, as described throughout the specification and in the appended claims, and are therefore not repeated here. Different variations of the method may be also applied, as described in connection with the various example embodiments.

[0120] An apparatus, such as for example a network device configured to implement one or more network functions or entities, may be configured to perform or cause performance of any aspect of the method(s) described herein. Further, a computer program or a computer program product may comprise instructions for causing, when executed, an apparatus to perform any aspect of the method(s) described herein. Further, an apparatus may comprise means for performing any aspect of the method(s) described herein. According to an example embodiment, the means comprises at least one processor, and memory including program code, the at least one processor, and program code configured to, when executed by the at least one processor, cause performance of any aspect of the method(s).

[0121] 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.

[0122] 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.

[0123] 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 orthose 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.

[0124] The steps or operations 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 scope of the subject matter described herein. Aspects of any of the example embodiments described above may be combined with aspects of any of the other example embodiments described to form further example embodiments without losing the effect sought.

[0125] 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.

[0126] Although subjects may be referred to as ‘first’ or ‘second’ subjects, this does not necessarily indicate any order or importance of the subjects. Instead, such attributes may be used solely for the purpose of making a difference between subjects.

[0127] 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 embodiments, those skilled in the art could make numerous alterations to the disclosed embodiments without departing from scope of this specification.

Claims

CLAIMS1. A computer-implemented method, comprising: determining (1301) a machine learning model configured to predict performance indicator values of a plurality of cells of a cellular communication network, wherein the plurality of cells comprises a first cell and one or more other cells, and wherein the machine learning model is configured to receive input data comprising: network topology information of the plurality of cells, spatial traffic distribution information comprising at least one timing advance value distribution value of at least the first cell, and antenna configuration information of the first cell, wherein the antenna configuration information is indicative of a tilt angle of an antenna of the first cell; predicting (1302), by the machine learning model based on the input data, the performance indicator values of the plurality of cells for a plurality of antenna configurations of the first cell; selecting (1303) a new antenna configuration for the first cell based on the performance indicator values of the plurality of cells predicted for the plurality of antenna configurations of the first cell; and causing (1304) tilting of the antenna of the first cell based on the new antenna configuration.

2. The method according to claim 1, wherein the antenna configuration information or the new antenna configuration comprises one or more of the following: the tilt angle of the antenna of the first cell, a half-power distance of a radiation pattern of the antenna of the first cell, a half-power beam width of the radiation pattern of the antenna of the first cell, a half-power angle of a vertical radiation pattern of the antenna of the first cell in a horizontal direction of a main lobe of the radiation pattern of the antenna of the first cell, oran attenuation level of a horizontal direction of the vertical radiation pattern in the horizontal direction of the main lobe of the radiation pattern of the antenna of the first cell.

3. The method according to claim 1 or 2, wherein the performance indicator values comprise one or more of the following: a channel quality indicator, a multiple-input multiple output channel rank indicator, a spectral efficiency, or a data volume.

4. The method according to any of claims 1 to 3, wherein the input data further comprises: a received signal strength of a signal of the first cell as measured by users of the one or more other cells, or an interference level of the signal of the first cell at the one or more other cells.

5. The method according to any of claims 1 to 4, further comprising: causing the tilting of the antenna of the first cell based on the new antenna configuration, in response to determining that a difference between at least one performance indicator value of the plurality of cells predicted for the new antenna configuration and at least one measured performance indicator value of the plurality of cells associated with a current antenna configuration of the first cell exceeds a first threshold.

6. The method according to any of claims 1 to 5, further comprising: predicting, by the machine learning model based on the input data, the performance indicator values of the plurality of cells for a current antenna configuration of the first cell; and determining not to use the machine learning model for selecting the new antenna configuration, in response to determining that a difference between at leastone performance indicator value of the plurality of cells predicted for the current antenna configuration and at least one measured performance indicator of the plurality of cells associated with the current antenna configuration exceeds a second threshold.

7. The method according to any of claims 1 to 6, further comprising: updating the machine learning model based on input data comprising the new antenna configuration and performance indicator values measured with the new antenna configuration.

8. The method according to any of claims 1 to 7, further comprising: determining a second machine learning model configured to predict at least one timing advance distribution value of the first cell, wherein the second machine learning model is configured to receive second input data comprising: traffic load level information of the one or more other cells, and the antenna configuration information of the first cell, wherein the second input data does not comprise the spatial traffic distribution information of the first cell; predicting, by the second machine learning model based on the second input data, the at least one timing advance distribution value of the first cell; providing the predicted timing advance value distribution value of the first cell to the machine learning model as the spatial traffic distribution information of the first cell for prediction of the performance indicator values of the plurality of cells.

9. The method according to 8, wherein the second input data further comprises network topology information.

10. The method according to claim 9, wherein the network topology information comprises: a minimum distance between a transmission site of the first cell and any other transmission site of the cellular communication network, ora number of cells overlapping with the first cell.

11. The method according to any of claims 1 to 10, wherein the antenna configuration information comprises: a height level of the antenna of the first cell.

12. The method according to any of claims 8 to 11, wherein the traffic load level information of the one or more other cells comprises physical resource block utilization rate values of the one or more other cells.

13. The method according to any of claims 1 to 12, wherein the input data to the machine learning model further comprises at least one of: traffic load level information of the one or more other cells, a type of propagation environment of the plurality of cells, a frequency band of the plurality of cells, or an identity of a manufacturer of network equipment of the plurality of cells.

14. The method according to any of claims 1 to 13, further comprising: selecting the new antenna configuration based on determining that the new antenna configuration provides best predicted performance among the plurality of antenna configurations.

15. The method according to any of claims 1 to 14, further comprising: normalizing the performance indicator values of the plurality of cells predicted for the plurality of antenna configurations; summing normalized performance indicator values for each antenna configuration of the plurality of antenna configurations; and selecting the new antenna configuration based on comparing the summed normalized performance indicator values of each antenna configuration of the plurality of antenna configurations.

16. An apparatus (200) comprising means for performing the method according to any of claims 1 to 15.

17. The apparatus (200) according to claim 16, wherein the means comprises at least one processor (202); and at least one memory (204) storing instructions that, when executed by the at least one processor (202), cause the apparatus (200) to perform the method according to any of claims 1 to 15.

18. A computer program comprising instructions configured to, when executed by an apparatus (200), cause the apparatus (200) at least to perform the method according to any of claims 1 to 15.

Images