A method for reducing energy consumption in a network

Abstract

A method for reducing energy consumption in a network is provided. A trained machine learning model to applied input data to select one or more network energy saving techniques that are predicted to result in a network energy consumption less than a threshold consumption amount, the trained machine learning model having been trained on historical data received from one or more nodes of the network. The historical data comprises a respective network energy consumption associated with each of one or more implemented network energy saving techniques. The selected one or more network energy saving techniques are implemented in at least one node of the network.

Description

[0001] A Method for Reducing Energy Consumption in a Network

[0002] FIELD

[0003] The disclosure relates generally to the field of networks, such as, for example, cellular networks. More particularly, the disclosure relates to a method for reducing energy consumption in a network, a computer program and a computer-readable storage medium.

[0004] GLOSSARY

[0005] 3GPP - Third Generation Partnership Project

[0006] Al - Artificial Intelligence

[0007] BS - Base Station

[0008] DRX - Discontinuous Reception

[0009] DTX - Discontinuous Transmission eNB - evolvedNodeB

[0010] E-UTRAN - Evolved Universal Terrestrial Radio Access Network gNB - gNodeB

[0011] ML - Machine Learning

[0012] NR - New Radio (5G)

[0013] RAN - Radio Access Network

[0014] RF - Radio Frequency

[0015] QoE - Quality of Experience

[0016] QoS - Quality of Service

[0017] UE - User Equipment

[0018] BACKGROUND

[0019] Network energy saving is important in many different types of networks, including computer networks and mobile (cellular) networks. For example, large computer networks, such as those used to support cloud-based services, can result in significant energy consumption. In another example, wireless networking devices are generally battery-powered, and therefore require energy efficient computing infrastructure to make the most of the limited source of energy.

[0020] Cellular network energy saving, such as Radio Access Network (RAN) energy saving for example, has been of particular interest in the telecommunications industry in recent years. The RAN aims to provide nation-wide coverage and service capacity, which requires a large number of radio sites from which signals can be transmitted and received. The number of radio sites is typically significantly larger than the number of core network nodes (often by a factor of 10,000 or more). Thus, the RAN represents a significant portion of a service provider’s network power consumption. RAN energy efficiency is therefore an important factor in enabling network energy savings. However, this RAN energy efficiency should not come at the cost of sufficient coverage and / or capacity of the network.

[0021] Careful design of network planning can reduce network energy consumption. Nevertheless, further network energy saving in various networks, including active radio networks, is desirable. However, implementing appropriate network energy saving techniques is challenging and complex, due to the various factors that that can impact the network, such as traffic load, mobility and advanced applications (for example, in extended reality applications).

[0022] Overcoming the issues noted above is desirable.

[0023] SUMMARY

[0024] Against this background, there is provided a method for reducing energy consumption in a network. Additional aspects appear in the description and claims.

[0025] In accordance with a first aspect, there is provided a method for reducing energy consumption in a network, the method comprising: applying a trained machine learning model to input data to select one or more network energy saving techniques that are predicted to result in a network energy consumption less than a threshold consumption amount, the trained machine learning model having been trained on historical data received from one or more nodes of the network, the historical data comprising a respective network energy consumption associated with each of one or more implemented network energy saving techniques; and implementing the selected one or more network energy saving techniques in at least one node of the network.

[0026] Thus, it is possible to determine a network energy saving technique that can provide a network energy consumption less than a threshold consumption amount for a given scenario. This may enable an increase network energy saving.

[0027] In one implementation, the network is a cellular network. Optionally, implementing the selected one or more network energy saving techniques in the at least one node of the network may comprise implementing the selected one or more network energy saving techniques in at least one cell in the cellular network. In one implementation, the cellular network may comprise a Radio Access Network (which may include, for example, a GRAN, GERAN, UTRAN, E-UTRAN, and / or another type of RAN).

[0028] In another implementation, the historical data further may comprise one or more network configuration parameters for implementing a respective one of the one or more implemented network energy saving techniques. Accordingly, applying the model may comprise determining a set of network configuration parameters for implementing the selected one or more network energy saving techniques based on the input data. Implementing the selected one or more network energy saving techniques in the at least one node may then comprise operating the at least one node according to the set of network configuration parameters.

[0029] The historical data including the one or more network configuration parameters may enable more accurate determination of the network energy saving technique. For example, a network energy saving technique may only provide a network energy consumption less than a threshold value for certain network configuration parameters. Thus, by determining the network configuration parameters based on the historical data, an appropriate implementation of the network saving technique can be achieved (rather than, for example, implementing the network saving technique with typical, average or pre-determined network configuration parameters, which may not provide the necessary energy saving).

[0030] Optionally, the input data comprises and / or the historical data further comprises information indicating network characteristics. Accordingly, selecting the one or more network energy saving techniques may comprise selecting the one or more network energy saving techniques for which the information indicating network characteristics of the historical data is the same as, sufficiently similar to or most similar to the information indicating network characteristics of the input data. Since the network energy saving provided by a technique may depend on the network characteristics, training the machine model on historical data including the network characteristics may provide a model that can more accurately predict a network energy saving technique resulting in a network energy consumption less than the threshold amount. This may in turn result in more improved network energy savings, since, for example, an incorrect network energy saving technique is less likely to be implemented.

[0031] Optionally, the information indicating network characteristics may comprise one or more of: a network load; one or more network traffic characteristics; one or more capabilities of a node in the network; a quality of service, QoS, parameter; a quality of experience, QoE, parameter; a time of day; and, when the network is a cellular network and the node comprises a network cell, neighbouring cell information.

[0032] In one implementation, applying the model further may comprise selecting one or more network energy saving techniques that are predicted to result in a QoS and / or QoE parameter above a threshold level. Network energy saving techniques can come at the risk reduced or limited QoE and / or QoS. This is undesirable in many scenarios, such as extended reality scenarios or live-streaming scenarios, where even a slight delay in signals can disorient or irritate the user. Thus, a network energy saving technique that provides a network energy consumption less than a threshold consumption amount and a QoS and / or QoE parameter above a threshold level is useful.

[0033] In another implementation, applying the model may further comprise selecting one or more network energy saving techniques that optimise the network energy consumption and the QoS and / or QoE parameter. Thus, a high quality user experience can be provided for a given network energy consumption.

[0034] The QoS parameter may comprise one or more of the QoS parameters described is described in 3GPP Technical Specification (TS) 23.203 (in Section 6.1.7.2, for example). For example, the QoS parameter may include a QoS Class Identifier (QCI) characteristic, an Allocation and Retention Priority (ARP) characteristic, a guaranteed bitrate (GBR) and / or a maximum bit rate (MBR). More specifically, the QCI may include a resource type (GBR or non-GBR), a priority level, packet delay budget, packet error loss rate, maximum data burst volume, data rate average window. The ARP characteristic may include information regarding the priority level, the pre-emption capability and the pre-emption vulnerability. A one-to-one mapping of the standardised QCI parameters to standardised characteristics is provided in Tables 6.1 .7-A and B in 3GPP TS 23.203. Other QoS parameters are possible. For example, the QoS parameter may also or instead comprise one or more of: a reference signal received power, RSRP; a received signal strength indicator, RSSI; a reference signal received quality, RSRQ; a reference signal time difference, RSTD; a line-of-sight, LOS, indicator or a non-line-of-sight, NLOS, indicator; a network throughput; a signal-to-noise ratio, SNR; block error rate, BLER; error vector magnitude, EVM; and channel state information, CSI, parameters. The CSI parameters may comprise a channel quality indicator, CQI; a precoding matrix indicator, PMI; SS / PBCH resource block indicator, SSBRI; a layer indicator, LI; CSI reference signal resource indicator; and / or a rank indicator, RL The QoS parameter used for selecting the one or more network energy saving techniques may be selected depending on the network scenario (for example, the type of network).

[0035] The QoE parameter may comprise one or more of the QoE parameters described in 3GPP TS 26.247 (in Section 10, for example). For instance, the QoE parameter may include an average throughput, an initial playout delay, a buffer level, a play list or playout, a playout delay for media start-up, device information, and so on. Other QoE parameters are possible. For example, the QoE parameter may also or instead comprise one or more of: a user satisfaction indicator; information regarding a stall event; a playback time; a rebuffering ratio; an indication of a failure; an average media bit rate; a start time; and jitter. The information regarding the stall event may comprise an indication that a stall event has occurred, a stalling ratio, a frequency of stall events, a duration of the stall event and / or a duration of one or more previous stall events, a time duration since a previous stall and / or a total number of stalls. Optionally, the input data may comprise real-time data received from one or more nodes of the network and / or predicted data for the one or more nodes of the network. Using real-time data enable the model to accurately select a network energy saving technique that can result in a network energy consumption below a threshold consumption amount. In turn, the network energy consumption of the network can be more accurately controlled.

[0036] The selected one or more network energy saving techniques may comprise a combination of two or more network energy saving techniques. Thus, further network energy savings may be possible. For example, the individual techniques may not be sufficient to provide a network energy consumption below the threshold level, but it may be determined that the combination of techniques is sufficient.

[0037] In one implementation, applying the model may comprise comparing a predicted network energy consumption of each of the one or more implemented network energy saving techniques and / or a combination of the one or more implemented network energy saving techniques to select the one or more network energy saving techniques, wherein the predicted network energy consumption is predicted based on the input data. Additionally or alternatively, applying the model may comprise computing an estimated energy saving of each of the one or more network energy saving techniques and / or a combination of the one or more implemented network energy saving techniques based on the respective associated network energy consumption. Thus, the network energy saving techniques predicted to result in a network energy consumption less than a threshold amount can be determined in a straightforward manner.

[0038] Optionally, computing the estimated network energy saving may comprise computing an estimated energy saving of a group of cells by calculating a power usage of a cell in which the one or more network energy saving techniques were implemented and a power usage of one or more neighbouring cells. In another implementation, predicting the network energy consumption by computing a predicted network energy consumption of a group of cells by calculating a predicted power usage of the at least one cell and a predicted power usage of one or more adjacent cells. Thus, cases can be identified where, although an individual cell may achieve a network energy consumption lower than the threshold value, the overall network energy consumption may nevertheless increase. Accordingly, a more accurate network energy technique selection is possible, which in turn may improve the network energy consumption.

[0039] In an implementation, comparing the predicted network energy consumption of each of the one or more implemented network energy saving techniques to select the one or more network energy saving techniques comprises selecting the one or more network energy saving techniques predicted to result in the lowest network energy consumption and / or the greatest network energy saving. Thus, the network energy consumption and / or network energy saving can be improved.

[0040] In another implementation, the method may further comprise, following the step of implementing the selected one or more network energy techniques, receiving further input data from the at least one node and applying the model to the further input data to select a further one or more network energy saving techniques. Receiving further input data and applying the model to the further input data allows the network energy consumption to remain below the threshold consumption amount, even when network circumstances change.

[0041] Optionally, when the selected further one or more network energy saving techniques differ from the selected one or more network energy saving techniques, the method may further comprise implementing the selected further one or more network energy saving techniques in the at least one node. Thus, the network consumption can be adjusted dynamically based on real-time information.

[0042] In a further implementation, applying the model to the further input data may comprise determining a further set of network configuration parameters for the selected further one or more network energy saving techniques based on the further input data. When the further set of network configuration parameters differs from the set of network configuration parameters by more than a threshold amount, the method may further comprise adjusting the operation of the network according to the further set of network configuration parameters.

[0043] The one or more implemented network energy saving techniques and / or the selected one or more network energy saving techniques may comprises one or more of: activating or deactivating a base station in the at least one cell for a period of time; implementing discontinuous transmission and / or discontinuous transmission in at least one cell of a cellular network; enabling or disabling a frequency layer for a time period; enabling or disabling an RF channel for a time period; activating or deactivating a secondary cell for a time period; and deactivating the at least one cell based on low cell activity. The discontinuous transmission and / or discontinuous reception may be implemented by switching one or more carriers on and / or off at the symbol level, subframe level or radio frame level.

[0044] Optionally, the method may further comprise training the machine learning model based on the historical data received from one or more network nodes.

[0045] In one implementation, training the machine learning model may comprise computing an estimated network energy saving of each of the one or more network energy saving techniques based on the network energy consumption. This may provide a straightforward manner of enabling accurate prediction of which network energy saving technique(s) will result in a network energy consumption less than the threshold consumption value.

[0046] The methods described above may be implemented as a computer program comprising instructions to operate a computer or computer system. The computer program may be stored on a computer-readable storage medium (which may be, for example, a non- transitory computer-readable medium).

[0047] The above methods may be implemented in a system comprising a network comprising at least one node and a computing device configured to operate the at least one node according to the selected one or more network energy saving techniques.

[0048] It should be noted that any feature described herein may be used with any particular aspect or embodiment of the invention. Moreover, the combination of any specific apparatus, structural or method features is also provided, even if that combination is not explicitly disclosed.

[0049] The invention will now be described with reference to the attached drawings depicting different embodiments thereof, the drawings being provided purely by way of example and not limitation.

[0050] BRIEF DESCRIPTION OF DRAWINGS

[0051] The invention may be put into practice in a number of ways, and preferred embodiments will now be described by way of example only and with reference to the accompanying drawings, in which:

[0052] Figure 1 illustrates an exemplary cellular network in which the methods disclosed herein may be implemented;

[0053] Figure 2 illustrates a flowchart diagram of a method for reducing energy consumption in a network;

[0054] Figure 3 illustrates a plurality of nodes in which network energy saving techniques have been implemented and an associated network energy consumption of the plurality of nodes;

[0055] Figure 4 shows an example of training and using a machine learning model for selecting a network energy saving technique; and

[0056] Figure 5 illustrates an example playout of media for which a number of different quality of experience parameters may be assessed.

[0057] It should be noted that the Figures are illustrated in schematic form for simplicity and are not necessarily drawn to scale. Like features are provided with the same (or similar) reference numerals. DESCRIPTION OF PREFERRED EMBODIMENTS

[0058] There are many techniques designed by 3GPP and other standards bodies, as well as implementation-based mechanisms, available for network energy savings. For example, in Release 17, the Third Generation Partnership Program (3GPP) considered network energy saving at different levels in the network architecture. An implementation-based approach may be used for intra-eNB scenarios. Intra-eNB scenarios relate handovers between cells when a UE moves from one sector (or cell) managed by an eNB to another sector (or cell) managed by the same eNB. An implementation-based approach may also be used in inter- eNB scenarios - for example, if both eNBs are provided by or controlled by the same vendor (proprietary mechanism or an Operation Administration and Maintenance, OAM, based mechanism). In inter-eNB and intra-eNB scenarios, inter-eNB or intra-eNB signalling (via a transport interface used to connect eNodeB, such as the X2 interface, for example) may enable switching an eNB on or off in both dual connectivity (for instance, 4G and 5G) and standalone network deployment scenarios. Inter-eNB scenarios relate to handovers between cells when a UE moves from a sector (or cell) managed by one eNB to a sector (or cell) controlled by a neighbouring (different) eNB. Inter-RAT signalling (which may be over a transport interface, such as S1 , for example) to enable eNB on / off was also studied.

[0059] Other example techniques include switching one or more frequency carriers on / off (shutting down a frequency layer for a period of time), switching one or more RF channel on / off, activation / deactivation of secondary cell (in a dual connective scenario), and cell DRX / DTX (time domain) and power domain techniques (which may include adjusting a transmission power with multi-input multiple-output, MIMO, layer adjustments). Recent developments in DRX / DTX operation have looked into network energy saving techniques for short timescales, such as symbol-level, subframe-level or radio frame-level switching on / off of carriers. As would be understood by the skilled person, a frame is divided into a plurality of subframes, which in turn are divided into a plurality of slots. Each slot is divided into a plurality of symbols, which is the smallest level of granularity (1 ms). In DTX / DRX, a cell is on and serving UEs in the cell. A DTX / DRX pattern is provided to the UE so that the UE is informed of when a service can be received and when to access the network. A delay in a change in configuration is much reduced in a DRX / DTX pattern-based method compared to, for example, cell on / off methods. Thus, improved service may be provided for the UEs. Symbol level and subframe level DTX / DRX may not be visible to the UE.

[0060] Operations and maintenance based mechanisms for network energy saving were also considered in Release 17. For example, downlink transmission of a cell may be turned off or the cell may be configured as a deactivated (dormant) cell based on the low activity of the cell (for example, a low traffic load). The cell may be turned off by shutting down a base station for a period of time. The dormant cell may then be reactivated (turned on) based on a trigger condition - for example, packet arrival or UE arrival in the cell. The deactivated cell transmits periodic discovery signals. Thus, the UE may be configured to measure the discovery signal for radio resource management. The investigation was addressed under self-optimisation network (SON) and functionality for cell activation / deactivation for energy saving and is described in 3GPP Technical Specification (TS) 36.423 and 3GPP TS 36.413. The main features addressed in this technique are as follows:

[0061] - An eNB hands over to another cell and the source cell is switched off.

[0062] Peer (neighbouring) eNBs are informed that the source cell has been switched off. This may be implemented using an eNB configuration update procedure - for example, via the X2 interface. In another example this may be implemented using an eNB direct information transfer procedure - for example, via the S1 interface.

[0063] The neighbouring cells maintain the cell configuration of the dormant cell.

[0064] The dormant cell is reactivated when required (for example, when a UE moves into the cell) by a non-capacity boosting cell. The activation procedure may be implemented over the X2 interface or S1 interface.

[0065] There may be minimum time before the source cell is switched off to prevent idle mode UE camping and incoming handover.

[0066] Network energy saving has also been investigated for NR networks, which is described in 3GPP TS 38.300. Similarly to the case of an Evolved Universal Terrestrial Radio Access Network (E-UTRAN), network energy saving is enabled in NR based on traffic load, UE arrival, and packet arrival. Multi-RAT Dual Connectivity (MR-DC), centralised unit- distributed unit (CU-DU) split architecture, and virtual network functions (that is, having fewer hardware units) are scenarios where network energy saving has been investigated for NR. Various features such as massive multi-input multiple-output (MIMO), 5G NR reference signals (for example, temporary reference signals, T-RS, or demodulation reference signals, DMRS), dormant secondary cells and deactivated secondary cell groups can also be used to provide network energy savings. DMRS are specific reference signals transmitted by a base station (BS) and may be cell-specific or UE-specific. The DMRS are used by the UE for (radio) channel estimation and frequency synchronisation. Beamformed DMRS may be used - that is, beamforming may be used to transmit some reference signals but not others. DMRS is operated within scheduled resources and are transmitted only when necessary.

[0067] Open-RAN Working Groups have also looked into network energy saving. Some aspects investigated are discussed in the O-RAN Work Group 1 Network Energy Saving Use Cases Technical Report 3 version 1 .00 are carrier and AI / ML assisted cell switch off / on and AI / ML based RF channel reconfiguration.

[0068] Whilst many techniques exist for network energy saving, existing systems typically implement cell on / off without determining whether this technique would result in a network energy saving for the given scenario. That is, existing systems do not seek to identify which network energy technique would provide a network energy saving for the given scenario, or which network energy saving technique would provide an optimal network energy saving. This is because the resulting network energy saving from these techniques may depend on many factors, including a network deployment scenario, network load, traffic characteristics, network node capabilities, a quality of service (QoS) requirement, and a quality of experience (QoE) requirement. Furthermore, these factors may vary in a time-dependent and / or location-dependent manner, or may otherwise not remain constant. The network deployment scenario may include parameters such as a location of the device, a type of device (for example, a firewall, switch, User Equipment, remote radio head (also known as a remote radio unit in wireless networks) and so on), an application of the device (for example, in extended reality, XR, or live-streaming), a type of cell or site (for example, a microcell / microsite, a small cell, which may include femtocells, picocells and microcells, or another type of cell or site), a type of network (for example, a heterogeneous network, standalone network, a dual connectivity network, or another type of network), information regarding a network implementation (for example, that carrier aggregation is being implemented, that high frequency and low frequency bands are being implemented (for example, FR1 and FR2 in 5G NR), or other information regarding the implementation), for instance.

[0069] Thus, identifying a most appropriate or optimal technique or techniques for network energy saving for a given scenario is a complex procedure. If an inappropriate or sub- optimal action is taken (for example, switching off a wrong cell), this may seriously deteriorate network performance. For instance, a wrong offloading of traffic could, in fact, increase the overall network energy consumption and / or potentially deteriorate the service quality. In other words, although each of the specified network energy saving (NES) techniques can enable energy savings, identifying which NES technique (NEST) would provide the an energy saving for a given scenario, particularly an optimal energy saving, is not straightforward, due to the dynamic nature of the load, user traffic, and other cell involvements, and so on. Thus, efficient and stable utilization of network energy saving features requires careful implementation and optimisation.

[0070] In view of these issues, the present disclosure relates to a method using an Artificial Intelligence (Al) / Machine Learning (ML) based mechanism to predict a technique that will provide a network energy saving at a given time and for a given scenario. In particular, a trained machine learning model is applied to input data to select a network energy technique (or combination of techniques) that is predicted to result in a network energy saving. The selected one or more techniques is then implemented in the network (for example, in at least one node of the network). An example cellular network 100 in which the proposed methods may be implemented is illustrated in Figure 1 . The cellular network comprises a plurality of cells 140A, 140B, 140C. Although three cells are illustrated in Figure 1 , it will be appreciated that more (or fewer) cells may be present in the network. A transceiver or base station 120A, 120B, 120C is positioned in each cell 140A-C. The base stations 120A-C may be gNBs or eNBs, or may be another type of node (for example, for 6G cellular network technologies and / or beyond). Each base station 120A-C can contain one or more radio units that typically operate on different frequencies or radio bands. Furthermore, each radio unit or remote radio unit (RRU) may provide cellular services for different cellular network technologies (for example, 2G, 3G, 4G, and / or 5G or beyond).

[0071] Although not shown in Figure 1 , additional network entities may be present in the cells. For example, one or more of the plurality of cells 140A-C may comprise secondary cells. A secondary cell is a typically a smaller cell within a master cell that can provide a capacity boost to the master cell. Secondary cells may be used in carrier aggregation (combining multiple RF carriers into a single logical channel) or dual connectivity where the secondary cell is connected to the master cell via a (ideal or non-ideal) network interface. One or more of the plurality of cells 140A-C may also (or instead) comprise additional transceivers or base stations. For example, three transceivers may typically be implemented in the plurality of cells 140A-C.

[0072] Referring to Figure 2, there is provided a flowchart diagram of a method 200 for training and applying a machine-learning model to select one or more network energy saving techniques (NESTs).

[0073] In step 201 , historical data is obtained from one or more network nodes (for example, from one or more of the base stations 120A-C). For example, the one or more network nodes may send the historical data to another entity, and the historical data may be sent upon request from the another entity. The another entity may be a server configured to perform at least some steps of the method 200.

[0074] The historical data sent in step 201 comprises a network energy consumption associated with one or more network energy saving techniques that have been implemented at a respective one or more of the network nodes. The historical data may comprise additional data, as will be discussed in further detail below.

[0075] The one or more network energy saving techniques may comprise any one or more of the network energy saving techniques discussed above (for instance, as described in 3GPP TS 38.300, 3GPP TS 36.423, 3GPP TS 36.413 and / or O-RAN Work Group 1 Network Energy Saving Use Cases Technical Report 3 version 1 .00). For example, the one or more network energy saving techniques may comprises activating or deactivating a base station (for example, base station 120A) in at least one cell (for instance, cell 140A) for a period of time. In another example, the one or more NESTs may additionally or alternatively comprise implementing discontinuous transmission and / or discontinuous reception in the at least one cell of the network, which may be implemented as discussed above. For example, the discontinuous transmission and / or discontinuous reception may be implemented by switching one or more (frequency) carriers on or off at the symbol level, subframe level or radio frame level. In another example, the one or more NESTs may additionally or alternatively comprise enabling or disabling a frequency layer for a time period. In a further example, the one or more NESTs may also or instead comprise enabling or disabling an RF channel for a time period. In yet another example, the one or more NESTs may additionally or alternatively comprise activating or deactivating a secondary cell for a time period. Additionally or alternatively, the one or more NESTs may comprise deactivating the at least one cell based on low cell activity. In some examples, the one or more NESTs may comprise a combination of techniques. For example, cell on / off and frequency carrier on / off may be implemented at the node. It will be appreciated that any other appropriate network energy saving technique may be applied.

[0076] The historical data may be collected at a plurality of nodes and at least some of the plurality of nodes may be different types of node (for example, gNB, eNB, BTS, another node of a current network, a node or base station of a 6G or other future network, and so on). Thus, the historical data can represent various different implementations of the network. The historical data may also represent various deployment scenarios and / or user scenarios. For example, one deployment scenario and / or user scenario may be an extended reality scenario (including augmented reality, virtual reality and mixed reality). Different deployment scenarios and / or user scenarios may result in different network requirements. For instance, in extended reality scenarios, even a slight delay in signals can result in disorientation of a user. Thus, the signal timing may be prioritised more highly in these types of scenarios.

[0077] The historical data may further comprise one or more network configuration parameters for implementing the one or more network energy saving techniques associated with the network energy consumption. The historical data may additionally or alternatively comprise information indicating characteristics of the network. The information indicating network characteristics may comprise an indication of a network load, one or more network traffic characteristics, one or more capabilities of a node in the network, a quality of service (QoS) parameter, a quality of experience (QoE) parameter, temporal information, and / or an indication of the node in which a network energy saving technique was implemented. The temporal information may indicate the time of day or time of year or other temporal information for the historical data. Network traffic characteristics may include bandwidth, delay, jitter and loss. Network traffic characteristics may be related or linked to QoS and / or QoE parameters. The historical data may additionally or alternatively comprise one or more performance indicators or key performance indicators (KPI). For example, a telecommunications system may have one or more KPIs that describe, for example, a minimum acceptable throughput for different cellular services provided by each base station 120A-C.

[0078] In cases where the network is a cellular network, for example as illustrated in Figure 1 , and the node comprises a network cell (for example, cell 140A) or base station (for example, base stations 120A), the historical data may comprise neighbouring cell information (for example, information regarding the cells 1406 and 140C). The neighbouring cell information may comprise information regarding nodes in the neighbouring cells (for example, information regarding base stations 120B and C)

[0079] A QoS parameter (which may also be termed a QoS metric) may be a measurement or value indicating the performance of a service, such as telephony, a computer network, or a cloud computing service, for example. The QoS parameter may be as discussed in 3GPP TS 23.107 or may be one or more of the QoS parameters described is described in 3GPP Technical Specification (TS) 23.203, Section 6.1 .7.2. For example, the QoS parameter may include a QoS Class Identifier (QCI) characteristic, an Allocation and Retention Priority (ARP) characteristic, a guaranteed bitrate (GBR) or a maximum bit rate (MBR). More specifically, the QoS parameter may include a resource type (GBR or non-GBR), a priority level, packet delay budget, packet error loss rate, maximum data burst volume, data rate average window. The ARP characteristic may include information regarding the priority level, the pre-emption capability and the pre-emption vulnerability. A one-to-one mapping of the standardised QoS parameters described in 3GPP TS 23.203 is provided in Tables 6.1 .7-A and B therein.

[0080] The QoS parameter may be calculated, for example, based on packet loss, bit rate, a network throughput, transmission delay, availability, jitter, and so on. The QoS parameter may be measured by a plug-in or application on a computing device (for example, a UE) or by a base station or node in the network. When the QoS parameter includes QCI or ARP characteristics, GBR or MBR, the QoS parameter may be measured at the service level (for example, per Service Data Flow, SDF or per SDF aggregate). The QCI may be a scalar value that is used as a reference to node specific parameters that control packet forwarding treatment (for example, scheduling weights, admission thresholds, queue management thresholds, link layer protocol configuration, and so on) and that have been pre-configured by the operator owning the node (for example, an eNodeB or another network node). This may be implemented in the access network by the QCI referencing node specific parameters that control packet forwarding treatment (for example, scheduling weights, admission thresholds, queue management thresholds, link layer protocol configuration, and so on), that have been pre-configured by the operator at a specific node(s) (for example, the eNodeB or the another network node). When required by operator policy, the eNodeB (or the another node) can be configured to also use the ARP priority level in addition to the QCI characteristic to control the packet forwarding treatment in the eNodeB (or the another node) for SDFs having high priority ARPs.

[0081] Further examples of QoS parameters include a reference signal received power (RSRP), a received signal strength indicator (RSSI), a reference signal received quality (RSRQ), a reference signal time difference (RSTD), a line-of-sight (LOS) indicator or a non- line-of-sight (NLOS) indicator, a rate of delivery over a communication network (network throughput), a signal-to-noise ratio (SNR), block error rate (BLER), error vector magnitude (EVM), channel state information (CSI) parameters. The CSI parameters may comprise a channel quality indicator (CQI), a precoding matrix indicator (PMI), an SS / PBCH resource block indicator (SSBRI), a layer indicator (LI), CSI reference signal resource indicator, and / or a rank indicator (Rl). Other QoS and CSI parameters are also possible.

[0082] A QoE parameter (which may also be termed a QoE metric) may be a measurement or value associated with a user experience with a service. For example, QoE information can be collected for a specific end user service or end user service type from UEs in a specified area. The collected information may be transmitted to another entity (for example, a server), where the information can be analysed and / or KPIs can be calculated.

[0083] The collection may be requested by an operator technician via a management system to the traffic network. As the network may not have any knowledge of which UEs have the capability to record the requested data, the UEs may report whether they have this capability or not when a session is set up. A UE having a capability that matches the request from the management system may be requested to start recording the requested information when constraints of the request are met. The QoE parameter may be measured by a plug-in or application on a the UE. The UE may make the recorded data available to the management system.

[0084] In another example, an OAM-based method could also be used to control the QoE parameter collection. An QAM system may be provided with an indication that a recording session has started and the QAM system may be allowed to modify the QoE measurement configuration. For example, the QoE configured area may be adjusted if the number of sessions is too small or too large. The indication may also be used to determine whether or not to terminate the QoE information collection - for example, if a sufficient number of recording sessions have been started.

[0085] Collecting QoE information from an end user service type from and from a specific user may allow an operator to obtain QoE information for the end user service type for an individual user due to customer complaint or testing of the quality for a new end user service type before launching a service more broadly.

[0086] A detailed explanation of the use and setting of QoE metrics for progressive download is provided in 3GPP TS 26.247. For example, the QoE configuration may only be evaluated by the client at the start of a QoE measurement and reporting session (“QoE session”) associated with a streaming session. This may include evaluation of any filtering criteria such as by geographical area. Client evaluation of all measurement and reporting criteria for an ongoing QoE session may be unaffected by any QoE configuration changes received during that session. That is, any changes to the QoE configuration may only affect QoE sessions started after these configuration changes have been received.

[0087] The QoE metric may include an average throughout. An example definition of the average throughput is provided in Table 1 .

[0088] Table 1 - Average Throughput Definition

[0089] The QoE metric may also or instead include an initial playout delay. The initial playout delay may only be logged at the point in time when playout of streaming media (for example, a video) begins. An example definition of the initial playout delay is provided in Table 2.

[0090] Table 2 - Initial Playout Delay Definition

[0091] The QoE metric may also or instead include a play list or playout. Decoded samples are generally rendered in presentation time sequence, at or close to its specified presentation time. A compact representation of the information flow can thus be constructed from a list of time periods during which samples of a single representation were continuously rendered, such that each was presented at its specified presentation time to some specific level of accuracy (for example, + / -10 ms). Such a sequence of periods of continuous delivery may be started by a user action that requests playout to begin at a specified media time (for instance, a "play", "seek" or "resume" action) and may continues until playout stops, which may be due to a user action, the end of the content, or a permanent failure. Table 30 below provides an example definition of the play list metric.

[0092] Table 3 - Play List Description

[0093] A representation is one of a complete set or subset of media content components comprising the media content during a defined period. The trace may include entries for different representations that overlap in time, which may be due to multiple representations are being rendered simultaneously - for example one audio and one video representation.

[0094] The play list may include user actions regarding start / stop, but may also include other non-user actions such as adaptation and rebuffering. Thus, the play list may be used to derive many other metrics, and an example calculation of a few stalling-related metrics is described with reference to Figure 5. Figure 5 illustrates an example of playout characteristics for a 60-second video. The video is stalled during a first period 501 for five seconds. During a second period 502 of ten seconds, the video plays with a first representation. Subsequently, there is a further stalled period 503 that lasts for 15 seconds. Following the stalled period 503, the video plays for a period 504 of 20 seconds with a second representation. Finally, the video plays for a period 505 of 30 seconds with a first representation. Thus, although the video duration is 60 seconds (corresponding to the sum of periods 502, 504 and 506), it can be seen from Figure 5 that the time that a user spends during playout is longer.

[0095] Assuming that a user selects to start the 60-second video at wall-clock time 09:00:00 (hh:mm:ss), this may result in the following (simplified) playout list being reported by the client:

[0096] Playlist

[0097] Entry#1 start = 09:00:00 mstart = 00:00:00 starttype = New playout request

[0098] Trace

[0099] Traceentry#1 representationid = 1 start = 09:00:05 sstart = 00:00:00 duration = 10 seconds stopreason = rebuffering

[0100] Traceentry#2 representationid = 2 start = 09:00:30 sstart = 00:00:10 duration = 20 seconds stopreason = representation switch

[0101] Traceentry#3 representationid = 1 start = 09:00:50 sstart = 00:00:30 duration = 30 seconds stopreason = end of content

[0102] The number of stalling occurrences may be calculated by counting how many times a stop reason is specified as "rebuffering". The time duration for a stalling event may be calculated based on the time difference between the end time of a trace entry with stopreason equal to "rebuffering", and the start time of the next trace entry. In the example above, the stalling starts at "Traceentry#1 , (start + duration)" = 09:00:05 + 10 secs = 09:00:15, and ends at "Traceentry#2, start" = 09:00:30. Thus the length of the stalling is 15 seconds.

[0103] Although Figure 5 is described with reference to playout, it will be understood that similar considerations may apply to other scenarios including, for example, playback.

[0104] The QoE parameter may also or instead include a playout delay for media start-up. This metric may indicate a waiting time that a user experiences for media start-up.

[0105] The QoE parameter may also or instead include device information. This metric may include information about the displayed video resolution, as well as the physical screen characteristics. If the video is rendered in full-screen mode, the video resolution usually coincides with the characteristics of the full physical display. If the video is rendered in a smaller subwindow, the characteristics of the actual video window shown may be logged. The physical screen width and the horizontal field-of-view may also be logged.

[0106] The device information may be logged at the start of each QoE reporting period, and may also be logged whenever the characteristics changes during the session (for instance, if the UE is rotated from horizontal to vertical orientation, or if the video subwindow size is changed).

[0107] Other QoE metrics discussed in 3GPP TS 26.247 may also or instead be used.

[0108] Other examples of QoE parameters include a user satisfaction indicator, an initial buffer delay or stall, an indication of (or information regarding) a stall event, a playback time, a rebuffering ratio, an indication of a start failure event (for instance, a media start failure), an average media bit rate, a media start time, and / or jitter. QoE metrics such as, for example, a playback time and a start time may indicate that a user has been waiting a significant time. Other QoE parameters may also or instead be used.

[0109] Jitter is a measurement of deviation from true periodicity of a signal assumed to be periodic in nature, often in reference to a clock signal. Jitter can be quantified in the same terms as any time-varying signals, for example, in terms of root-mean-square (RMS) displacement, peak-to-peak displacement, spectral density, and so on. The jitter can be measured by number of different metrics, such as absolute jitter, (maximum) time interval error, period jitter, and cycle-to-cycle jitter, for example.

[0110] The information regarding the stall event may include an indication that a stall even has occurred, a stalling ratio, a frequency of stall events, a duration of the stall event(s), a time duration since a previous stall and / or a total number of stalls. A stalling ratio is a percentage / fraction of the time that a user experiences stalling issues. That is, the stall ratio may be defined as - total use ti:me +total stall ti:—me . Stall events may occur when a buffer is emptied as a result of a download throughput that is lower than a bit rate. The stall event may be an initial buffer stall, for example. The rebuffering ratio is a percentage / fraction of time that a user experiences buffering total rebuffering time issues. That is, the rebuffering ratio may be determined as total use time +total rebuffering time'

[0111] An indication of a failure event may comprise an indication (for example, a notification) that an event has not started within a pre-determined time period of an initiation attempt. For example, an indication of a failure event may be transmitted when an initial packet is not successfully delivered to a user device within the pre-determined time period. The initiation attempt may be a user-initiated event. For example, the initiation attempt may be a user clicking on a graphical control element (for example, a button) to initiate playback of media, for instance. The media may include a video file or audio file. If the media does not start within the pre-determined time period after the click, the user device may send the indication of the failure event.

[0112] Referring again to Figure 2, at step 202, one or more machine learning algorithms are used to construct a machine learning model based on at least a subset of the historical data. The one or more ML algorithms may be run at one of the base stations 120A-C, a computing device (for example, by a UE in one of the cells 140A-C), a third party server, or a combination of these.

[0113] Any appropriate machine learning algorithm may be used. For example, a supervised learning algorithm may search through a hypothesis space to find a suitable hypothesis that can make accurate predictions for a particular problem. Supervised learning algorithms include support-vector machines, linear or logistic regression, naive Bayes, decision trees, k-nearest neighbour algorithm, neural networks and similarity networks. A multiple classifier or ensemble method (which use multiple learning algorithms or hypotheses) may improve the accuracy of the results. For example, a random forest (random decision forest) is an ensemble learning method that operates by constructing a multitude of decision trees at training time. Unsupervised learning algorithms, such as clustering methods, anomaly detection methods and / or approaches for learning latent variable models, for example, may also or instead be used. Clustering methods may include hierarchical clustering, k-means, mixture models, DBSCAN and OPTICS algorithm. Anomaly detection methods may include local outlier factor and isolation forest methods. Approaches for learning latent variable models may include an expectation-maximisation algorithm, method of moments and blind signal separation techniques. Weak or semisupervised learning may also or instead be used.

[0114] At step 203, the machine learning model is trained based on at least a subset of the historical data. Training the model may comprise computing an estimated network energy saving of each of the one or more NESTs based on the network energy consumptions in the historical data. During or after training, the model may be tuned, as will be discussed with reference to Figure 4.

[0115] At step 204, the trained model is applied to input data to select one or more network energy saving techniques that are predicted to result in a network energy consumption less than a threshold amount. Following or as part of step 204, the model may output a prediction of the network energy saving, as will be discussed in further detail with reference to Figure 4.

[0116] The input data may be real-time data obtained from at least one node. Thus, the trained model can select one or more NESTs based on how the node is currently being used. Since, as discussed above, various factors including the network deployment scenario, time (for example, time of day) and user implementation affect the network energy saving possible, using real-time data may enable a more accurate prediction of which NEST(s) would result in a network energy consumption less than a threshold amount. Thus, using real-time data may improve the network energy consumption reduction possible.

[0117] In another example, the input data may be predicted data. The predicted data may be predicted by the trained model or another entity based on the historical data. For example, it may be determined (by the model or another entity) that a user typically enters a particular cell region at a particular time. This may correspond to the user arriving at their workplace or home, for example. The trained model may thus generate or receive this predicted data to use as input data to select one or more NESTs for the predicted future scenario. This may enable more efficient implementation of the one or more NESTs, as the one or more NESTs can be determined and / or configured ahead of a time in which they are to be implemented.

[0118] The input data may comprise a number of different types of data. For example, the input data may correspond to one or more network configuration parameters for a particular NEST. The model may thus be used to predict a set of network configuration parameters for the particular NEST that result in the network energy consumption less than a (non-zero) threshold consumption amount. In other words, selecting the one or more NESTs may comprise determining the set of network configuration parameters for implementing the particular NEST. In cases where a combination of NESTs for network energy saving is selected by the model, the set of network configuration parameters comprises network configuration parameters for the combination of NESTs. In another example, the input data may comprise further historical data, which may be used to assess or tune the model, for example. In a further example, the input data may comprise information relating to the network. For example, the input data may specify at least one node for which a network energy saving technique is to be selected. In another example, where the network is a cellular network, the input data may comprise information relating to a cell in the network and, optionally, information relating to neighbouring cells.

[0119] Applying the model to select the one or more NESTs may comprise comparing a predicted network energy consumption of each of (or at least some of) the one or more historically implemented NESTs. The method 200 may thus comprise predicting the network energy consumption of each of (or at least some of) the one or more historically implemented NESTs based on the input data. This may include predicting a network energy consumption of a combination of NESTs.

[0120] In another example, applying the model may additionally or alternatively comprise computing an estimated energy saving of each of (or at least some of) the one or more NESTs based on the respective network energy consumption in the historical data.

[0121] In cases where the network is a cellular network, it is useful to consider the power saving (network energy saving) of a group of cells, rather than only the at least one cell in which the selected one or more NESTs are to be implemented. The method 200 may thus further comprise predicting the network energy consumption by computing a predicted network energy consumption of a group of cells by calculating a predicted power usage of the at least one cell and a predicted power usage of one or more adjacent (neighbouring) cells. In examples where applying the model comprises computing an estimated network energy saving, computing the estimated energy saving may comprise computing an estimated energy saving of a group of cells by calculating a power usage of a cell in which the one or more network energy saving techniques were implemented and a power usage of one or more neighbouring cells.

[0122] Comparing the predicted network energy consumption of the one or more implemented NESTs to select the one or more NESTs may comprise selecting the one or more NESTs predicted to result in the lowest network energy consumption and / or the greatest network energy saving.

[0123] The threshold amount may correspond to the network energy consumption prior to implementing the selected one or more network energy saving techniques. Thus, the trained model may select one or more NESTs that results in a network energy saving compared to the current operation of the network. In some examples, the threshold amount may be set such that selecting the one or more NESTs results in the lowest network energy consumption and / or the greatest network energy saving. For example, the threshold amount may be set by the model based on the historical data and / or dynamically updated based on additional (for example, input) data.

[0124] At step 205, the one or more selected network energy saving techniques are implemented in at least one node of the network. In other words, the at least one node is operated according to the one or more selected NESTs. The method 200 may thus be a method of reducing energy consumption in a network.

[0125] Optionally, at step 206, further input data may be received from the at least one node following the implementation of the one or more NESTs. The method 200 may then return to step 204 to apply the model to the further input data to select a further one or more network energy saving techniques. Thus, the method 200 may allow fine-tuning or adjustment of the selected one or more NESTs. The fine-tuning or adjustment may comprise selecting a new one or more NESTs or reconfiguring parameters of the previously selected one or more NESTs. As discussed above, various factors affecting a network energy consumption may change over time, or may otherwise vary dynamically. Thus, receiving further input data and applying the model to the further input data allows the network energy consumption to remain below the threshold consumption amount, even when circumstances change. In other words, the selected one or more NESTs can be updated dynamically. This may allow the optimal NEST to be identified and implemented as the circumstances change.

[0126] In some examples, the selected further one or more NESTs may not differ from (or may be sufficiently similar to) the selected one or more NESTs. The selected further one or more NESTs may be determined to be sufficiently similar to the selected one or more NESTs in cases defined to be sufficiently similar (for example, by a user, Al or ML algorithm) or otherwise are determined to be sufficiently similar. Where the selected further one or more NESTs do not differ from / are sufficiently similar to the selected one or more NESTs, the method 200 may terminate at step 204 and the initial selected one or more NESTs may be maintained. That is, the at least one node may continue to be operated according to the initial selected one or more NESTs. Alternatively, the initial selected one or more NESTs may be maintained, but the method 200 may iteratively perform at least steps 204 and 206. This may continue until the selected further one or more NESTs differ from (or are not sufficiently similar to) the selected one or more NESTs.

[0127] When the selected further one or more network energy saving techniques differ from (or are not sufficiently similar to) the selected one or more network energy saving techniques, step 205 may be repeated with the selected further one or more NESTs. That is, the further one or more network energy saving techniques may be implemented in the at least one node. This may occur after one or more additional iterations of steps 204 and 206.

[0128] Similarly to discussed above with respect to the input data, applying the model to the further input data in step 204 may comprise determining a further set of network configuration parameters for the selected further one or more network energy saving techniques based on the further input data. In cases where the further set of network configuration parameters does not differ from (or is sufficiently similar to) the initial set of network configuration parameters, the method 200 may terminate at step 204 and the initial set of network configuration parameters may be maintained. Alternatively, the initial selected one or more NESTs may be maintained, but the method 200 may iteratively perform at least steps 204 and 206, as discussed above. This may continue until the further set of network configuration parameters differ from (or are not sufficiently similar to) the initial set of network configuration parameters.

[0129] In examples, the further set of network configuration parameters may differ from (or not be sufficiently similar to) the set of network configuration parameters. This may be determined to be the case if the further set of network configuration parameters differ from the set of network configuration parameters by more than a threshold amount, are defined (by a user, Al or ML algorithm) to differ, or otherwise are determined to be not sufficiently similar. This may be because the further set of network configuration parameters result in an increased network energy saving compared to the initial set of network configuration parameters, or this may be because the initial set of network configuration parameters no longer result in the required network energy saving. This may be as a result of the network conditions having changed. Accordingly, the method 200 may further comprise adjusting the operation of the network according to the further set of network configuration parameters. In other words, the set of network configuration parameters can be updated dynamically. This may allow the optimal set of network configuration parameters to be identified and implemented as the circumstances change.

[0130] Steps 201 to 203 may be performed by a different entity to that performing steps 204 to 206. For example, the steps of generating and training the model may be performed by a first computing device, whilst the steps of applying the model, implementing the selected one or more NESTs and receiving further input data may be performed by a second computing device. The first computing device may be a server, for example. The second computing device may be the transceiver or base station in a cellular network. In other examples, the second computing device may be a server.

[0131] Once the model has been created and trained, it can be applied to input data without a need to create a new model or (re-)train the model. Thus, steps 201 to 203 may not be performed as part of the method 200. For example, the trained machine learning model may be received by the second computing device such that the second computing device can perform steps 204 to 206 without the need to perform steps 201 to 203.

[0132] As discussed above, network energy saving is an important factor in networks, particularly for large networks. However, network energy saving techniques can come at the risk of reduced or limited service (for example, reduced or limited coverage) and / or reduced or limited QoE and / or QoS. Thus, whilst it is desirable to reduce, optimise (for example, minimise) or at least limit network energy consumption, other factors may be taken into account. For example, in extended reality scenarios, even a slight delay in signals can result in disorientation of a user. Similarly, a low average media bit rate may cause a user to stop using a particular service. Thus, if a user is unable to use, or is not satisfied with using, a device due to the implemented NEST, this is undesirable.

[0133] The inventor has recognised maintaining various QoS and / or QoE parameters whilst also aiming to reduce network energy consumption is desirable. This balance can be difficult to achieve when also aiming to reduce network energy consumption using existing methods. Existing methods typically implement cell on / off for network energy saving, without investigating the network energy saving achieved in practice and without determining a QoS and / or QoE received by a user. Thus, in worst case scenarios, the network energy consumption is increased and the quality and / or coverage of the service is also poor.

[0134] In view of these issues, at step 204, applying the model may further comprise selecting one or more network energy saving techniques that are predicted to result in a QoS and / or QoE parameter above a threshold level. The input data may accordingly comprise a QoE and / or QoS matrix for a computing device in the network (for example, a UE in a cellular network). The matrix may define the threshold level of a respective QoE and / or QoS parameter. In some examples, selecting the one or more NESTs at step 204 may comprise selecting one or more NESTs that optimise the network energy consumption and the QoS and / or QoE parameter.

[0135] In some examples, selecting the one or more NESTs may comprise selecting the one or more NESTs that are predicted to result in more than one QoS and / or QoE parameter above a respective (non-zero) threshold level. For example, the selected NEST(s) may be predicted to result in the greatest number of QoS and / or QoE parameters above the respective threshold levels. In examples where more than one NEST is predicted to result in more than one QoS and / or QoE parameter above the respective threshold level, each of the NESTs may be implemented. In other examples, one or more of the NESTs may be implemented based on, for example, the network deployment and / or user scenario. For example, whilst a NEST may be predicted to result in a low packet loss, the user may be using voice over IP or another type of voice call, which may not require particularly low packet loss.

[0136] The QoS and / or QoE parameter may relate to UEs in the at least one cell in which the selected one or more NESTs are to be implemented (that is, when the network is a cellular network). In some examples, the QoS and / or QoE parameter may also relate to UEs in one or more neighbouring cells. For example, whilst a NEST may be predicted to result in a QoE and / or QoS parameter for users in the at least one cell above a threshold level, the QoE and / or QoS for users in a neighbouring cell may not be above the threshold level. Thus, the NEST may not be selected for implementation in the at least one cell. In other examples, the NEST may be implemented although the QoE and / or QoS parameter is not above a threshold level for users in the at least one cell and / or the neighbouring cells above a threshold level. For example, where no NESTs are predicted to result in a QoE and / or QoS parameter the threshold level, one or more NESTs that maximise the QoE and / or QoS parameter may be selected.

[0137] Figure 3 illustrates one exemplary embodiment in which particular network energy saving techniques are implemented at a node. Three nodes 320A-C are shown in Figure 3 for illustrative purposes only. More or fewer nodes may be present. Furthermore, the implementation of network energy saving techniques may differ from that shown in Figure 3.

[0138] A first network energy saving technique is implemented at node 320A, resulting in a first network energy consumption. This data 360 may be stored at the node 320A and / or may be transmitted to a computing device. The computing device may be a server, for example, a database server. The data 360 may be used as historical data (for example, for generating or training an ML model) or as input data (for example, as real-time data to be input into the ML model to determine a further (different) NEST).

[0139] Similarly, a second network energy saving technique is implemented at node 320B, resulting in a second network energy consumption. At node 320C, the first network energy saving technique is implemented, resulting in a third network energy consumption. Again, data 360 may be stored at the respective nodes 320B, 320C and / or may be transmitted to the computing device.

[0140] Since the network energy consumption may depend on a number of factors, although the first network energy saving technique has been implemented in nodes 320A and 320C, the resulting first and third network energy consumptions may not be the same. In other examples, the first and third network energy consumptions may be the same (at least to within a threshold tolerance). For example, the network conditions for the node 320A and 320C may be the same, or sufficiently similar, such that the first and third network energy consumptions are comparable.

[0141] As discussed above in respect of Figure 2, a trained model is applied to input data to select one or more network energy saving techniques that are predicted to result in a network energy consumption less than a (non-zero) threshold consumption amount. The selected one or more network energy saving techniques are then implemented in at least one node of the network. This may be achieved by operating the at least one node according to a set of network configuration parameters determined by the model.

[0142] For example, the historical data may further comprise one or more network configuration parameters for implementing the respective NEST at the node. The one or more machine learning algorithms may thus identify patterns or trends based on the network configuration parameters and the network energy saving. Similarly, the model may be trained on the historical data to predict network configuration parameters. Applying the model to the input data may thus comprise determining a set of network configuration parameters for implementing the selected one or more NESTs, based on the input data. For example, the model may determine that a particular set of network configuration parameters are predicted to result in a network energy consumption less than a threshold consumption amount for the selected one or more NESTs. Other sets of network configuration parameters may be determined by the model not to result in the network energy consumption less than the threshold consumption amount for the selected one or more NESTs and so may not be used to operate the at least one network node. In other examples, the model may determine that the particular set of network configuration parameters are predicted to optimise (for example, minimise) the network energy consumption.

[0143] Accordingly, implementing the selected one or more NESTs in the at least one node may comprise operating the at least one node according to the set of network configuration parameters.

[0144] The historical data and / or input data may comprise information indicating network characteristics or conditions. In cases where the input data and historical data comprise this information, selecting the one or more network energy saving techniques may comprise selecting the one or more network energy saving techniques for which the information indicating network characteristics of the historical data is the same as, sufficiently similar to or most similar to the information indicating network characteristics of the input data.

[0145] The information being sufficiently similar (or the network conditions being sufficiently similar) may mean that the information is the same to within a threshold tolerance, defined (for example, by a user, an Al or a machine learning algorithm) as being corresponding or similar, or otherwise corresponding.

[0146] For example, the data 360 may comprise a temporal indication that is within a predetermined time period of, or overlaps with, a temporal indication of the input data. For instance, node 320A may have produced the second network energy consumption between 5 and 6PM. The input data may comprise a time window of between 5:30 and 6:30PM. Thus, as the time windows of the historical and input data overlap, the information or network conditions may be considered sufficiently similar. In another example, the historical data may comprise information that the node 320C was activated at 11 :30AM and the input data may comprise a time of 11 :57AM. The temporal information may thus be considered sufficiently similar where the input data time is within the pre-determined time period of the historical data time (for example, the pre-determined time period may be thirty minutes).

[0147] In other examples, the information (or network conditions) being sufficiently similar may mean that the operation of the nodes or the network is similar. For example, a gNB node may operate similarly to a next generation eNB (ng-eNB) node in that the gNB and the ng-eNB may allow UEs to connect to a 5G core network via a radio interface. In particular, the gNB may allow UEs configured to operate according to 5G provisions to connect to a 5G core network using a 5G radio interface, while the ng-eNB may allow UEs configured to operate according to 4G LTE provisions to connect to the 5G core network via a 4G radio interface. Different types of small cells may operate similarly to each other (for example, a picocell may operate similarly to a microcell). Networks involving high frequency and low frequency bands may operate similarly to each other.

[0148] In some examples, the information (or network conditions) may be determined as being sufficiently similar where the information is most similar to the information indicating the network characteristics of the input data. It will be appreciated that other criteria may be used to identify the network conditions and / or information as being sufficiently similar.

[0149] Figure 4 illustrates a flowchart method illustrating one exemplary implementation of the method 200. Steps 401 , 402, 403 and 404 may correspond to steps 201 -204 discussed above with reference to Figure 2. At step 401 , real-time data from one or more nodes may be collected in addition to, or instead of, historical data. This may also be the case in step 201 . Using real-time data for the one or more ML algorithms may enable a model that can more accurately predict and / or select one or more NESTs. For example, historical data may not accurately represent current usage of the network and / or NESTs. For instance, historical data may suggest that a particular NEST would not result in a network energy consumption below a threshold value, but the network infrastructure may have been updated or changed since the historical data was obtained. Real-time data may thus show the implementation of NESTs with current network infrastructure.

[0150] Step 403 (corresponding to step 203 in Figure 2) may comprise a step of adjusting parameters of the model for a specific NEST or specific NESTs. This is shown at step 411 in Figure 4. That is, the model may be tuned by, for example, adjusting internal coefficients or weights for the model that were found by the one or more ML algorithms, based on the specific NEST(s). In another example, output parameters of the model may be tuned. Realtime input data may be used to tune the model (as well as to compare the outcome of the NES techniques for the given scenario). Other model tuning is also possible, for example by adjusting hyperparameters such as, for instance, a maximum depth for a decision tree or a number of trees included in a random forest.

[0151] Where the historical data comprises one or more network configuration parameters for implementing a respective one of the one or more NESTs, the model parameters may be adjusted in step 411 to allow prediction of a set of network configuration parameters for the specific NEST(s). The specific NEST(s) may include a combination of NESTs. In this case, one or more network configuration parameters for each of the NESTs in the combination and / or network configuration parameters for the combination of NESTs may be used (for example, input into the model) to adjust the model parameters.

[0152] Once the model has been trained in step 403 and, optionally, fine-tuning of the model has been performed at step 411 , the predictive model is output at step 412. At step 413, input data or new data is received from one or more nodes in the network. The input data may be received directly from the one or more nodes or via another computing device (for example, a database server). The input data may be as discussed in respect of step 204. The input data may correspond to a specific network energy saving technique (for example, a network energy technique currently implemented at the one or more nodes). The input data may comprise neighbouring cell information when the network is a cellular network and the one or more nodes are in a cell of, or are the cell of, the cellular network.

[0153] At step 404, the model is applied to the input data received at step 413. Optionally, a QoE and / or QoS matrix may be input into the model as well at step 414, as discussed above with reference to Figure 2.

[0154] At step 415, the model may output a prediction of the network energy saving. The predicted network energy saving may be relative to the threshold consumption amount or the current network energy consumption (that is, prior to implementing the selected one or more NESTs). The predicted network energy saving may represent an overall network energy saving, or may represent a saving in part of the network. The predicted network energy saving may be output as a percentage saving. The model may additionally or alternatively output a prediction of the network energy consumption.

[0155] At step 415, the model may additionally or alternatively output the selected one or more NESTs. In some examples, the model may only output the selected NEST(s) determined to be optimal. Determining that the selected NEST(s) are optimal may comprise determining that the selected NEST(s) resulting in the greatest energy saving, the lowest network energy consumption, the highest QoS and / or QoE parameters and / or the greatest number of QoS and / or QoE parameters above the threshold level.

[0156] In other examples, step 415 may correspond to step 205. In other words, the model may not output the network energy saving and / or the selected NEST(s), but may instead proceed to implement the selected NEST(s).

[0157] The method may be implemented by a network entity or station of a cellular network, the network entity or station being configured to operate in accordance with the method. Examples of a network entity in this context may include a cell, an entity configured for operating or controlling the cell, a provisioning platform, a global / provisioning API platform. Other examples may include a network node of a cellular network (for example, an eNB, a gNB or a base station of a 6G network, a future network or another network). For example, the network entity may include a processor and at least one communication interface, particularly comprising one or both of a transmitter and receiver. A user terminal may likewise include a processor and at least one communication interface, particularly comprising one or both of a transmitter and receiver. A controller for a network entity may also be considered.

[0158] Any of the methods described herein may be implemented as a computer program. The computer program may be configured to control a network node or entity to perform any method according to the disclosure.

[0159] Although the methods described herein are described primarily based on LTE base stations (eNBs) and 5G NR base stations (gNBs), it will be understood that the methods described herein may be used with any current or future (for example, 6G and beyond) cellular network, in which the base stations may be referred to by other terminology.

[0160] Furthermore, although the methods described herein have been described mainly with reference to a cellular networks, it will be appreciated that the methods may be implemented in other types of network. For example, the network may be a computer network. Network energy saving techniques that may be used by a computer network include switching off a switch port in the computer network for a minimum period of time, routing network traffic via a different traffic path, disabling a network link for a period of time, and so on. QoE and / or QoS parameters for computer networks may be measured in manners similar to those discussed above.

[0161] The methods described herein may be implemented with computer system configurations including hand-held devices, microprocessor systems, microprocessor- based or programmable consumer electronics, minicomputers, mainframe computers and the like. The embodiments can also be practiced in distributed computing environments, where tasks are performed by remote processing devices that are linked through a network.

[0162] The computer system may include a processor, such as a central processing unit (CPU). The processor may execute logic in the form of a software program. The computer system may include a memory including volatile and non-volatile storage medium. The different parts of the system may be connected using a network (e.g. wireless networks and wired networks). The computer system may include one or more interfaces. The computer may contain a suitable operating system such as UNIX (including Linux) or Windows (RTM), for example.

[0163] Certain embodiments can also be embodied as computer-readable code on a non- transitory computer-readable medium. The computer readable medium may be any data storage device than can store data, which can thereafter be read by a computer system. Examples of the computer readable medium include hard drives, network attached storage (NAS), read-only memory, random-access memory, CD-ROMs, CD-Rs, CD-RWs, magnetic tapes, and other optical and non-optical data storage devices. The computer readable medium can also be distributed over a network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion. Although embodiments according to the disclosure have been described with reference to particular types of devices and applications (particularly UEs) and the embodiments have particular advantages in such case, as discussed herein, approaches according to the disclosure may be applied to other types of device and / or application (for example, desktop computers). Each feature disclosed in this specification, unless stated otherwise, may be replaced by alternative features serving the same, equivalent or similar purpose. Thus, unless stated otherwise, each feature disclosed is one example only of a generic series of equivalent or similar features.

[0164] All of the aspects and / or features disclosed in this specification may be combined in any combination, except combinations where at least some of such features and / or steps are mutually exclusive. In particular, the preferred features of the disclosure are applicable to all aspects and embodiments of the disclosure and may be used in any combination. Likewise, features described in non-essential combinations may be used separately (not in combination).

[0165] As used herein, including in the claims, unless the context indicates otherwise, singular forms of the terms herein are to be construed as including the plural form and vice versa. For instance, unless the context indicates otherwise, a singular reference herein including in the claims, such as “a” or “an” (such as a technique) means “one or more” (for instance, one or more techniques).

[0166] Throughout the description and claims of this disclosure, the words “comprise”, “including”, “having” and “contain” and variations of the words, for example “comprising” and “comprises” or similar, mean “including but not limited to”, and are not intended to (and do not) exclude other components. Also, the use of “or” is inclusive, such that the phrase “A or B” is true when “A” is true, “B is true”, or both “A” and “B” are true.

[0167] The use of any and all examples, or exemplary language (“for instance”, “such as”, “for example” and like language) provided herein, is intended merely to better illustrate the disclosure and does not indicate a limitation on the scope of the disclosure unless otherwise claimed. No language in the specification should be construed as indicating any nonclaimed element as essential to the practice of the disclosure.

[0168] The terms “first” and “second” may be reversed without changing the scope of the invention. That is, an element termed a “first” element (e.g., a first NEST) may instead be termed a “second” element (e.g., a second NEST) and an element termed a “second” element (e.g., a second NEST) may instead be considered a “first” element (e.g. a first NEST). Any steps described in this specification may be performed in any order or simultaneously unless stated or the context requires otherwise. Moreover, where a step is described as being performed after a step, this does not preclude intervening steps being performed.

[0169] It is also to be understood that, for any given component or embodiment described herein, any of the possible candidates or alternatives listed for that component may generally be used individually or in combination with one another, unless implicitly or explicitly understood or stated otherwise. It will be understood that any list of such candidates or alternatives is merely illustrative, not limiting, unless implicitly or explicitly understood or stated otherwise.

[0170] In this detailed description of the various embodiments, for the purposes of explanation, numerous specific details are set forth to provide a thorough understanding of the embodiments disclosed. One skilled in the art will appreciate, however, that these various embodiments may be practiced with or without these specific details. Furthermore, one skilled in the art can readily appreciate that the specific sequences in which methods are presented and performed are illustrative and it is contemplated that the sequences can be varied and still remain within the scope of the various embodiments disclosed herein.

[0171] All literature and similar materials cited in this application, including but not limited to patents, patent applications, articles, books, treaties and internet web pages are expressly incorporated by reference in their entirety for any purpose. Unless otherwise described, all technical and scientific terms used herein have a meaning as is commonly understood by one of ordinary skill in the art to which the various embodiments described herein belongs.

Claims

Claims:1 . A method for reducing energy consumption in a network, the method comprising: applying a trained machine learning model to input data to select one or more network energy saving techniques that are predicted to result in a network energy consumption less than a threshold consumption amount, the trained machine learning model having been trained on historical data received from one or more nodes of the network, the historical data comprising a respective network energy consumption associated with each of one or more implemented network energy saving techniques; and implementing the selected one or more network energy saving techniques in at least one node of the network.

2. The method of any previous claim, wherein the network is a cellular network.

3. The method of claim 2, wherein implementing the selected one or more network energy saving techniques in the at least one node of the network comprises implementing the selected one or more network energy saving techniques in at least one cell in the cellular network.

4. The method of claim any previous claim, wherein: the historical data further comprises one or more network configuration parameters for implementing a respective one of the one or more implemented network energy saving techniques; applying the model comprises determining a set of network configuration parameters for implementing the selected one or more network energy saving techniques based on the input data; and implementing the selected one or more network energy saving techniques in the at least one node comprises operating the at least one node according to the set of network configuration parameters.

5. The method of any previous claim, wherein the input data comprises and / or the historical data further comprises information indicating network characteristics.

6. The method of claim 5 when the input data and the historical data comprise information indicating network characteristics, wherein selecting the one or more network energy saving techniques comprises selecting the one or more networkenergy saving techniques for which the information indicating network characteristics of the historical data is the same as, sufficiently similar to or most similar to the information indicating network characteristics of the input data.

7. The method of claim 5 or claim 6, wherein the information indicating network characteristics comprises one or more of: a network load; one or more network traffic characteristics; one or more capabilities of a node in the network; a quality of service, QoS, parameter; a quality of experience, QoE, parameter; a time of day; and, when the network is a cellular network and the node comprises a network cell, neighbouring cell information.

8. The method of claim 7 when the information indicating network characteristics comprises a QoS parameter and / or a QoE, parameter, wherein applying the model further comprises selecting one or more network energy saving techniques that are predicted to result in a QoS and / or QoE parameter above a threshold level.

9. The method of claim 8, wherein applying the model further comprises selecting one or more network energy saving techniques that optimise the network energy consumption and the QoS and / or QoE parameter.

10. The method of any one of claims 7 to 9, wherein the QoS parameter comprises one or more of: a QoS Class Identifier (QCI) characteristic; an Allocation and Retention Priority (ARP) characteristic; a guaranteed bitrate (GBR); a maximum bit rate (MBR); a reference signal received power, RSRP; a received signal strength indicator, RSSI; a reference signal received quality, RSRQ; a reference signal time difference, RSTD; a line-of-sight, LOS, indicator or a non-line-of-sight, NLOS, indicator; a network throughput; a signal-to-noise ratio, SNR; block error rate, BLER; error vector magnitude, EVM; and channel state information, CSI, parameters; and / or the QoE parameter comprises one or more of:a user satisfaction indicator; an average throughput; an initial playout delay; a buffer level; a play list; a playout delay for media start-up; device information; information regarding a stall event; a playback time; a rebuffering ratio; an indication of a failure; an average media bit rate; a start time; and jitter.11 . The method of claim 10 when the QoE parameter comprises the information regarding the stall event, wherein the information regarding the stall event comprises an indication that a stall event has occurred, a stalling ratio, a frequency of stall events, a duration of the stall event and / or a duration of one or more previous stall events, a time duration since a previous stall and / or a total number of stalls.

12. The method of claim 10 when the QoS parameter comprises CSI parameters, wherein the CSI parameters comprise one of more of: a channel quality indicator, CQI; a precoding matrix indicator, PMI;SS / PBCH resource block indicator, SSBRI; a layer indicator, LI;CSI reference signal resource indicator; and / or a rank indicator, RL13. The method of any previous claim, wherein the input data comprises real-time data received from one or more nodes of the network and / or predicted data for the one or more nodes of the network.

14. The method of any previous claim, wherein the selected one or more network energy saving techniques comprises a combination of two or more network energy saving techniques.

15. The method of any previous claim, wherein applying the model comprises: comparing a predicted network energy consumption of each of the one or more implemented network energy saving techniques and / or a combination of the one or more implemented network energy saving techniques to select the one or more network energy saving techniques, wherein the predicted network energy consumption is predicted based on the input data; and / or computing an estimated energy saving of each of the one or more network energy saving techniques and / or a combination of the one or more implemented network energy saving techniques based on the respective associated network energy consumption.

16. The method of claim 15 when dependent on claim 2 or 3, wherein computing the estimated network energy saving comprises computing an estimated energy saving of a group of cells by calculating a power usage of a cell in which the one or more network energy saving techniques were implemented and a power usage of one or more neighbouring cells; and / or predicting the network energy consumption by computing a predicted network energy consumption of a group of cells by calculating a predicted power usage of the at least one cell and a predicted power usage of one or more adjacent cells.

17. The method of claim 15 or claim 16, wherein comparing the predicted network energy consumption of each of the one or more implemented network energy saving techniques to select the one or more network energy saving techniques comprises selecting the one or more network energy saving techniques predicted to result in the lowest network energy consumption and / or the greatest network energy saving.

18. The method of any previous claim, further comprising, following the step of implementing the selected one or more network energy techniques, receiving further input data from the at least one node and applying the model to the further input data to select a further one or more network energy saving techniques.

19. The method of claim 18, further comprising, when the selected further one or more network energy saving techniques differ from the selected one or more network energy saving techniques, implementing the selected further one or more network energy saving techniques in the at least one node.

20. The method of claim 19 when dependent on claim 4, wherein applying the model to the further input data comprises determining a further set of network configuration parameters for the selected further one or more network energy saving techniques based on the further input data and further comprising, when the further set of network configuration parameters differs from the set of network configuration parameters by more than a threshold amount, adjusting the operation of the network according to the further set of network configuration parameters.21 . The method of any previous claim, wherein the one or more implemented network energy saving techniques and / or the selected one or more network energy saving techniques comprises one or more of: activating or deactivating a base station in the at least one cell for a period of time; implementing discontinuous transmission and / or discontinuous reception in the at least one cell of a cellular network; enabling or disabling a frequency layer for a time period; enabling or disabling an RF channel for a time period; activating or deactivating a secondary cell for a time period; and deactivating the at least one cell based on low cell activity.

22. The method of claim 21 when the one or more implemented network energy saving techniques and / or the selected one or more network energy saving techniques comprise implementing discontinuous transmission and / or discontinuous reception, wherein the discontinuous transmission and / or discontinuous reception is implemented by switching one or more carriers on or off at the symbol level, subframe level or radio frame level.

23. The method of any previous claim, further comprising training the machine learning model based on the historical data received from one or more network nodes.

24. The method of claim 23, wherein training the machine learning model comprises computing an estimated network energy saving of each of the one or more network energy saving techniques based on the network energy consumption.

25. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any preceding claim.

26. A computer-readable storage medium having stored thereon the computer program of claim 25.

Images