Bandwidth assignment and energy conservation

Abstract

A computer-implemented method is provided performed by a computing device. The method includes receiving (804) data including a cell state for a cell in the communication network. The method further includes determining (812) a bandwidth assignment for the cell based on (i) a first machine learning, ML, agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) a second ML agent that is aware of a key performance indicator, KPI, for the cell and provides a recommended bandwidth selection based on an observed physical resource block, PRB, value for the cell state. Related methods and apparatus are also provided.

Description

Bandwidth Assignment and Energy Conservation TECHNICAL FIELD

[0001] The present disclosure relates generally to computer-implemented methods performed by a computing device for bandwidth assignment and energy conservation in a communication network, and related methods and devices. BACKGROUND

[0002] Long Term Evolution (LTE) is the fourth-generation broadband wireless access technology developed by Third Generation Partnership Project (3GPP) that may provide higher user data rates, improved system throughput, reduced latency, improve coverage and reduced cost. In deployment on a Communication Service Provider (CSP) network, multiple radio sites can be configured for an LTE radio system.

[0003] During deployment, assignment of bandwidth for each LTE site / cell is an important operation since both data rate and system throughput depend on the availability of bandwidth for services in LTE systems in different spectrums. A total of six system bandwidths are supported by the LTE: 1.4MHz, 3MHz, 5MHz, 10MHz, 15MHz and 20MHz.

[0004] Selection of bandwidth can play a vital role in optimizing the network for energy efficiency. Variation of bandwidth on different nodes can impact performance if not selected and set appropriately.

[0005] Machine learning (ML) based models may be expected to give promising performance to improve the selection and operations related to reduction of energy consumption on radio base stations (RBS or BS) by assigning variable bandwidth enabled by a ML based model.

[0006] Offline policy learning (OPL) is a technique in ML where an agent learns a policy or strategy using pre-existing data collected by unknown behavior policies, without requiring real- time interaction with the environment.

[0007] In OPL, the ML agent may only have access to a fixed dataset and, therefore, no active exploration is involved. OPL can allow the ML agent to learn from a dataset offline, which may enable the ML agent to make informed decisions when facing similar situations in the future. OPL may be particularly useful in scenarios where direct interaction with the environment is costly, time-consuming, and / or not feasible.

[0008] There has been a surge in interest in OPL due to, e.g., its relevance to practical scenarios where abundant logged experiences are available, but direct interaction with the environment is often limited.

[0009] However, the data-driven nature of OPL may lead to several difficulties and challenges. For example, outcomes for actions that were not chosen in the historical data cannot be observed, which are referred to as counterfactual outcomes. Such unseen state action pairs may be erroneously estimated to have unrealistic values and, thus, may cause extrapolation error. SUMMARY

[0010] There currently exist certain challenges. Modifying configurable aspects of a network node (e.g., a BS) can result in varying power consumption levels, which may be a problem when optimizing a network. Some approaches to assigning bandwidth in communication networks (such as an LTE system) may be primarily static. Therefore, selecting a most favorable configuration may be of high importance to reduce energy consumed by the entire network. However, when there is high traffic or low traffic, a higher bandwidth may be overprovisioned which may cause increased energy consumption. Among various adjustable configurations, bandwidth may be one important factor affecting power consumption.

[0011] Some approaches may manually adjust bandwidth for a site through a trial-and-error process, instead of being self-organized for example. However, such manual adjustment may not result in minimal energy consumption and may not adapt quickly to evolving network conditions. Consequently, there may be a need for automated bandwidth management for improved energy- efficiency in a communication network while preserving adequate network performance.

[0012] Certain aspects of the disclosure and their embodiments may provide solutions to these or other challenges.

[0013] Some embodiments provide a computer-implemented method performed by a computing device. The method includes receiving data including a cell state for a cell in the communication network. The method further includes determining a bandwidth assignment for the cell based on (i) a first ML agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) a second ML agent that is aware of a key performance indicator, KPI, for the cell and provides a recommended bandwidth selection based on an observed physical resource block, PRB, value for the cell state.

[0014] Other embodiments provide a computing device. The computing device comprises a first ML agent and a second ML agent in a communication network; at least one processor; and at least one memory connected to the at least one processor and storing program code that is executed by the at least one processor to perform operations. The operations include to receive data including a cell state for a cell in the communication network. The operations further include to determine a bandwidth assignment for the cell based on (i) the first ML agent that identifies acandidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) the second ML agent that is aware of a KPI for the cell and provides a recommended bandwidth selection based on an observed PRB value for the cell state.

[0015] Some embodiments provide a computer program comprising program code to be executed by at least one processor of a computing device comprising a first ML agent and a second ML agent in a communication network. Execution of the program code causes the computing device to perform operations. The operations include to receive data including a cell state for a cell in the communication network. The operations further include to determine a bandwidth assignment for the cell based on (i) the first ML agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) the second ML agent that is aware of a KPI for the cell and provides a recommended bandwidth selection based on an observed PRB value for the cell state.

[0016] Other embodiments include a computer program product comprising a non-transitory storage medium including program code to be executed by at least one processor of a computing device comprising a first ML agent and a second ML agent in a communication network. Execution of the program code causes the computing device to perform operations. The operations include to receive data including a cell state for a cell in the communication network. The operations further include to determine a bandwidth assignment for the cell based on (i) the first ML agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) the second ML agent that is aware of a KPI for the cell and provides a recommended bandwidth selection based on an observed PRB value for the cell state.

[0017] Certain embodiments may provide one or more of the following technical advantage(s). Based on the operations of the first ML model and the second ML model, the bandwidth assignment and network performance may be both realistic and practical implementation on real-world communication networks. Further, energy savings may be achieved, which may contribute to improved sustainability by reducing operating expenditures that are associated with carbon dioxide emissions. When the bandwidth based on the ML model recommendation, dynamically changes in time, during a 24 day, or weekly, the power used from the grid , is also changing, and are taking and produced from different cleaned and non-clean energy source with variable carbon dioxide emission. If the emission of the produced source is high e.g., from a coal plant, the emissions are high. At the same time, we shall consider reducing energy usage of the radio-based station and reducing the power from the grid. So, by reducing the power from grid, carbon dioxide emissions are lowered. Moreover, based on the inclusion of a determination and consideration of how much degradation impact there may be on networkperformance, e.g., on latency, throughput, coverage, etc., energy conservation and network performance may be balanced in a decision on whether or not to change a bandwidth of a cell(s), which may enable increased energy savings during bandwidth changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are included to provide a further understanding of the disclosure and are incorporated in and constitute a part of this application, illustrate certain non-limiting embodiments of the present disclosure. In the drawings:

[0019] Figure 1A is a schematic diagram illustrating an example pipeline of operations in accordance with some embodiments;

[0020] Figure 1B is a schematic diagram illustrating an example LTE network populated with three base stations;

[0021] Figure 2 is a plot of a running average of energy consumption for radios in accordance with some embodiments;

[0022] Figure 3 is a plot of a correlation matrix in accordance with some embodiments;

[0023] Figure 4 is a plot of hourly PRB thresholds for different bandwidths and regions in accordance with some embodiments;

[0024] Figure 5 is a plot of predicted hourly energy of a logging policy and trained policies under different weight values in accordance with some embodiments;

[0025] Figure 6 includes three plots of energy consumption for various bandwidths in accordance with some embodiments;

[0026] Figure 7 includes plots for examples of two recommendation engines showing energy- KPI trade-off curves in accordance with some embodiments;

[0027] Figure 8 is a flowchart illustrating operations of a computing device according to some embodiments;

[0028] Figure 9 is a block diagram of computing device according to some embodiments;

[0029] Figure 10 is a block diagram of a communication system in accordance with some embodiments;

[0030] Figure 11 is a block diagram of a network node in accordance with some embodiments; and

[0031] Figure 12 is a block diagram of a virtualization environment in accordance with some embodiments.DETAILED DESCRIPTION

[0032] Some of the embodiments contemplated herein will now be described more fully with reference to the accompanying drawings. Embodiments are provided by way of example to convey the scope of the subject matter to those skilled in the art, in which examples of embodiments of the present disclosure are shown. Inventive concepts may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art. It should also be noted that these embodiments are not mutually exclusive. Components from one embodiment may be tacitly assumed to be present / used in another embodiment.

[0033] As communication networks e.g., LTE networks, fifth generation (5G) networks, sixth generation (6G) networks, etc. continue to expand, their energy consumption may be a concern due to increased costs and environmental impact.

[0034] In some approaches, for example, modifying configurable aspects of a network node may result in varying power consumption levels, which may be a problem when optimizing the communication network. Moreover, in some approaches, assigning bandwidth in communication networks may be primarily a static approach. Therefore, selecting a most favorable configuration may be of high importance in order to try to reduce energy consumed by a cell(s) or an entire network. However, when there is high traffic or low traffic, a higher bandwidth may be overprovisioned which may cause increased energy consumption. Among adjustable configurations, bandwidth may be an important factor that can affect power consumption.

[0035] Other approaches may use manual adjustment of bandwidth for a cell site through a trial-and-error process, instead of being self-organized for example. Such an approach, however, may fail to result in minimal or acceptable energy consumption and may not quickly adapt to network conditions.

[0036] Some other approaches that may focus on energy-efficiency of LTE systems may intelligently switch on / off cells or set cells to a sleep mode with use of ML, for example. Such operations, however, may be designed to handle peak traffic periods, which may account for approximately 28% of network uptime for example, as opposed to handling peak and non-peak traffic periods.

[0037] Other approaches may include operations for performing wireless communication on a bandwidth part (BWP) that include performing communication on a first BWP, changing the first BWP to a second BWP based on a time period related to the BWP, and performingcommunication on the second BWP. Such operations may include a control unit that determines at least one feasible operation of a ML-based device based on information determined or generated using a data analysis process a ML process.

[0038] In some other approaches, when only historical data is extracted from a real-world communication network, an ML model tries to learn a pattern by exploiting the historical data. As a consequence, the ML model can be vulnerable to the quality of the historical data including, for example, that some bandwidths may have a zero or low probability of being selected and / or the size of the data or subset of data may be too large or too small.

[0039] In contrast, some embodiments of the present disclosure include operations that dynamically configure cell bandwidth while achieving acceptable network performance while reducing energy consumption. Thus, energy may be saved in a communication network based on dynamically assigning a bandwidth (e.g., an optimal bandwidth) without affecting network performance or achieving adequate network performance.

[0040] Examples herein include a ML model that formulates a task as an offline contextual bandits problem and implements offline learning and evaluation processes. While some examples herein are discussed in the context of an LTE network, the present disclosure is not so limited. Operations herein may be performed in other networks, such as 5G, 6G, etc.

[0041] Moreover, some embodiments include operations based on two ML agents: a first ML agent referred to herein an “energy agent” and a second ML agent referred to herein as a “teacher agent”. The energy agent and the teacher agent of some embodiments are separate, where the energy agent is directed to energy efficiency and the teacher agent is directed to consideration of how much (if any) degradation impact may occur to network performance if the bandwidth is changed.

[0042] In some examples, the energy agent and the teach agent correlate network operation and bandwidth assignments within a bandwidth recommendation engine, while at the same time having uninterrupted network service.

[0043] In some other examples, a CSP (for example) may have flexibility to customize a desired percentage of energy savings by adjusting a hyper-parameter that determines the relative weight of the two ML agents in a final bandwidth decision.

[0044] Other examples include offline evaluation results including an assessment of an example of the ML model and a demonstration of an ability to conserve energy without compromising network performance.

[0045] Some examples include operations that use the ML agents to optimize energy consumption in a communication network by dynamically assigning optimal bandwidth torespective network nodes, while providing no compromise on performance and a key performance indicator (KPI). KPIs can include, for example, latency, throughput, and / or energy efficiency. At the same time, the ML model can be adjusted for customized energy savings capabilities by ranking radio type, bandwidth, and related physical resource block (PRB) usage, for example.

[0046] Moreover, in some examples, the ML model can consider static power plateaus (e.g., of a radio of BS in idle or sleep mode) that are detected and a dynamic character of a radio of a BS on higher PRBs.

[0047] Some examples herein further include that respective cells of a BS can be configured with one bandwidth per cell level where the recommendation engine recommends downlink (DL) bandwidths for respective cells in a target communication network (e.g., in an LTE network).

[0048] Orthogonal frequency-division multiple access (OFDMA) divides the time-frequency domain into multiple orthogonal subcarriers referred to as resource blocks (RBs), which are shared between multiple user equipments (UEs). Available RBs can be assigned to UEs based on requirements of the UEs, while the unassigned RBs can be turned off to minimize power consumption and reduce interference.

[0049] Power consumption at a BS can be classified into two categories: static power consumption (e.g., during idle or sleep mode) and dynamic power consumption. Static power consumption can primarily pertain to the hardware components (e.g., a radio) within the BS and can be relatively constant for a static mode. In contrast, dynamic power consumption (also referred to herein as communicational power) can vary based on the traffic load between the BS and UEs. The static power consumption is highly dependent on the hardware components, and with newer generation of radio even though the static power consumption is reduced, the power consumption is still there.

[0050] A bandwidth decides a capacity for traffic in a BS because the bandwidth determines the total available RBs to serve UEs and, therefore, does not change and is fixed in some networks. For example, channel bandwidths for an LTE network may be as follows: Channel bandwidth (MHz) 1.4 3 5 10 15 20 Number of RBs 6 15 25 50 75 100

[0051] Thus, in some examples, operations are included that may minimize dynamic power consumption for high RBs and static power consumption for low RB by allocating an DL bandwidth (e.g., an optimal DL bandwidth) to respective cells in a BS.

[0052] As discussed further herein regarding Figures 2 and 6, an evaluation of an example according to some embodiments showed that bandwidth variation for low RBs impacts staticpower consumption (which may have not known previously) as well as dynamic power consumption.

[0053] In some examples, data was exploited and a relationship was found between bandwidth, energy consumption, and KPIs of multiple network nodes, static power plateaus, and dynamic levels; and the recommendation engine was trained to recommend optimal bandwidths for static power consumption and the dynamic power consumption.

[0054] Based on finding the example radio static plateaus and dynamic power consumption levels, a technical advantage may include enhancing energy savings. For example, as discussed further herein, an evaluation of example shows energy savings observed by a first recommendation engine of about 14% with no KPI degradation, and energy savings observed by a second recommendation engine of about 6%.

[0055] In some examples, a computing device can recommend bandwidth every hour for respective cells to save energy without affecting the quality of service (QoS). While some examples herein are discussed in the non-limiting context of a computing device that can recommend bandwidth every hour, the present disclosure is not so limited and a bandwidth can be recommended for other time periods e.g., a bandwidth recommendation multiple times in a 24 hour period.

[0056] Figure 1A is a block diagram showing an example pipeline 100 of operation according to some embodiments. Pipeline 100 includes operation that correlate between site bandwidth, energy consumption of the site, and a bandwidth(s) recommendation. As shown, the example pipeline 100 includes: ^ A first operation for data preparation 104 of historical data 102. The data preparation includes preprocessing 104a, feature selection 104b, and analysis and / or visualization 104c. The data preparation 104 outputs a training data set. The historical data 102 can be stored, for example, in node in a communication system e.g., communication system 1000 shown in Figure 10 or in a cloud-based node. ^ A second operation performs data augmentation 106 on the training data set. The data augmentation 106 includes splitting the training data to create a balanced training set of data where each output class (or target class) is represented by the same number of input samples. In an example, the output class can be bandwidth and the input samples are cell states corresponding to their cell bandwidth. Data augmentation 106 can be performed, for example, in node in a communication system e.g., network node 1010 in Figure 10 or another node in Figure 10 comprising a computing device or in a cloud- based node e.g., computing device 1018.^ A third operation, performed by a computing device that includes a recommendation engine 108, to determine a recommended bandwidth assignment(s) 118 based on periodically received input cell states. As shown, the recommendation engine 108 includes two ML agents: a first ML agent 112 that comprises an energy-driven agent and a second ML agent 110 that comprises a teacher agent. The recommendation engine 108 can be included in a computing device e.g., the computing device 900 discussed further herein with respect to Figure 9. The computing device, for example, can be a node in a communication system e.g., network node 1010 in Figure 10, or another node comprising a computing device in communication system 1000 in Figure 10 or cloud-based computing device e.g., computing device 1018 in Figure 10. ^ The first ML agent 112 is trained using augmented data from data augmentation 106 within an offline contextual bandits framework with an objective of selecting the bandwidth that minimizes energy consumption. On the other hand, the second ML agent 110 is KPI-aware, which, in an example, always selects bandwidths without harming the KPI performance based on insights gained from historical data 102. The first ML agent 112 and the second ML agent can be included in the computing device e.g., as shown in Figure 9. ^ Further, in some examples, the recommendation engine 108 determines a static power plateau e.g., for low RBs and dynamic variation e.g., for high RBs, as discussed further herein in connection with Figure 6. ^ A fourth operation includes policy evaluation 116 to test and verify the recommendation engine 108 using a designated test set, as discussed further herein. Policy evaluation 116 can be performed, for example, in node in a communication system e.g., network node 1010 in Figure 10 or another node in Figure 10 comprising a computing device or in a cloud-based node e.g., computing device 1018.

[0057] Operations to automatically select a bandwidth (e.g., an optimal bandwidth) are now discussed further.

[0058] As shown in the example in Figure 1B, one example includes an LTE network populated with N BSs 130a, 103b, 130c; and the sectors 120a, 120b, 120c in each BS 130a, 130b. 130c, respectively, can be further broken down into cells 122a-c, 124a-c, and 126a-c, respectively, which also can be denoted by C୨ൌ ^c^,୨, cଶ,୨, … , c୩,୨^, where c୧,୨represents cell ^^ from all ^^ cells of BS ^^, j ൌ 1,2, … , N; s୨ൌ ^B^,୨, Bଶ,୨, … , B୩,୨^ is the configuration of BS ^^, where B୧,୨indicates the bandwidth of cell ^^ from BS ^^. In this example, the carrier frequency is fixed and is not modified.

[0059] In this example, the configuration of the cells in the network can further be denoted byS ൌ ^s^, sଶ, … , s^^. The power consumption of cell ^^^,^ in a certain time period ^^ is denoted byp^୧,୨ , and the network bandwidth configuration that can minimize energy is formally given by:S୭୮^ ൌ arg min ∑ ∑ p^ ൌ arg m ^ୗ ୧ ୨ ୧,୨ ୗin P (Equation 1)where P^denotes the power consumption of the whole network.

[0060] In this example, QoS of the network is guaranteed by achieving specified KPI demands. The KPIs used to describe QoS can include throughput, latency, interference, etc. Further, in this example, the selected KPI collections I୧,୨for cell c୧,୨are specified to be above threshold θ୧,୨. The problem now becomes an optimization problem subject to constraints in thefollowing Equation 2:S୭୮^ ൌ arg min ∑ ∑ p^ ൌ ^ୗ ୧ ୨ ୧,୨ arg mୗin Ps.t. Ii,j < θi,j (Equation 2)

[0061] In this example, if the total energy consumption under all possible bandwidth configurations can be known, an optimal solution can be found by performing an exhaustive search on the set of samples൫S, P^, I୧,୨൯.

[0062] The first ML agent 112 is now discussed further. The first ML agent 112 can be described, in some examples, as an energy driven agent 112. The energy-driven agent 112 can learn from a training set of ^^^బafter data augmentation 106 to greedily choose the bandwidth that minimizes energy consumption for a current cell state(s), including the static power plateau of the bandwidth and the dynamic variation of the bandwidth, as discussed further with respect to Figure 6.

[0063] A regressor r^ is trained for the first ML agent 112 in this example to predict a reward associated with each possible action a୧∈ ^^ given a specific state. The reward is designed as the negative linear-scaled energy within range [−1, 0], which can later be reversely transformed in this example to hourly power consumption during evaluation. In this way, the regressor ^̂^ also can be an hourly energy predictor. Consequently, the policy π^s^ of the energy driven agent 112 in this example is defined as: π^s^ൌ arg mୟ∈a ^x^ ^̂^^^^, ^^^(Equation 3)

[0064] In this example, the policy makes deterministic choices, where the probability of selecting an action is either 1 or 0. In this example, a deep neural network was trained for regression. However, if the network is trained as normal regressor, it can be tricky to achieve highly accurate predictions and can be biased considering the limited data and model size.

[0065] To address the problem, in this example, a process can be used that implements a weighted squared loss. See e.g., T A. Balakrishnan, D. Bouneffouf, N. Mattei, and F. Rossi, “Incorporating behavioral constraints in online ai systems”, Proceedings of the AAAI Conference on Artificial Intelligence, vol.33, no.01, pp.3–11 at page 18 (2019). This weighted loss takes into account the conditional action probability of the logging policy and incorporates a threshold τ to bond the estimation, providing an unbiased prediction for the new policy value. The loss is denoted by: ൫୰ି^^ ^ మL^r, r^^ ൌ ୰ ^,ୟ ൯ (Equation 4)where π^^a୧|s୧^ is the for state ^^^by the logging policy π^.

[0066] With reference to offline reinforcement learning algorithm Batch-Constrained deep Q- learning (BCQ), in this example, a generative model Gனwas trained as the state-conditioned model of the data set ^^గబ, therefore Gன^a|s^ ^ π^^a|s^. For larger values of ^^ , less accuracy of ^^னisrequired as its robust property. The loss is further re-formulated as:,r^ ൌ ୰ି୰^^ ^ మL^r ^ ൫ ^,ୟ ൯୫ୟ^^ୋಡ^a|s^,த^ (Equation 5)

[0067] The second ML agent 110 is now discussed further. The second ML agent 110 can be described, in some examples, as a teacher agent 110. Continuing with the above example, historical data 102 was extracted from a LTE network, where all bandwidths are manually set based on expert knowledge.

[0068] More specifically, the bandwidths are tailored to handle traffic in peak hours. Utilizing this information, operations include to determine the minimum and maximum average hourly utilized PRB values supported by each bandwidth from the historical data 102.

[0069] Thus, in this example, the teacher agent 110 is KPI-aware and provides expert recommendations for bandwidth selection based on the observed hourly mean PRB value in a given state. Each bandwidth corresponds to a range of supported PRB values, and the teacher agent 110 assesses which range the state falls within, identifies suitable bandwidth (action) options, and assigns recommendation scores to all four available actions. Therefore, the reward of the teacher agent 110 is defined as vector r^^^⃗ ൌ ^rୟ^బ, rୟ^భ, rୟ^మ, rୟ^య൧, where rୟ^^∈ ^0,1^ is the score for the ith action in ^^ ൌ ^5MHz, 10MHz,is proportional to the level ofrecommendation for the current state, which means a higher score indicates a stronger recommendation for the corresponding action. Moreover, a customized weight σ ∈ ^0,1^ is introduced in the final policy π^, which is given asπ^^s^ൌ arg mୟ∈a^x^^^^ ^^௧^^^1 െ ^^^^̂^^^^, ^^^^  (Equation 6)where ^^ is a user-configurable the energy-driven agent 112 and theKPI-aware agent 110.

[0071] In some examples, the value of ^^ is a user-configurable weight that balances between the energy-driven agent 112 and the KPI-aware agent 110.

[0072] Further, in some examples, by adjusting ^^, the recommendation engine 108 can be customized to various scenarios, ranging from a purely energy-saving approach to a more conservative KPI-oriented approach. This flexibility may allow for exploring the trade-off between energy-saving percentage and KPI degradation as the value of σ is changed.

[0073] Policy evaluation 116 is now discussed further. In this example, during policy evaluation three criteria are examined: 1. How much energy can be saved with the recommendation engine 108? 2. Is KPI affected by saving energy? If so, how much KPI degradation is caused? 3. What is the trade-off between energy-saving percentage and KPI degradation?

[0074] Regarding the first criteria, as previously discussed in this example, a reward (e.g., energy) predictor^ r^s, a^ and probability estimator ^^னare trained, which can again be utilized in policy evaluation 116.

[0075] Further, in this example, Direct Method (DM) and Doubly Robust (DR) estimation are applied, which are two conventional offline evaluation processes for contextual bandits, to predict the average reward in n recommendation rounds under the policy.

[0076] By scaling this predicted reward back to average energy consumption, it can be compared with the original mean consumption of the network over n rounds, thereby calculating the percentage of energy saved. Given that the trained policy makes a determined decision (Equation 6), the DM estimator can be simplified to the following Equation 7: (Equation 7)where I( ^) is an indicator function evaluating to one if its argument is true and zero otherwise.

[0077] Similarly, the DR estimator can be re-written as the following Equation 8:(Equation 8)

[0078] For the second and third criterion, the previously discussed PRB value range for each bandwidth is used to assess whether the selected bandwidth can support current traffic state. This may allow for calculating a rate of incorrect selections, which may serve as an indicator of the degree of KPI degradation. By adjusting the value of ^^ from 1 to 0, how the final policy πσ prioritizes energy efficiency may be observed, the trade-off between energy-saving percentage and KPI degradation may be observed.

[0079] Determination of a recommended bandwidth(s) will now be discussed further.

[0080] As previously discussed, the energy consumption of a cell is directly influenced by its serving radio and the traffic load (e.g., low and high PRB utilization). Figure 2 investigates this relationship. Figure 2 is a plot of running average of energy consumption for four radio, where “radio_x” is a radio type and the number following the radio type (e.g., “31” following :radio_a”) is its encoded number (e.g., rank). As shown in Figure 2, the running average of consumed energy against the hourly mean DL PRB shows that the consumed energy is positively related to the PRB value.

[0081] Continuing with the example, thirteen features other than bandwidth and consumed energy were selected 104b during data preparation 104 to formulate the cell states, including DL PRB utilization, energy configuration parameters, radio type, locations, number of antennas, etc. The continuous features in the state were further normalized for the example process, and are shown in correlation matrix shown in Figure 3.

[0082] Given the features and the correlation matrix, in this example, the maximal average hourly utilized PRB supported by each bandwidth was determined, as shown in Figure 2. These values were implemented later as PRB thresholds to assess the correctness of bandwidth selection in maintaining KPIs and QoS.

[0083] Further, also in this example, an analysis was conducted to investigate the impact of both bandwidth and radio type on energy consumption, related to static power plateaus at low RB and dynamic variation for higher RB. By comparing the energy consumption of cells with the same radio type as shown in Figure 4 for regions, R, R1-R4, but different bandwidths, it can be determined which bandwidth option tends to result in higher energy consumption. Thisdetermination may help to identify that the recommended bandwidth(s) aligns with an energy- saving objective.

[0084] An evaluation of results of this example and energy savings was performed. The recommendation engine 108 was tested using both unbiased and biased reward sets for the teacher agent 110, which are referred to herein as Engine 1 and Engine 2, respectively herein. For Engine 1 with unbiased teacher reward, Engine 1 had no preference over certain bandwidths and the embedded teacher agent 110 assigns the same score for supportive bandwidths. While for the biased Engine 2, Engine 2 tended to select the smallest bandwidth that can handle current traffic to achieve bandwidth-efficiency.

[0085] The performance of Engines 1 and 2 were evaluated using the same ^^ values for comparison. During the experiment, the result of saved energy and KPI influence did not linearly respond to the ^^ value, so six ^^ values of different intervals were chosen. Figure 5 shows box plots of rewards from all test samples under multiple policies corresponding to the varied ^^ values.

[0086] Accuracy of the embedded reward (hourly energy) regressor ^̂^ also was assessed by having the recommendation engine 108 select same actions as that in the test data, thus mimicking the logging policy ^^^. In this way, the reward of the recommendation engine 108 becomes the prediction of ^^^’s reward. By comparing this prediction with real value, the accuracy of the regressor ^̂^ can be evaluated as shown in Figure 6.

[0087] Corresponding relative energy values and energy saving rates for this evaluation are summarized in the following Table 1: Logging Policy Engine 1 Engine 2 π0(True) Real mean hourly energy X π0 (Predict) DM estimation: X - 0.53 Accuracy: 99.7% DR Estimation: X + 1.3 Accuracy: 99.4% Trained Policy DM DR Saving Rate DM DR Saving Rate πσσ = 0 X–30.43 X–30.42 15.1% X–30.43 X–30.42 15.1% σ = 0.1 X–29.81 X–29.45 14.7% X-24.21 X-24.14 12.0% σ = 0.2 X–28.33 X–28.33 14.0% X-20.05 X-19.93 9.9% σ = 0.3 X–28.34 X–28.18 13.9% X-14.15 X-14.02 6.9% σ = 0.9 X–28.34 X–28.18 13.9% X-12.9 X-12.82 6.4%σ = 1 X–12.9 X–12.56 6.4% X-12.9 X-12.82 6.4%

[0088] Energy-KPI trade-off of each policy is shown for this evaluation in the following Table 2:

[0089] As shown in Tables 1 and 2, it can be observed that Engine 1 with unbiased teacher agent 110 leads to lower KPI degradation and a higher percentage of saved energy. However, as ^^ value increases from 0.1 to 1, the percentage of saved energy for Engine 2 ranges from 6% to 12%, while for Engine 1 it ranges from 14% to 15%.

[0090] This indicates that Engine 1 maintains a relatively consistent energy-saving outcome over a wide range of ^^ values, which may mean that Engine 1 may be more challenging to customize the saved energy specifically in this scenario. Moreover, both types of engines demonstrate a non-linear relationship between the two criteria.

[0091] To better understand the trade-off property, this evaluation examined smaller σ values between 0 to 0.3 as shown in the following Table 3:

[0092] To account for the slight fluctuations in the results obtained from different test data, 5000 rounds of repeated testing were conducted and the outcomes were averaged for both Engine 1 and Engine 2. In each round, a randomly selected subset of test data was evaluated using apredefined set of ^^ values. Through trial and error, a specific set of σ values was designed for each of Engine 1 and Engine 2 that effectively captured the trade-off curve.

[0093] Analysis of the trade-off curve for this example showed that Engine 1 achieved the best energy saving percentage at approximately 14% with zero KPI degradation, while Engine 2 attained about 6% energy savings.

[0094] The comprehensive numerical results of the trade-off curves is shown in Table 3, revealing a distinct non-linear relationship. As the ^^ value decreases and the energy-driven agent’s 112 influence grows stronger, the points on the curve become denser along the x-axis direction in Figure 7. This indicates that when the energy consumption is already relatively low, even a slight increase in the percentage of saved energy will result in significant KPI degradation. Furthermore, Engine 1 utilizing an unbiased teacher agent’s 110 reward was particularly sensitive to small ^^ values in terms of KPI degradation.

[0095] In another example, the indexes in Table 3 can be used to enable the operations of the process to determine and consider how much the degradation impact will be (e.g., different grades), and whether or not to recommend a change of the bandwidth. Some conditions, for example, for degradation (e.g., latency, throughput, coverage, etc.) may be acceptable for some uses to enable more savings during bandwidth changes.

[0096] Operations of a computing device can be performed by the computing device 900 of Figure 9. Operations of the computing device implemented using the structure of Figure 9 are now discussed with reference to the flow chart of Figure 8 according to some embodiments of the present disclosure. Operations 800, 802, 806-810, and 814 from the flow chart of Figure 8 may be optional with respect to some embodiments of computing devices and related methods. Modules may be stored in memory 905, the first ML agent 112, and / or the second ML agent 110 of Figure 9, for example, and these modules may provide instructions so that when the instructions of a module are executed by respective computing device processor 903 (also referred to herein as processing circuitry), computing device 900 performs respective operations of the flow chart of Figure 8. While the computing device 900 is shown as including the first ML agent 112 and / or the second ML agent, embodiments herein are not so limited. In other embodiments, the first ML agent 112 and / or the second ML agent 110 can be outside the computing device 900 in a cloud- based node (e.g., a server) or in a distributed network (e.g., in different cloud-based nodes).

[0097] Referring to Figure 8, in some embodiments, a computer-implemented method is provided that is performed by a computing device, e.g., for bandwidth assignment and energy conservation in a communication network. The method includes receiving (804) data including a cell state for a cell in the communication network. The method further includes determining (812)a bandwidth assignment for the cell based on (i) a first ML agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) a second ML agent that is aware of a KPI for the cell and provides a recommended bandwidth selection based on an observed PRB value for the cell state.

[0098] Determining (812), in some embodiments, uses a specified weight applied to the second ML agent to determine the bandwidth assignment.

[0099] In some embodiments, the specified weight balances a trade-off between the first ML agent and the second ML agent by adjusting their weight in the final bandwidth recommendation. The higher the weight is, the more trust and reliance is placed on a certain agent.

[0100] In some embodiments, the method further includes repeating (814) the receiving (804) and the determining (812) on a plurality of occasions in a period (e.g., a twenty-four-hour period) achieved by adjustable working and inference frequency.

[0101] The received data can include augmented data including a plurality of data sets for a plurality of bandwidths obtained by counterfactual data augmentation. Counterfactual data augmentation refers to a process to add hypothetical samples in an imbalanced data set for those data classes having an insufficient amount of samples. By doing so, the dataset can be balanced to avoid, or diminish, a biased training result for the ML agent. The counterfactual data augmentation can be applied to the plurality of data sets. A respective data set can include a state of the cell in a time period, a bandwidth for the state of the cell in the time period, and a reward representing energy consumption of the cell for the state and the bandwidth.

[0102] The plurality of data sets can be obtained from configuration management data for a first periodicity and performance management data for a second periodicity for a period of time for a plurality of base stations in the communication network.

[0103] The configuration management data can include one or more of the following configuration management parameters per cell per base station for the plurality of base stations (i) a downlink bandwidth, (ii) a radio type, (iii) an energy consumption parameter, and (iv) a number of antenna.

[0104] The performance management data can include one or more of the following, a state of the communication network including traffic information expressed as a number of PRBs in uplink per cell per base station for the plurality of base stations, downlink per cell per base station for the plurality of base stations, and energy consumption per a radio unit of a base station per cell.

[0105] In some embodiments, the method further includes processing (800) historical data to associate at least one configuration management parameter of the cell with the cell state to determine a consumed energy of the cell; and obtaining (802) the plurality of data sets.

[0106] In some embodiments, the first ML agent is trained using the augmented data within an offline contextual bandits framework to select a bandwidth that minimizes energy consumption for the cell state. The cell state includes a static energy level of the bandwidth and a variation of the energy level of the bandwidth.

[0107] Further, in some embodiments, the second ML agent learns from historical data a plurality of observed PRB values for a time periodicity supported by respective bandwidths in configured bandwidths to handle traffic in peak time periods.

[0108] The respective bandwidths can correspond to a respective range of supported PRB values.

[0109] In some embodiments, the second ML agent (i) determines which range the cell state falls within, (ii) identifies candidate bandwidth options from the range, and (iii) assigns respective recommendation scores to the candidate bandwidth options.

[0110] In some embodiments, a respective recommendation score has a value in a first range.

[0111] Further, in some embodiments, the respective recommendation score that has a higher value in the range indicates a stronger recommendation for the respective candidate bandwidth option.

[0112] Determining (812) the bandwidth assignment can include a plurality of bandwidth assignments for a plurality of cells in the communication network that minimize a total energy consumption of the plurality of cells in the communication network.

[0113] In some embodiments, a quality of service of the communication network satisfies a plurality of KPIs for the plurality of cells for the plurality of bandwidth assignments that minimize the total energy consumption.

[0114] The specified weight, in some embodiments, is adjustable over a second range of values between a first end and a second end of the second range. In some embodiments, (i) the first end of the range results in the bandwidth assignment for the cell that maximizes energy conservation for the cell state, and (ii) the second end of the range results in the bandwidth assignment that maximizes the KPI for the cell.

[0115] In some embodiments, the method further includes evaluating (810) the bandwidth assignment based on one or more of (i) an amount of energy saved, (ii) whether the KPI is affected by the energy saved and, if so, an amount of degradation of the KPI, and (iii) assessing a trade-off between the amount of energy saved and degradation of the KPI.

[0116] In some embodiments, the energy consumption of the cell is associated with a serving radio of the cell and a PRB utilization of the cell, and the method further includes ranking (806) aplurality of serving radios of respective cells in the plurality of cells based on respective energy consumption given an identical PRB value.

[0117] Further, in some embodiments, determining (808) a plurality of respective threshold values for respective KPIs in a plurality of KPIs is based on a maximal average utilized PRB supported by respective bandwidths for a periodicity.

[0118] As shown in Figure 9, the computing device 900 includes processor 903 that is operatively coupled to memory 905, network interface 907, first ML agent 112, second ML agent 110, and / or any other component, or any combination thereof. Certain computing devices may utilize all or a subset of the components shown in Figure 9. The level of integration between the components may vary from one computing device to another computing device. Further, certain computing devices may contain multiple instances of a component, such as multiple processors, memories, etc.

[0119] The processor 903 is configured to process instructions and data and may be configured to implement any sequential state machine operative to execute instructions stored as machine-readable computer programs in the memory 905, first ML agent 112, and / or second ML agent 110. The processor 903 may be implemented as one or more hardware-implemented state machines (e.g., in discrete logic, field-programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), etc.); programmable logic together with appropriate firmware; one or more stored computer programs, general-purpose processors, such as a microprocessor or digital signal processor (DSP), together with appropriate software; or any combination of the above. For example, the processor 903 may include multiple central processing units (CPUs).

[0120] In the example, the network interface 907 may be configured to provide an interface or interfaces to an input device, output device, or one or more input and / or output devices. Examples of an output device include a display, a monitor, a printer, another output device, or any combination thereof. An input device may allow a user to capture information into the computing device 900. Examples of an input device include a touch-sensitive or presence-sensitive display, a camera (e.g., a digital camera, a digital video camera, a web camera, etc.), a microphone, a sensor, a mouse, a trackball, a directional pad, a trackpad, a scroll wheel, a smartcard, and the like. The presence-sensitive display may include a capacitive or resistive touch sensor to sense input from a user. A sensor may be, for instance, a force sensor, an optical sensor, a proximity sensor, a biometric sensor, etc., or any combination thereof. An output device may use the same type of interface port as an input device. For example, a Universal Serial Bus (USB) port may be used to provide an input device and an output device.

[0121] The memory 905, first ML agent 112, and / or second ML agent 110 may be or be configured to include memory such as random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic disks, optical disks, hard disks, removable cartridges, flash drives, and so forth. In one example, the memory 905, first ML agent 112, and / or second ML agent 110 includes one or more application programs, such as an operating system, web browser application, a widget, gadget engine, or other application, and corresponding data. The memory 905, first ML agent 112, and / or second ML agent 110 may store, for use by the computing device 900, any of a variety of various operating systems or combinations of operating systems.

[0122] The memory 905, first ML agent 112, and / or second ML agent 110 may be configured to include a number of physical drive units, such as redundant array of independent disks (RAID), flash memory, USB flash drive, external hard disk drive, thumb drive, pen drive, key drive, high- density digital versatile disc (HD-DVD) optical disc drive, internal hard disk drive, Blu-Ray optical disc drive, holographic digital data storage (HDDS) optical disc drive, external mini-dual in-line memory module (DIMM), synchronous dynamic random access memory (SDRAM), external micro-DIMM SDRAM, smartcard memory such as tamper resistant module in the form of a universal integrated circuit card (UICC) including one or more subscriber identity modules (SIMs), such as a USIM and / or ISIM, other memory, or any combination thereof. The UICC may for example be an embedded UICC (eUICC), integrated UICC (iUICC) or a removable UICC commonly known as ‘SIM card.’ The memory 905, first ML agent 112, and / or second ML agent 110 may allow the computing device 900 to access instructions, application programs and the like, stored on transitory or non-transitory memory media, to off-load data, or to upload data. An article of manufacture, such as one utilizing a communication system may be tangibly embodied as or in the memory 905, first ML agent 112, and / or second ML agent 110 which may be or comprise a device-readable storage medium.

[0123] The processor 903 may be configured to communicate with a network using the network interface 907. The network interface 907 may comprise one or more communication subsystems. The network interface 907 may include one or more transceivers used to communicate, such as by communicating with one or more remote transceivers of another device capable of wireless communication (e.g., another computing device, local computing device, edge node, cloud node, etc.). Each transceiver may include a transmitter and / or a receiver appropriate to provide network communications (e.g., optical, electrical, and so forth).

[0124] In the illustrated embodiment, communication functions of the network interface 907 may include cellular communication, Wi-Fi communication, LPWAN communication, data communication, voice communication, multimedia communication, short-range communications such as Bluetooth, near-field communication, location-based communication such as the use of the global positioning system (GPS) to determine a location, another like communication function, or any combination thereof. Communications may be implemented according to one or more communication protocols and / or standards, such as IEEE 802.11, Code Division Multiplexing Access (CDMA), Wideband Code Division Multiple Access (WCDMA), global system for mobile communications (GSM), long term evolution (LTE), New Radio (NR), Universal Mobile Telecommunications System (UMTS), WiMax, Ethernet, transmission control protocol / internet protocol (TCP / IP), synchronous optical networking (SONET), Asynchronous Transfer Mode (ATM), QUIC, Hypertext Transfer Protocol (HTTP), and so forth.

[0125] Functions implemented by some embodiments may be virtualized. In the present context, virtualizing means creating virtual versions of apparatuses or computing devices which may include virtualizing hardware platforms, storage devices and networking resources. As used herein, virtualization can be applied to any device described herein, or components thereof, and relates to an implementation in which at least a portion of the functionality is implemented as one or more virtual components. Some or all of the functions described herein may be implemented as virtual components executed by one or more VMs implemented in one or more virtual environments hosted by one or more of hardware nodes, such as a hardware computing device that operates as an edge node or cloud node. Further, in embodiments the virtual node may be entirely virtualized.

[0126] Applications (which may alternatively be called software instances, virtual appliances, network functions, virtual nodes, virtual network functions, etc.) can be run in the virtualization environment to implement some of the features, functions, and / or benefits of some of the embodiments disclosed herein.

[0127] Although the computing device described herein may include the illustrated combination of hardware components, other embodiments may comprise computing devices with different combinations of components. It is to be understood that these computing devices may comprise any suitable combination of hardware and / or software needed to perform the tasks, features, functions and methods disclosed herein. Determining, calculating, obtaining or similar operations described herein may be performed by processing circuitry, which may process information by, for example, converting the obtained information into other information, comparing the obtained information or converted information to information stored in thecomputing device, and / or performing one or more operations based on the obtained information or converted information, and as a result of said processing making a determination. Moreover, while components are depicted as single boxes located within a larger box, or nested within multiple boxes, in practice, computing devices may comprise multiple different physical components that make up a single illustrated component, and functionality may be partitioned between separate components. For example, a communication interface may be configured to include any of the components described herein, and / or the functionality of the components may be partitioned between the processor and the network interface. In another example, non- computationally intensive functions of any of such components may be implemented in software or firmware and computationally intensive functions may be implemented in hardware.

[0128] In certain embodiments, some or all of the functionality described herein may be provided by processing circuitry executing instructions stored on in memory, which in certain embodiments may be a computer program product in the form of a non-transitory computer- readable storage medium. In alternative embodiments, some or all of the functionalities may be provided by the processing circuitry without executing instructions stored on a separate or discrete device-readable storage medium, such as in a hard-wired manner. In any of those particular embodiments, whether executing instructions stored on a non-transitory computer-readable storage medium or not, the processing circuitry can be configured to perform the described functionality. The benefits provided by such functionality are not limited to the processing circuitry alone or to other components of the computing device, but are enjoyed by the computing device as a whole, and / or by end users and a wireless network generally.

[0129] In certain embodiments, a computing device (900, 1018, 1010) is provided. The computing device (900, 1018, 1010) includes a first ML agent (112) and a second ML agent (110) in a communication network (1000); at least one processor (903, 11302); and at least one memory (905, 11304) connected to the at least one processor (903, 11302) and storing program code that is executed by the at least one processor to perform operations. The operations include to perform some or all of the functionality described herein.

[0130] In certain embodiments, a computer program including program code to be executed by a processor (903, 11302) of a computing device (900, 1018, 1010) including a first ML agent (112) and a second ML agent (110) in a communication network (1000). Execution of the program code causes the computing device to perform operations. The operations include to perform some or all of the functionality described herein.

[0131] In certain embodiments, a computer program product including a non-transitory storage medium (905, 11304) including program code (909, 11324) to be executed by a processor(903, 11302) of a computing device (900, 1018, 1010) including a first ML agent (112) and a second ML agent (110) in a communication network (1000). Execution of the program code causes the computing device to perform operations. The operations include to perform some, or all of the functionality described herein.

[0132] Figure 10 shows an example of a communication system 1000 in accordance with some embodiments.

[0133] In the example, the communication system 1000 (also referred to herein as a communication network) includes a telecommunication network 1002 that includes an access network 1004, such as a RAN, and a core network 1006, which includes one or more core network nodes 1008. The access network 1004 includes one or more access network nodes, such as network nodes 1010a and 1010b (one or more of which may be generally referred to as network nodes 1010), or any other similar 3GPP access node or non-3GPP access point. The network nodes 1010 facilitate direct or indirect connection of user equipment (UE) (also referred to herein as “user device”), such as by connecting UEs 1012a, 1012b, 1012c, and 1012d (one or more of which may be generally referred to as UEs 1012) to the core network 1006 over one or more wireless connections. A computing device (e.g., computing device 900) can be a cloud-based computing device 1018, network node 1010, or another node comprising a computing device in communication system 1000.

[0134] Example wireless communications over a wireless connection include transmitting and / or receiving wireless signals using electromagnetic waves, radio waves, infrared waves, and / or other types of signals suitable for conveying information without the use of wires, cables, or other material conductors. Moreover, in different embodiments, the communication system 1000 may include any number of wired or wireless networks, network nodes, UEs, and / or any other components or systems that may facilitate or participate in the communication of data and / or signals whether via wired or wireless connections. The communication system 1000 may include and / or interface with any type of communication, telecommunication, data, cellular, radio network, and / or other similar type of system.

[0135] The UEs 1012 may be any of a wide variety of communication devices, including wireless devices arranged, configured, and / or operable to communicate wirelessly with the network nodes 1010 and other communication devices. Similarly, the network nodes 1010 are arranged, capable, configured, and / or operable to communicate directly or indirectly with the UEs 1012 and / or with other network nodes or equipment in the telecommunication network 1002 to enable and / or provide network access, such as wireless network access, and / or to perform other functions, such as administration in the telecommunication network 1002.

[0136] In the depicted example, the core network 1006 connects the network nodes 1010 to one or more hosts, such as host 1016. These connections may be direct or indirect via one or more intermediary networks or devices. In other examples, network nodes may be directly coupled to hosts. The core network 1006 includes one more core network nodes (e.g., core network node 1008) that are structured with hardware and software components. Features of these components may be substantially similar to those described with respect to the UEs, network nodes, and / or hosts, such that the descriptions thereof are generally applicable to the corresponding components of the core network node 1008. Example core network nodes include functions of one or more of a Mobile Switching Center (MSC), Mobility Management Entity (MME), Home Subscriber Server (HSS), Access and Mobility Management Function (AMF), Session Management Function (SMF), Authentication Server Function (AUSF), Subscription Identifier De-concealing function (SIDF), Unified Data Management (UDM), Security Edge Protection Proxy (SEPP), Network Exposure Function (NEF), and / or a User Plane Function (UPF).

[0137] The host 1016 may be under the ownership or control of a service provider other than an operator or provider of the access network 1004 and / or the telecommunication network 1002, and may be operated by the service provider or on behalf of the service provider. The host 1016 may host a variety of applications to provide one or more service. Examples of such applications include live and pre-recorded audio / video content, data collection services such as retrieving and compiling data on various ambient conditions detected by a plurality of UEs, analytics functionality, social media, functions for controlling or otherwise interacting with remote devices, functions for an alarm and surveillance center, or any other such function performed by a server.

[0138] As a whole, the communication system 1000 of Figure 10 enables connectivity between the UEs, network nodes, and hosts. In that sense, the communication system may be configured to operate according to predefined rules or procedures, such as specific standards that include, but are not limited to: GSM; UMTS; LTE, and / or other suitable 2G, 3G, 4G, 5G standards, or any applicable future generation standard (e.g., 6G); wireless local area network (WLAN) standards, such as the Institute of Electrical and Electronics Engineers (IEEE) 802.11 standards (WiFi); and / or any other appropriate wireless communication standard, such as the Worldwide Interoperability for Microwave Access (WiMax), Bluetooth, Z-Wave, Near Field Communication (NFC) ZigBee, LiFi, and / or any low-power wide-area network (LPWAN) standards such as LoRa and Sigfox.

[0139] In some examples, the telecommunication network 1002 is a cellular network that implements 3GPP standardized features. Accordingly, the telecommunications network 1002 may support network slicing to provide different logical networks to different devices that areconnected to the telecommunication network 1002. For example, the telecommunications network 1002 may provide URLLC services to some UEs, while providing eMBB services to other UEs, and / or mMTC / Massive IoT services to yet further UEs.

[0140] In some examples, the UEs 1012 are configured to transmit and / or receive information without direct human interaction. For instance, a UE may be designed to transmit information to the access network 1004 on a predetermined schedule, when triggered by an internal or external event, or in response to requests from the access network 1004. Additionally, a UE may be configured for operating in single- or multi-RAT or multi-standard mode. For example, a UE may operate with any one or combination of Wi-Fi, NR and LTE, i.e. being configured for multi-radio dual connectivity (MR-DC), such as E-UTRAN (Evolved-UMTS Terrestrial Radio Access Network) New Radio – Dual Connectivity (EN-DC).

[0141] In the example, the hub 1014 communicates with the access network 1004 to facilitate indirect communication between one or more UEs (e.g., UE 1012c and / or 1012d) and network nodes (e.g., network node 1010b). In some examples, the hub 1014 may be a controller, router, content source and analytics, or any of the other communication devices described herein regarding UEs. For example, the hub 1014 may be a broadband router enabling access to the core network 1006 for the UEs. As another example, the hub 1014 may be a controller that sends commands or instructions to one or more actuators in the UEs. Commands or instructions may be received from the UEs, network nodes 1010, or by executable code, script, process, or other instructions in the hub 1014. As another example, the hub 1014 may be a data collector that acts as temporary storage for UE data and, in some embodiments, may perform analysis or other processing of the data. As another example, the hub 1014 may be a content source. For example, for a UE that is a VR headset, display, loudspeaker or other media delivery device, the hub 1014 may retrieve VR assets, video, audio, or other media or data related to sensory information via a network node, which the hub 1014 then provides to the UE either directly, after performing local processing, and / or after adding additional local content. In still another example, the hub 1014 acts as a proxy server or orchestrator for the UEs, in particular if one or more of the UEs are low energy IoT devices.

[0142] The hub 1014 may have a constant / persistent or intermittent connection to the network node 1010b. The hub 1014 may also allow for a different communication scheme and / or schedule between the hub 1014 and UEs (e.g., UE 1012c and / or 1012d), and between the hub 1014 and the core network 1006. In other examples, the hub 1014 is connected to the core network 1006 and / or one or more UEs via a wired connection. Moreover, the hub 1014 may be configured to connect to an M2M service provider over the access network 1004 and / or to another UE over a directconnection. In some scenarios, UEs may establish a wireless connection with the network nodes 1010 while still connected via the hub 1014 via a wired or wireless connection. In some embodiments, the hub 1014 may be a dedicated hub – that is, a hub whose primary function is to route communications to / from the UEs from / to the network node 1010b. In other embodiments, the hub 1014 may be a non-dedicated hub – that is, a device which is capable of operating to route communications between the UEs and network node 1010b, but which is additionally capable of operating as a communication start and / or end point for certain data channels.

[0143] Figure 11 shows a network node 11300 in accordance with some embodiments. As used herein, network node refers to equipment capable, configured, arranged and / or operable to communicate directly or indirectly with a UE and / or with other network nodes, computing devices, or equipment, in a telecommunication network. Examples of network nodes include, but are not limited to, access points (APs) (e.g., radio access points), base stations (BSs) (e.g., radio base stations, Node Bs, evolved Node Bs (eNBs) and NR NodeBs (gNBs)), computing devices, O-RAN nodes or components of an O-RAN node (e.g., O-RU, O-DU, O-CU).

[0144] Base stations may be categorized based on the amount of coverage they provide (or, stated differently, their transmit power level) and so, depending on the provided amount of coverage, may be referred to as femto base stations, pico base stations, micro base stations, or macro base stations. A base station may be a relay node or a relay donor node controlling a relay. A network node may also include one or more (or all) parts of a distributed radio base station such as centralized digital units, distributed units (e.g., in an O-RAN access node) and / or remote radio units (RRUs), sometimes referred to as Remote Radio Heads (RRHs). Such remote radio units may or may not be integrated with an antenna as an antenna integrated radio. Parts of a distributed radio base station may also be referred to as nodes in a distributed antenna system (DAS).

[0145] Other examples of network nodes include multiple transmission point (multi-TRP) 5G access nodes, multi-standard radio (MSR) equipment such as MSR BSs, network controllers such as radio network controllers (RNCs) or base station controllers (BSCs), base transceiver stations (BTSs), transmission points, transmission nodes, multi-cell / multicast coordination entities (MCEs), Operation and Maintenance (O&M) nodes, Operations Support System (OSS) nodes, Self-Organizing Network (SON) nodes, positioning nodes (e.g., Evolved Serving Mobile Location Centers (E-SMLCs)), and / or Minimization of Drive Tests (MDTs).

[0146] The network node 11300 includes a processing circuitry 11302, a memory 11304 which includes first ML agent 112 and second ML agent 110, a communication interface 11306, and a power source 11308. The network node 11300 may be composed of multiple physically separate components (e.g., a NodeB component and a RNC component, or a BTS component anda BSC component, etc.), which may each have their own respective components. In certain scenarios in which the network node 11300 comprises multiple separate components (e.g., BTS and BSC components), one or more of the separate components may be shared among several network nodes. For example, a single RNC may control multiple NodeBs. In such a scenario, each unique NodeB and RNC pair, may in some instances be considered a single separate network node. In some embodiments, the network node 11300 may be configured to support multiple radio access technologies (RATs). In such embodiments, some components may be duplicated (e.g., separate memory 11304 for different RATs) and some components may be reused (e.g., a same antenna 11310 may be shared by different RATs). The network node 11300 may also include multiple sets of the various illustrated components for different wireless technologies integrated into network node 11300, for example GSM, WCDMA, LTE, NR, WiFi, Zigbee, Z-wave, LoRaWAN, Radio Frequency Identification (RFID) or Bluetooth wireless technologies. These wireless technologies may be integrated into the same or different chip or set of chips and other components within network node 11300.

[0147] The processing circuitry 11302 may comprise a combination of one or more of a microprocessor, controller, microcontroller, central processing unit, digital signal processor, application-specific integrated circuit, field programmable gate array, or any other suitable computing device, resource, or combination of hardware, software and / or encoded logic operable to provide, either alone or in conjunction with other network node 11300 components, such as the memory 11304, to provide network node 11300 functionality.

[0148] In some embodiments, the processing circuitry 11302 includes a system on a chip (SOC). In some embodiments, the processing circuitry 11302 includes one or more of radio frequency (RF) transceiver circuitry 11312 and baseband processing circuitry 11314. In some embodiments, the radio frequency (RF) transceiver circuitry 11312 and the baseband processing circuitry 11314 may be on separate chips (or sets of chips), boards, or units, such as radio units and digital units. In alternative embodiments, part or all of RF transceiver circuitry 11312 and baseband processing circuitry 11314 may be on the same chip or set of chips, boards, or units.

[0149] The memory 11304, including first ML agent 112 and second ML agent 110, may comprise any form of volatile or non-volatile computer-readable memory including, without limitation, persistent storage, solid-state memory, remotely mounted memory, magnetic media, optical media, random access memory (RAM), read-only memory (ROM), mass storage media (for example, a hard disk), removable storage media (for example, a flash drive, a Compact Disk (CD) or a Digital Video Disk (DVD)), and / or any other volatile or non-volatile, non-transitory device-readable and / or computer-executable memory devices that store information, data, and / orinstructions that may be used by the processing circuitry 11302. The memory 11304, including first ML agent 112 and second ML agent 110, may store any suitable instructions, data, or information, including a computer program, software, an application including one or more of logic, rules, code, tables, and / or other instructions capable of being executed by the processing circuitry 11302 and utilized by the network node 11300. The memory 11304, including first ML agent 112 and second ML agent 110, may be used to store any calculations made by the processing circuitry 11302 and / or any data received via the communication interface 11306. In some embodiments, the processing circuitry 11302 and memory 11304, including first ML agent 112 and second ML agent 110, is integrated.

[0150] The communication interface 11306 is used in wired or wireless communication of signaling and / or data between a network node, access network, and / or UE. As illustrated, the communication interface 11306 comprises port(s) / terminal(s) 11316 to send and receive data, for example to and from a network over a wired connection. The communication interface 11306 also includes radio front-end circuitry 11318 that may be coupled to, or in certain embodiments a part of, the antenna 11310. Radio front-end circuitry 11318 comprises filters 11320 and amplifiers 11322. The radio front-end circuitry 11318 may be connected to an antenna 11310 and processing circuitry 11302. The radio front-end circuitry may be configured to condition signals communicated between antenna 11310 and processing circuitry 11302. The radio front-end circuitry 11318 may receive digital data that is to be sent out to other network nodes or UEs via a wireless connection. The radio front-end circuitry 11318 may convert the digital data into a radio signal having the appropriate channel and bandwidth parameters using a combination of filters 11320 and / or amplifiers 11322. The radio signal may then be transmitted via the antenna 11310. Similarly, when receiving data, the antenna 11310 may collect radio signals which are then converted into digital data by the radio front-end circuitry 11318. The digital data may be passed to the processing circuitry 11302. In other embodiments, the communication interface may comprise different components and / or different combinations of components.

[0151] In certain alternative embodiments, the network node 11300 does not include separate radio front-end circuitry 11318, instead, the processing circuitry 11302 includes radio front-end circuitry and is connected to the antenna 11310. Similarly, in some embodiments, all or some of the RF transceiver circuitry 11312 is part of the communication interface 11306. In still other embodiments, the communication interface 11306 includes one or more ports or terminals 11316, the radio front-end circuitry 11318, and the RF transceiver circuitry 11312, as part of a radio unit (not shown), and the communication interface 11306 communicates with the baseband processing circuitry 11314, which is part of a digital unit (not shown).

[0152] The antenna 11310 may include one or more antennas, or antenna arrays, configured to send and / or receive wireless signals. The antenna 11310 may be coupled to the radio front-end circuitry 11318 and may be any type of antenna capable of transmitting and receiving data and / or signals wirelessly. In certain embodiments, the antenna 11310 is separate from the network node 11300 and connectable to the network node 11300 through an interface or port.

[0153] The antenna 11310, communication interface 11306, and / or the processing circuitry 11302 may be configured to perform any receiving operations and / or certain obtaining operations described herein as being performed by the network node. Any information, data and / or signals may be received from a UE, another network node and / or any other network equipment. Similarly, the antenna 11310, the communication interface 11306, and / or the processing circuitry 11302 may be configured to perform any transmitting operations described herein as being performed by the network node. Any information, data and / or signals may be transmitted to a UE, another network node and / or any other network equipment.

[0154] The power source 11308 provides power to the various components of network node 11300 in a form suitable for the respective components (e.g., at a voltage and current level needed for each respective component). The power source 11308 may further comprise, or be coupled to, power management circuitry to supply the components of the network node 11300 with power for performing the functionality described herein. For example, the network node 11300 may be connectable to an external power source (e.g., the power grid, an electricity outlet) via an input circuitry or interface such as an electrical cable, whereby the external power source supplies power to power circuitry of the power source 11308. As a further example, the power source 11308 may comprise a source of power in the form of a battery or battery pack which is connected to, or integrated in, power circuitry. The battery may provide backup power should the external power source fail.

[0155] Embodiments of the network node 11300 may include additional components beyond those shown in Figure 11 for providing certain aspects of the network node’s functionality, including any of the functionality described herein and / or any functionality necessary to support the subject matter described herein. For example, the network node 11300 may include user interface equipment to allow input of information into the network node 11300 and to allow output of information from the network node 11300. This may allow a user to perform diagnostic, maintenance, repair, and other administrative functions for the network node 11300.

[0156] Figure 12 is a block diagram illustrating a virtualization environment 12500 in which functions implemented by some embodiments may be virtualized. In the present context, virtualizing means creating virtual versions of apparatuses or devices which may includevirtualizing hardware platforms, storage devices and networking resources. As used herein, virtualization can be applied to any device described herein, or components thereof, and relates to an implementation in which at least a portion of the functionality is implemented as one or more virtual components. Some or all of the functions described herein may be implemented as virtual components executed by one or more VMs implemented in one or more virtual environments 12500 hosted by one or more of hardware nodes, such as a hardware network node that operates as a computing device, network node, UE, core network node, or host. Further, in embodiments in which the virtual node does not require radio connectivity (e.g., a core network node or host), then the node may be entirely virtualized.

[0157] Applications 12502 (which may alternatively be called software instances, virtual appliances, network functions, virtual nodes, virtual network functions, etc.) are run in the virtualization environment 12500 to implement some of the features, functions, and / or benefits of some of the embodiments disclosed herein.

[0158] Hardware 12504 includes processing circuitry, memory that stores software and / or instructions executable by hardware processing circuitry, and / or other hardware devices as described herein, such as a network interface, input / output interface, and so forth. Software may be executed by the processing circuitry to instantiate one or more virtualization layers 12506 (also referred to as hypervisors or virtual machine monitors (VMMs)), provide VMs 12508a and 12508b (one or more of which may be generally referred to as VMs 12508), and / or perform any of the functions, features and / or benefits described in relation with some embodiments described herein. The virtualization layer 12506 may present a virtual operating platform that appears like networking hardware to the VMs 12508.

[0159] The VMs 12508 comprise virtual processing, virtual memory, virtual networking or interface and virtual storage, and may be run by a corresponding virtualization layer 12506. Different embodiments of the instance of a virtual appliance 12502 may be implemented on one or more of VMs 12508, and the implementations may be made in different ways. Virtualization of the hardware is in some contexts referred to as network function virtualization (NFV). NFV may be used to consolidate many network equipment types onto industry standard high volume server hardware, physical switches, and physical storage, which can be located in data centers, and customer premise equipment.

[0160] In the context of NFV, a VM 12508 may be a software implementation of a physical machine that runs programs as if they were executing on a physical, non-virtualized machine. Each of the VMs 12508, and that part of hardware 12504 that executes that VM, be it hardware dedicated to that VM and / or hardware shared by that VM with others of the VMs, forms separate virtualnetwork elements. Still in the context of NFV, a virtual network function is responsible for handling specific network functions that run in one or more VMs 12508 on top of the hardware 12504 and corresponds to the application 12502.

[0161] Hardware 12504 may be implemented in a standalone network node with generic or specific components. Hardware 12504 may implement some functions via virtualization. Alternatively, hardware 12504 may be part of a larger cluster of hardware (e.g. such as in a data center or CPE) where many hardware nodes work together and are managed via management and orchestration 12510, which, among others, oversees lifecycle management of applications 12502. In some embodiments, hardware 12504 is coupled to one or more radio units that each include one or more transmitters and one or more receivers that may be coupled to one or more antennas. Radio units may communicate directly with other hardware nodes via one or more appropriate network interfaces and may be used in combination with the virtual components to provide a virtual node with radio capabilities, such as a radio access node, base station, or a computing device. In some embodiments, some signaling can be provided with the use of a control system 12512 which may alternatively be used for communication between hardware nodes and radio units.

[0162] Further definitions and embodiments are discussed below.

[0163] In the above-description of certain embodiments of the present disclosure, it is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which concepts of the present disclosure belong. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of this specification and the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0164] When an element is referred to as being “connected”, “coupled”, “responsive”, or variants thereof to another element, it can be directly connected, coupled, or responsive to the other element or intervening elements may be present. In contrast, when an element is referred to as being “directly connected”, “directly coupled”, “directly responsive”, or variants thereof to another element, there are no intervening elements present. Like numbers refer to like elements throughout. Furthermore, “coupled”, “connected”, “responsive”, or variants thereof as used herein may include wirelessly coupled, connected, or responsive. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. Well-known functions or constructions may not be described in detail for brevity and / orclarity. The term “and / or” (abbreviated “ / ”) includes any and all combinations of one or more of the associated listed items.

[0165] It will be understood that although the terms first, second, third, etc. may be used herein to describe various elements / operations, these elements / operations should not be limited by these terms. These terms are only used to distinguish one element / operation from another element / operation. Thus a first element / operation in some embodiments could be termed a second element / operation in other embodiments without departing from the teachings of concepts of the present disclosure. The same reference numerals or the same reference designators denote the same or similar elements throughout the specification.

[0166] As used herein, the terms “comprise”, “comprising”, “comprises”, “include”, “including”, “includes”, “have”, “has”, “having”, or variants thereof are open-ended, and include one or more stated features, integers, elements, steps, components, or functions but does not preclude the presence or addition of one or more other features, integers, elements, steps, components, functions, or groups thereof. Furthermore, as used herein, the common abbreviation “e.g.”, which derives from the Latin phrase “exempli gratia,” may be used to introduce or specify a general example or examples of a previously mentioned item, and is not intended to be limiting of such item. The common abbreviation “i.e.”, which derives from the Latin phrase “id est,” may be used to specify a particular item from a more general recitation.

[0167] Example embodiments are described herein with reference to block diagrams and / or flowchart illustrations of computer-implemented methods, apparatus (systems and / or devices) and / or computer program products. It is understood that a block of the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, can be implemented by computer program instructions that are performed by one or more computer circuits. These computer program instructions may be provided to a processor circuit of a general purpose computer circuit, special purpose computer circuit, and / or other programmable data processing circuit to produce a machine, such that the instructions, which execute via the processor of the computer and / or other programmable data processing apparatus, transform and control transistors, values stored in memory locations, and other hardware components within such circuitry to implement the functions / acts specified in the block diagrams and / or flowchart block or blocks, and thereby create means (functionality) and / or structure for implementing the functions / acts specified in the block diagrams and / or flowchart block(s).

[0168] These computer program instructions may also be stored in a tangible computer- readable medium that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable mediumproduce an article of manufacture including instructions which implement the functions / acts specified in the block diagrams and / or flowchart block or blocks. Accordingly, embodiments of the present disclosure may be embodied in hardware and / or in software (including firmware, resident software, micro-code, etc.) that runs on a processor such as a digital signal processor, which may collectively be referred to as “circuitry,” “a module” or variants thereof.

[0169] It should also be noted that in some alternate implementations, the functions / acts noted in the blocks may occur out of the order noted in the flowcharts. For example, two blocks shown in succession may in fact be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality / acts involved. Moreover, the functionality of a given block of the flowcharts and / or block diagrams may be separated into multiple blocks and / or the functionality of two or more blocks of the flowcharts and / or block diagrams may be at least partially integrated. Finally, other blocks may be added / inserted between the blocks that are illustrated, and / or blocks / operations may be omitted without departing from the scope of the present disclosure. Moreover, although some of the diagrams include arrows on communication paths to show a primary direction of communication, it is to be understood that communication may occur in the opposite direction to the depicted arrows.

[0170] Many variations and modifications can be made to the embodiments without substantially departing from the principles of the present disclosure. All such variations and modifications are intended to be included herein within the scope of present disclosure. Accordingly, the above disclosed subject matter is to be considered illustrative, and not restrictive, and the examples of embodiments are intended to cover all such modifications, enhancements, and other embodiments, which fall within the spirit and scope of the present disclosure. Thus, to the maximum extent allowed by law, the scope of the present disclosure is to be determined by the broadest permissible interpretation of the present disclosure including the examples of embodiments and their equivalents, and shall not be restricted or limited by the foregoing detailed description.

Claims

Claims:

1. A computer-implemented method performed by a computing device, the method comprising: receiving (804) data including a cell state for a cell in the communication network; and determining (812) a bandwidth assignment for the cell based on (i) a first machine learning, ML, agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) a second ML agent that is aware of a key performance indicator, KPI, for the cell and provides a recommended bandwidth selection based on an observed physical resource block, PRB, value for the cell state.

2. The computer-implemented method of Claim 1, wherein the determining (812) uses a specified weight applied to the second ML agent to determine the bandwidth assignment.

3. The computer-implemented method of Claim 2, wherein the specified weight balances a trade-off between the first ML agent and the recommended bandwidth selection of the second ML agent.

4. The computer-implemented method of any one of Claims 1 to 3, further comprising: repeating (814) the receiving (804) and the determining (812) on a plurality of occasions in a period.

5. The computer-implemented method of any one of Claims 1 to 4, wherein the data comprises augmented data comprising a plurality of data sets for a plurality of bandwidths obtained by counterfactual data augmentation.

6. The computer-implemented method of Claim 5, wherein the counterfactual data augmentation is applied to the plurality of data sets, and a respective data set comprises a state of the cell in a time period, a bandwidth for the state of the cell in the time period, and a reward representing energy consumption of the cell for the state and the bandwidth.

7. The computer-implemented method of Claim 6, wherein the plurality of data sets are obtained from configuration management data for a first periodicity and performancemanagement data for a second periodicity for a period of time for a plurality of base stations in the communication network.

8. The computer-implemented method of Claim 7, wherein the configuration management data comprises one or more of the following configuration parameters per cell per base station for the plurality of base stations (i) a downlink bandwidth, (ii) a radio type, (iii) an energy consumption parameter, and (iv) a number of antennas.

9. The computer-implemented method of any one of Claims 7 to 8, wherein the performance management data comprises one or more of the following : a state of the communication network including traffic information expressed as a number of PRBs in uplink per cell per base station for the plurality of base stations, downlink per cell per base station for the plurality of base stations, and energy consumption per a radio unit of a base station per cell.

10. The computer-implemented method of any one of Claims 7 to 9, further comprising: processing (800) historical data to associate at least one configuration management parameter of the cell with the cell state to determine a consumed energy of the cell; and obtaining (802) the plurality of data sets.

11. The computer-implemented method of any one of Claims 5 to 10, wherein the first ML agent is trained using the augmented data within an offline contextual bandits framework to select a bandwidth that minimizes energy consumption for the cell state, and wherein the cell state includes a static energy level of the bandwidth and a variation of the energy level of the bandwidth.

12. The computer-implemented method of any one of Claims 1 to 11, wherein the second ML agent learns from historical data a plurality of observed PRB values for a time periodicity supported by respective bandwidths in configured bandwidths to handle traffic in peak time periods.

13. The computer-implemented method of Claim 12, wherein the respective bandwidths correspond to a respective range of supported PRB values.

14. The computer-implemented method of Claim 13, wherein the second ML agent (i) determines which range the cell state falls within, (ii) identifies candidate bandwidth options from the range, and (iii) assigns respective recommendation scores to the candidate bandwidth options.

15. The computer-implemented method of Claim 14, wherein the respective recommendation score has a value in a first range.

16. The computer-implemented method of Claim 15, wherein the respective recommendation score that has a higher value in the range indicates a stronger recommendation for the respective candidate bandwidth option.

17. The computer-implemented method of any one of Claims 1 to 16, wherein determining (812) the bandwidth assignment comprises a plurality of bandwidth assignments for a plurality of cells in the communication network that minimize a total energy consumption of the plurality of cells in the communication network.

18. The computer-implemented method of Claim 17, wherein a quality of service of the communication network satisfies a plurality of KPIs for the plurality of cells for the plurality of bandwidth assignments that minimize the total energy consumption.

19. The computer-implemented method of any one of Claims 2 to 18, wherein the specified weight is adjustable over a second range of values between a first end and a second end of the second range.

20. The computer-implemented method of Claim 19, wherein (i) the first end of the range results in the bandwidth assignment for the cell that maximizes energy conservation for the cell state, and (ii) the second end of the range results in the bandwidth assignment that maximizes the KPI for the cell.

21. The computer-implemented method of any one of Claims 1 to 20, further comprising: evaluating (810) the bandwidth assignment based on one or more of (i) an amount of energy saved, (ii) whether the KPI is affected by the energy saved and, if so, an amount ofdegradation of the KPI, and (iii) assessing a trade-off between the amount of energy saved and degradation of the KPI.

22. The computer-implemented method of any one of Claims 17 to 21, wherein the energy consumption of the cell is associated with a serving radio of the cell and a PRB utilization of the cell, the method further comprising: ranking (806) a plurality of serving radios of respective cells in the plurality of cells based on respective energy consumption given an identical PRB value.

23. The computer-implemented method of any one of Claims 17 to 22, further comprising: determining (808) a plurality of respective threshold values for respective KPIs in a plurality of KPIs is based on a maximal average utilized PRB supported by respective bandwidths for a periodicity.

24. A computing device (900, 1018, 1010) comprising: a first machine learning, ML, agent (112) and a second ML agent (110) in a communication network (1000); at least one processor (903, 11302); at least one memory (905, 11304) connected to the at least one processor (903, 11302) and storing program code that is executed by the at least one processor to perform operations comprising: receive data including a cell state for a cell in the communication network; and determine a bandwidth assignment for the cell based on (i) the first ML agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) the second ML agent that is aware of a key performance indicator, KPI, for the cell and provides a recommended bandwidth selection based on an observed physical resource block, PRB, value for the cell state.

25. The computing device of Claim 24, wherein the at least one memory (905, 11304) is connected to the at least one processor (903, 11302) and stores program code that is executed by the at least one processor to perform operations according to any one of Claims 2 to 23.

26. A computer program comprising program code to be executed by at least one processor (903, 11302) of a computing device (900, 1018, 1010) comprising a first machine learning, ML, agent (112) and a second ML agent (110) in a communication network (1000), whereby execution of the program code causes the computing device to perform operations comprising: receive data including a cell state for a cell in the communication network; and determine a bandwidth assignment for the cell based on (i) the first ML agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) the second ML agent that is aware of a key performance indicator, KPI, for the cell and provides a recommended bandwidth selection based on an observed physical resource block, PRB, value for the cell state.

27. The computer program of Claim 26, whereby execution of the program code causes the computing device (900, 1018, 1010) to perform operations according to any of Claims 2 to 23.

28. A computer program product comprising a non-transitory storage medium (905, 11304) including program code (909, 11324) to be executed by at least one processor (903, 11302) of a computing device (900, 1018, 1010) comprising a first machine learning, ML, agent (112) and a second ML agent (110) in a communication network (1000), whereby execution of the program code causes the computing device to perform operations comprising: receive data including a cell state for a cell in the communication network; and determine a bandwidth assignment for the cell based on (i) the first ML agent that identifies a candidate bandwidth for the cell that minimizes energy consumption for the cell state, and (ii) the second ML agent that is aware of a key performance indicator, KPI, for the cell and provides a recommended bandwidth selection based on an observed physical resource block, PRB, value for the cell state.

29. The computer program product of Claim 28, whereby execution of the program code causes the computing device to perform operations according to any of Claims 2 to 23.

Images