Bandwidth allocation and power savings

Abstract

A computer-implemented method performed by a computing device is provided. The method comprises receiving (804) data, the data comprising a cell state of a cell in the communication network. The method further includes determining (812) a bandwidth allocation for the cell based on (i) a first machine learning (ML) agent that identifies candidate bandwidths for the cell that minimize energy consumption for the cell state and (ii) a second ML agent that identifies candidate bandwidths for the cell that minimize energy consumption for the cell state, the second ML agent is aware of key performance indicators (KPIs) of the cell and provides recommended bandwidth selections based on physical resource block (PRB) values observed for the state of the cell. Related methods and apparatuses are also provided.

Description

Technical Field

[0001] This disclosure generally relates to computer-implemented methods and related methods and apparatus for bandwidth allocation and energy saving in communication networks, executed by computing devices. Background Technology

[0002] Long Term Evolution (LTE) is a fourth-generation broadband radio access technology developed by the 3rd Generation Partnership Project (3GPP). It can provide higher user data rates, increased system throughput, reduced latency, improved coverage, and lower costs. When deployed on a Communications Service Provider (CSP) network, multiple radio sites can be configured for an LTE radio system.

[0003] Allocating bandwidth to each LTE site / cell is a critical operation during deployment, as both data rates and system throughput depend on the availability of serving bandwidth in different spectrums for the LTE system. LTE supports six system bandwidths: 1.4MHz, 3MHz, 5MHz, 10MHz, 15MHz, and 20MHz.

[0004] Bandwidth selection can play a crucial role in optimizing network energy efficiency. If bandwidth differences between nodes are not properly selected and configured, performance can be negatively impacted.

[0005] Machine learning (ML)-based models can be expected to improve the selection and operation associated with energy reduction of radio base stations (RBS or BS) to achieve good performance by allocating variable bandwidth supported by ML-based models.

[0006] Offline policy learning (OPL) is a technique in machine learning in which an agent learns a policy or decision using pre-existing data collected from unknown behavioral policies without interacting with the environment in real time.

[0007] In OPL, the ML agent can access only a fixed dataset, thus avoiding active exploration. OPL allows the ML agent to learn from the dataset offline, enabling it to make informed decisions when faced with similar situations in the future. OPL can be particularly useful in scenarios where direct interaction with the environment is costly, time-consuming, and / or impractical.

[0008] Interest in OPL has surged due to its relevance to real-world scenarios, where, while extensive recording experience is available, direct interaction with the environment is often limited.

[0009] However, the data-driven nature of OPL can lead to several difficulties and challenges. For example, the consequences of actions not selected in historical data cannot be observed; these consequences are known as counterfactual results. Such unseen state-action pairs may be incorrectly estimated to have unrealistic values, potentially leading to extrapolation errors. Summary of the Invention

[0010] Several challenges exist. Modifying configurable aspects of network nodes (e.g., base stations) can lead to variations in power consumption levels, which can cause problems when optimizing the network. Some methods of allocating bandwidth in communication networks (e.g., LTE systems) may be largely static. Therefore, selecting the most advantageous configuration can be important for reducing energy consumption by the entire network. However, providing excessive bandwidth during periods of high or low traffic can lead to increased energy consumption. Among various tunable configurations, bandwidth is likely to be a significant factor affecting power consumption.

[0011] Some methods might involve manually adjusting site bandwidth through a trial-and-error process, rather than self-organizing. However, such manual adjustments may not achieve minimal energy consumption and may not be able to quickly adapt to changing network conditions. Therefore, communication networks may require automated bandwidth management to improve energy efficiency while maintaining adequate network performance.

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

[0013] Some embodiments provide a computer-implemented method executed by a computing device. The method includes receiving data including cell states of cells in a communication network. The method further includes determining bandwidth allocation for a cell based on (i) a first ML agent and (ii) a second ML agent, the first ML agent identifying candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understanding the cell's key performance indicators (KPIs) and providing recommended bandwidth selection based on physical resource block (PRB) values ​​observed for the cell state.

[0014] Other embodiments provide a computing device. The computing device includes: 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 executed by the at least one processor to perform operations. The operations include receiving data including cell states of cells in the communication network. The operations also include determining a bandwidth allocation for a cell based on (i) the first ML agent and (ii) the second ML agent, the first ML agent identifying candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understanding the cell's KPIs and providing a recommended bandwidth selection based on PRB values ​​observed for the cell state.

[0015] Some embodiments provide a computer program including program code to be executed by at least one processor of a computing device, the computing device including 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 receiving data, including cell states of cells in the communication network. The operations also include determining a bandwidth allocation for a cell based on (i) the first ML agent and (ii) the second ML agent, the first ML agent identifying candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understanding the cell's KPIs and providing a recommended bandwidth selection based on PRB values ​​observed 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, the computing device including 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. These operations include receiving data, including cell states of cells in the communication network. The operations also include determining a bandwidth allocation for a cell based on (i) the first ML agent and (ii) the second ML agent, the first ML agent identifying candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understanding the cell's KPIs and providing a recommended bandwidth selection based on PRB values ​​observed for the cell state.

[0017] Certain embodiments may provide one or more of the following technical advantages. Based on the operation of the first and second ML models, bandwidth allocation and network performance can be both realistically feasible and practically implementable in real-world communication networks. Furthermore, energy savings can be achieved, which helps improve sustainability by reducing operational expenditures associated with CO2 emissions. When the bandwidth recommended based on the ML model changes dynamically over a 24-day or weekly cycle, the electricity used from the grid also changes accordingly, and this electricity comes from different clean and non-clean energy sources with varying CO2 emissions. If emissions from emission sources (e.g., emissions from coal-fired power plants) are high, then carbon emissions are high. Simultaneously, we should consider reducing the energy use of radio base stations and reducing electricity from the grid. Therefore, by reducing electricity from the grid, CO2 emissions are reduced. Furthermore, based on determining and considering the extent to which network performance (e.g., latency, throughput, coverage, etc.) may be degraded, energy savings and network performance can be balanced when deciding whether to change the bandwidth of a cell, which can achieve improved energy savings during bandwidth changes. Attached Figure Description

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

[0019] Figure 1A This is a schematic diagram illustrating an example operational pipeline according to some embodiments;

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

[0021] Figure 2 This is a graph showing the average operating power consumption of a radio unit according to some embodiments;

[0022] Figure 3 This is a related matrix diagram based on some embodiments;

[0023] Figure 4 This is an hourly PRB threshold map based on different bandwidths and regions in some embodiments;

[0024] Figure 5 It is a predicted hourly energy map of the policy and the trained policy under different weight values ​​according to some embodiments;

[0025] Figure 6 This includes three power consumption graphs for different bandwidths according to some embodiments;

[0026] Figure 7 Includes example graphs of two recommendation engines, which illustrate energy consumption-KPI trade-off curves according to some embodiments;

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

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

[0029] Figure 10 This is a block diagram of a communication system according to some embodiments;

[0030] Figure 11 This is a block diagram of network nodes according to some embodiments; and

[0031] Figure 12 This is a block diagram of a virtualized environment according to some embodiments. Detailed Implementation

[0032] Some 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, wherein examples of embodiments of the present disclosure are shown. However, the inventive concept can be embodied in many different forms and should not be construed as limiting 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 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 assumed by default to exist / be used in another embodiment.

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

[0034] In some approaches, such as modifying the configurability of network nodes, power consumption levels may change, which can cause problems when optimizing communication networks. Furthermore, in some approaches, bandwidth allocation within the communication network may be primarily static. Therefore, selecting the most advantageous configuration can be important for reducing energy consumption by the cell or the entire network. However, providing excessive bandwidth during periods of high or low traffic can lead to increased energy consumption. In adjustable configurations, bandwidth can be a significant factor influencing power consumption.

[0035] Other methods might involve manually adjusting cell bandwidth through a trial-and-error process, rather than, for example, self-organization. However, this approach may not achieve the lowest or acceptable power consumption and may not be able to adapt quickly to network conditions.

[0036] Other approaches that focus on the energy efficiency of LTE systems can use, for example, ML (Multi-Level Mechanism) to intelligently turn cells on / off or put cells into sleep mode. However, this operation can be designed to handle peak traffic periods (which may, for example, account for about 28% of network uptime) rather than balancing peak and off-peak traffic periods.

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

[0038] In some other approaches, when only historical data is extracted from real-world communication networks, ML models attempt to learn patterns by utilizing historical data. Therefore, ML models may be susceptible to the quality of the historical data, including, for example, the probability of some bandwidths being selected might be zero or very low, and / or the size of the data or subset of data might be too large or too small.

[0039] In contrast, some embodiments of this disclosure include dynamically configuring cell bandwidth to achieve acceptable network performance while reducing energy consumption. Therefore, energy savings can be achieved in communication networks based on dynamically allocated bandwidth (e.g., optimal bandwidth) without compromising network performance or achieving sufficient network performance.

[0040] Examples in this paper include ML models that formulate tasks as offline context bandits problems and implement offline learning and evaluation processes. While some examples in this paper are discussed in the context of LTE networks, this disclosure is not limited thereto. The operations described in this paper can be performed in other networks, such as 5G, 6G, etc.

[0041] Furthermore, some embodiments include operations based on two ML agents: a first ML agent, referred to herein as the "energy agent"; and a second ML agent, referred herein as the "teacher agent". In some embodiments, the energy agent and the teacher agent are separate, wherein the energy agent relates to energy efficiency, and the teacher agent relates to considering how much network performance might be degraded (if any) in the event of bandwidth changes.

[0042] In some examples, the energy agent and teacher agent correlate network operations and bandwidth allocation within the bandwidth recommendation engine, while providing uninterrupted network service.

[0043] In some other examples, the CSP (for example) can flexibly customize the desired percentage of energy savings by adjusting a hyperparameter that determines the relative weight of the two ML agents in the final bandwidth decision.

[0044] Other examples include offline evaluation results, which include evaluations of examples of ML models and demonstrations of energy-saving capabilities without impacting network performance.

[0045] Some examples include using an ML agent to optimize energy consumption in a communication network by dynamically allocating optimal bandwidth to each network node without impacting performance and key performance indicators (KPIs). KPIs may include, for example, latency, throughput, and / or energy efficiency. Furthermore, ML models can be tailored to achieve customized energy-saving capabilities, for example, by sorting radio unit types, bandwidth, and associated physical resource blocks (PRB) usage.

[0046] In addition, in some examples, the ML model may take into account the detected static power stability value (e.g., the power stability value of the radio unit of the BS in idle or sleep mode) as well as the dynamic characteristics of the BS's radio unit at higher PRB.

[0047] Some examples in this article also include: Each cell of the BS can be configured with a bandwidth at the cell level, where the recommendation engine recommends downlink (DL) bandwidth for each cell in the target communication network (e.g., an LTE network).

[0048] Orthogonal Frequency Division Multiple Access (OFDMA) divides the time-frequency domain into multiple orthogonal subcarriers (called resource blocks (RBs)) shared among multiple user equipments (UEs). Available RBs can be allocated to UEs based on their needs, while unallocated RBs can be disabled to minimize power consumption and reduce interference.

[0049] Power consumption at the BS can be categorized into two types: static power consumption (e.g., during idle or sleep modes) and dynamic power consumption. Static power consumption can primarily involve hardware components within the BS (e.g., radio units) and can be relatively constant in static mode. In contrast, dynamic power consumption (also referred to herein as communication power consumption) can vary based on the traffic load between the BS and the UE. Static power consumption is highly dependent on hardware components, and for newer generation radio units, power consumption still exists even if static power consumption is reduced.

[0050] Bandwidth determines the service capacity in a BS (Base Station), because bandwidth determines the total number of RBs (Royalties Base Stations) available to serve UEs. Therefore, bandwidth does not change and is fixed in some networks. For example, the channel bandwidth of an LTE network can be as follows:

[0051]

[0052] Therefore, in some examples, the following operation is included: this operation can minimize the dynamic power consumption of high RBs and the static power consumption of low RBs by allocating DL bandwidth (e.g., optimal DL bandwidth) to each cell in the BS.

[0053] As this article is about Figure 2 and Figure 6 Further discussion reveals that evaluations of examples according to some embodiments show that bandwidth variations with low RB can affect both static power consumption (which may not have been previously known) and dynamic power consumption.

[0054] In some examples, data is used to discover relationships between bandwidth, energy consumption, and KPIs of multiple network nodes, static power stability, and dynamic levels, and a recommendation engine is trained to recommend optimal bandwidth for both static and dynamic power consumption.

[0055] Based on the observed static stability values ​​and dynamic power consumption levels of the example radio units, technical advantages can include enhanced energy efficiency. For example, as further discussed herein, the evaluation of the examples shows that energy savings of approximately 14% were observed by the first recommendation engine without KPI degradation, and energy savings of approximately 6% were observed by the second recommendation engine.

[0056] In some examples, the computing device may recommend bandwidth for each cell hourly to save energy without affecting the quality of service (QoS). While some examples in this document are discussed in a non-limiting context of a computing device capable of recommending bandwidth hourly, this disclosure is not limited thereto, and bandwidth may be recommended for other time periods, such as multiple times within a 24-hour period.

[0057] Figure 1A This is a block diagram illustrating an example operational pipeline 100 according to some embodiments. Pipeline 100 includes operations related to site bandwidth, site energy consumption, and bandwidth recommendation. As shown, the example pipeline 100 includes:

[0058] - The first operation of data preparation 104 for historical data 102. Data preparation includes preprocessing 104a, feature selection 104b, and analysis and / or visualization 104c. Data preparation 104 outputs a training dataset. Historical data 102 can be stored, for example, in a communication system (e.g., Figure 10 In the nodes of the communication system 1000 shown or in cloud-based nodes.

[0059] - The second operation performs data augmentation 106 on the training dataset. Data augmentation 106 includes splitting the training data to create a balanced training dataset, where each output class (or target class) is represented by the same number of input samples. In the example, the output class could be bandwidth, and the input samples are cell states corresponding to their cell bandwidths. This could be done, for example, in nodes of a communication system (e.g., ...). Figure 10 Network node 1010) or Figure 10 Data enhancement 106 is performed in another node, including computing devices or cloud-based nodes (e.g., computing device 1018).

[0060] A third operation performed by a computing device including recommendation engine 108 determines a recommended bandwidth allocation 118 based on periodically received input cell states. As shown, recommendation engine 108 includes two ML agents: a first ML agent 112, which includes an energy-driven agent; and a second ML agent 110, which includes a teacher agent. Recommendation engine 108 can be included in a computing device, for example, as described herein regarding... Figure 9 Further discussion focuses on computing device 900. The computing device may be, for example, a node in a communication system (e.g., Figure 10 Network node 1010) or including Figure 10 Another node of the computing device in the communication system 1000 or a cloud-based computing device (e.g., Figure 10 (Computing device 1018 in the middle).

[0061] - The first ML agent 112 is trained within the offline context robber algorithm framework using augmented data from data augmentation 106, with the goal of selecting bandwidth that minimizes energy consumption. On the other hand, the second ML agent 110 is KPI-aware, and in this example, it always selects bandwidth based on insights gained from historical data 102 without compromising KPI performance. The first ML agent 112 and the second ML agent can be included in, for example... Figure 9 In the computing device shown.

[0062] - Furthermore, in some examples, the recommendation engine 108 determines static power stability values ​​(e.g., for low RB) and dynamic changes (e.g., for high RB), as combined in this paper. Figure 6 Further discussion is needed.

[0063] - The fourth operation includes policy evaluation 116, further discussed below, for testing and validating the recommendation engine 108 using a specified test set. This can be done, for example, at nodes in a communication system (e.g., Figure 10 Network node 1010) or Figure 10 Policy evaluation 116 is performed in another node, including computing devices or cloud-based nodes (e.g., computing device 1018).

[0064] Now let's discuss the operation of automatically selecting bandwidth (e.g., optimal bandwidth).

[0065] like Figure 1B As shown in the example, one example includes an LTE network with N BS 30a, 130b, and 130c, and each of the BS 130a, 130b, and 130c, sectors 120a, 120b, and 120c can be further subdivided into cells 122a-c, 124a-c, and 126a-c, respectively, which can also be... It means that, among them, BS All Communities in a community , It is BS The configuration, in which, Instructions BS residential area The bandwidth. In this example, the carrier frequency is fixed and not modified.

[0066] In this example, the cell configuration in the network can also be determined by... Indicates a specific time period. Inner Community The power consumption is determined by The network bandwidth configuration that minimizes energy consumption is given in form by the following formula:

[0067] (Formula 1)

[0068] in, This represents the power consumption of the entire network.

[0069] In this example, network QoS is guaranteed by meeting specified KPI requirements. KPIs describing QoS can include throughput, latency, interference, etc. Furthermore, the cell... Selected KPI set Designated as above the threshold The problem now becomes an optimization problem constrained by the following constraints in Equation 2:

[0070]

[0071] Subject to: (Formula 2)

[0072] In this example, if the total energy consumption under all possible bandwidth configurations is known, it can be determined by analyzing the sample set. Perform an exhaustive search to find the optimal solution.

[0073] Now let's discuss the first ML agent 112 further. In some examples, the first ML agent 112 can be described as an energy-driven agent 112. The energy-driven agent 112 can be derived from the training set after data augmentation 106. The learning process involves greedily selecting the bandwidth that minimizes energy consumption (including the static power stability of the bandwidth and the dynamic changes in the bandwidth) under the current cell state, such as regarding... Figure 6 Further discussion is needed.

[0074] In this example, the regressor The first ML agent 112 is trained to predict each possible action given a specific state. The associated reward. This reward is designed as a negatively linearly scaled energy in the range [-1, 0], which in this example can be subsequently converted back to hourly power consumption during evaluation. In this way, the regressor It could also be an hourly energy predictor. Therefore, the strategy for energy-driven agent 112 in this example... Defined as:

[0075] (Formula 3)

[0076] In this example, the strategy makes a deterministic choice, where the probability of choosing an action is either 1 or 0. In this example, a deep neural network is trained for regression analysis. However, if the network were trained as a regular regressor, highly accurate predictions might be difficult to achieve, and predictions could be biased, given limitations in data and model size.

[0077] To address this issue, in this example, a process implementing weighted squared loss can be used. See, for example, “Incorporating behavioral constraints in online AI systems,” Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33, No. 1, pp. 3–11, 18 (2019). This weighted loss considers the probabilities of conditional actions of the recorded policy and introduces a threshold. This constrains the estimation, thus providing an unbiased prediction of the new policy value. The loss is expressed as follows:

[0078] (Formula 4)

[0079] in, It is a recording strategy Select status The following action The probability of.

[0080] Referring to the offline reinforcement learning algorithm Batch Constrained Deep Q-Learning (BCQ), in this example, the generative model Training as a dataset The state-condition model, therefore .for The larger value, due to its robustness, is suitable for... The accuracy requirement is relatively low. The loss is further re-formulated as:

[0081] (Formula 5)

[0082] Now let's discuss the second ML agent 110 further. In some examples, the second ML agent 110 can be described as a teacher agent 110. Continuing with the example above, historical data 102 is extracted from the LTE network, where all bandwidth is manually set based on expert knowledge.

[0083] More specifically, the bandwidth is customized to handle traffic during peak hours. By utilizing this information, the operation includes: determining the minimum and maximum average hourly PRB values ​​supported by each bandwidth from historical data 102.

[0084] Therefore, in this example, the teacher agent 110 is KPI-aware and provides expert recommendations for bandwidth selection based on the hourly average PRB values ​​observed in a given state. Each bandwidth corresponds to a range of supported PRB values, and the teacher agent 110 evaluates which range the state falls into, identifies appropriate bandwidth (action) options, and assigns recommendation scores to all four available actions. Thus, the reward for the teacher agent 110 is defined as a vector... ,in, yes The score of the i-th action in the sequence.

[0085] Furthermore, in this example, the recommendation score is proportional to the recommendation level in the current state, meaning a higher score indicates a stronger recommendation for the corresponding action. Additionally, in the final policy... Introducing custom weights It is given as:

[0086] (Formula 6)

[0087] in, It is a user-configurable weight that balances the energy-driven agent 112 and the KPI-aware agent 110.

[0088] In some examples, The value is a user-configurable weight that balances the energy-driven agent 112 and the KPI-aware agent 110.

[0089] Furthermore, in some examples, by adjusting The recommendation engine 108 can be customized for a wide range of scenarios, from purely energy-saving methods to more conservative KPI-driven approaches. This flexibility allows for exploration in changing... The value represents a trade-off between the percentage of energy savings and KPI degradation.

[0090] Now let's discuss strategy evaluation 116 further. In this example, three criteria are checked during strategy evaluation:

[0091] 1. How much energy can be saved by using the recommendation engine 108?

[0092] 2. Does energy conservation affect KPIs? If so, to what extent will it cause KPI degradation?

[0093] 3. How to balance the percentage of energy savings with the deterioration of KPIs?

[0094] Regarding the first criterion, as previously discussed in this example, training a reward (e.g., energy) predictor. and probability estimator These can be used again for strategy evaluation 116.

[0095] Furthermore, in this example, direct method (DM) and dual robust (DR) estimation are applied, which are two common offline evaluation processes used in the context robber algorithm to predict the average reward for n rounds of recommendations under this policy.

[0096] By scaling the predicted reward back to the average energy consumption, it can be compared to the original average energy consumption of the network over n rounds, thus calculating the percentage energy saving. Given a trained policy, making a deterministic decision (Equation 6), the DM estimator can be simplified to the following Equation 7:

[0097] (Formula 7)

[0098] in, It is an indicator function that evaluates to 1 if its parameter is true, and 0 otherwise.

[0099] Similarly, the DR estimator can be rewritten as the following formula 8:

[0100] (Formula 8)

[0101] For the second and third standards, the PRB value range for each bandwidth discussed earlier is used to assess whether the selected bandwidth can support the current traffic volume status. This allows for the calculation of the ratio of incorrect selections, which can serve as an indicator of the degree of KPI degradation. When the value is adjusted from 1 to 0, the final strategy can be observed. By prioritizing energy efficiency, a trade-off between energy savings percentage and KPI degradation can be observed.

[0102] The determination of the recommended bandwidth will now be discussed further.

[0103] As previously discussed, a cell’s energy consumption is directly affected by its serving radio units and traffic loads (e.g., low and high PRB utilization). Figure 2 This relationship was studied. Figure 2 This is a graph showing the operating average power consumption of four radio units, where "radio_x" is the radio unit type, and the number following the radio unit type (e.g., "31" after "radio_a") is its encoded number (e.g., rank). Figure 2 As shown, the relationship between the average operating energy consumption and the hourly average DL PRB indicates that energy consumption is positively correlated with the PRB value.

[0104] Continuing this example, during data preparation 104, 13 features other than bandwidth and power consumption are selected to construct the cell state, including DL PRB utilization, energy configuration parameters, radio unit type, location, and number of antennas. For the example process, the continuous features in this state are further normalized, and... Figure 3 The correlation matrix is ​​shown below.

[0105] In this example, given the features and correlation matrix, determine the maximum average hourly utilization (PRB) supported by each bandwidth, such as... Figure 2 As shown. These values ​​were then implemented as PRB thresholds to assess the correctness of bandwidth selection while maintaining KPI and QoS.

[0106] Furthermore, in this example, an analysis was conducted to investigate the impact of both bandwidth and radio unit type on power consumption, correlated with static power stability at low RB and dynamic variations at high RB. This was achieved through comparison. Figure 4 The energy consumption of cells with the same radio unit type but different bandwidths in areas R (R1 to R4) shown can determine which bandwidth option tends to lead to higher energy consumption. This determination can help identify recommended bandwidths that meet energy-saving targets.

[0107] The results and energy-saving performance evaluation of this example are presented. This paper tests the recommendation engine 108 using the unbiased reward set and biased reward set of the teacher agent 110 (referred to as Engine 1 and Engine 2 in this paper, respectively). For Engine 1 with unbiased teacher rewards, Engine 1 has no preference for certain bandwidths, and the embedded teacher agent 110 assigns the same score to supportive bandwidths. For the biased engine 2, Engine 2 tends to select the minimum bandwidth capable of handling the current traffic to achieve bandwidth efficiency.

[0108] Use the same Values ​​were compared to evaluate the performance of engines 1 and 2. During the experiment, the results regarding fuel efficiency and the impact of KPIs were compared with... The values ​​do not have a linear relationship, therefore six values ​​from different intervals are selected. value. Figure 5 It shows that in different Box plots of rewards for all test samples under various strategies corresponding to the values.

[0109] Embedded reward (energy per hour) returner The accuracy was also evaluated by having the recommendation engine 108 select the same actions as those in the test data, thus simulating the recording strategy. In this way, the reward for recommendation engine 108 becomes... The predicted value of the reward. By comparing this predicted value with the actual value, the regressor can be evaluated. The accuracy, such as Figure 6 As shown.

[0110] Table 1 below summarizes the corresponding relative energy values ​​and energy saving rates in this assessment:

[0111]

[0112] Table 2 below shows the energy-KPI trade-offs for each strategy in this evaluation:

[0113]

[0114] As shown in Tables 1 and 2, it can be observed that Engine 1 with unbiased teacher agent 110 results in lower KPI degradation and a higher energy saving percentage. However, with As the value increases from 0.1 to 1, the energy saving percentage of engine 2 ranges from 6% to 12%, while the energy saving percentage of engine 1 ranges from 14% to 15%.

[0115] This instruction: In Over a wide range of values, Engine 1 maintains relatively consistent fuel efficiency, which could mean that customizing Engine 1 for fuel efficiency in this scenario might be more challenging. Furthermore, both engine types exhibit a non-linear relationship between these two standards.

[0116] To better understand the nature of the trade-off, the assessment examined smaller values ​​between 0 and 0.3. The values ​​are shown in Table 3 below:

[0117]

[0118] To account for slight fluctuations in results obtained from different test data, 5000 rounds of repeated testing were conducted on Engine 1 and Engine 2, and the results were averaged. In each round, a predefined set of... The value is used to evaluate a randomly selected subset of test data. Through trial and error, a specific set of designs is developed for each of Engine 1 and Engine 2. The value effectively captures the trade-off curve.

[0119] Analysis of the trade-off curves for this example shows that Engine 1 achieved an optimal energy saving percentage of approximately 14% without KPI degradation, while Engine 2 achieved approximately 6% energy saving.

[0120] Table 3 shows the combined numerical results of the tradeoff curves, revealing a clear nonlinear relationship. With... The value decreased and the effect of energy-driven agent 112 was enhanced. Figure 7 In the curve, the points along the x-axis become more densely packed. This indicates that when energy consumption is already relatively low, even a slight increase in the percentage of energy savings will lead to a significant deterioration in KPIs. Furthermore, Engine 1, which utilizes the rewards of unbiased teacher agent 110, shows a smaller degree of KPI deterioration. The value is particularly sensitive.

[0121] In another example, the indexes in Table 3 can be used to enable the process to determine and account for the extent of degradation effects (e.g., different levels) and whether bandwidth changes are recommended. For example, some conditions of degradation (e.g., latency, throughput, coverage, etc.) may be acceptable for some uses to allow for greater energy savings during bandwidth changes.

[0122] The operation of computing devices can be controlled by Figure 9 The computing device 900 is used to execute the commands. Reference is now made to some embodiments of this disclosure. Figure 8 Flowchart discussion using Figure 9 The structure enables the operation of computing devices. Figure 8Operations 800, 802, 806 to 810 and 814 in the flowchart may be optional for some embodiments of the computing device and related methods. For example, the module may be stored in... Figure 9 The memory 905, the first ML agent 112 and / or the second ML agent 110 are contained therein, and these modules can provide instructions such that when the instructions of the modules are executed by the corresponding computing device processor 903 (also referred to herein as processing circuitry), the computing device 900 performs... Figure 8 The flowchart describes the corresponding operations. Although computing device 900 is shown as including a first ML agent 112 and / or a second ML agent, the embodiments herein are not limited thereto. In other embodiments, the first ML agent 112 and / or the second ML agent 110 may be in a cloud-based node (e.g., a server) or a distributed network (e.g., different cloud-based nodes) outside of computing device 900.

[0123] refer to Figure 8 In some embodiments, a computer-implemented method for bandwidth allocation and energy saving in a communication network, performed by a computing device, is provided. The method includes: receiving (804) data including cell states of cells in the communication network. The method further includes determining (812) a bandwidth allocation for a cell based on (i) a first ML agent and (ii) a second ML agent, the first ML agent identifying candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understanding the cell's KPIs and providing a recommended bandwidth selection based on PRB values ​​observed for the cell state.

[0124] In some embodiments, it is determined (812) that a specified weight applied to a second ML agent is used to determine bandwidth allocation.

[0125] In some embodiments, specifying weights balances the trade-off between the first and second ML agents by adjusting their weights in the final bandwidth recommendation. Higher weights indicate greater trust and reliance on a particular agent.

[0126] In some embodiments, the method further includes repeating the receiving (804) and determining (812) at multiple times within a time period (e.g., a 24-hour time period) achieved by adjustable working and inference frequencies.

[0127] The received data may include augmented data, which consists of multiple datasets for different bandwidths obtained through counterfactual data augmentation. Counterfactual data augmentation is the process of adding hypothetical samples to an imbalanced dataset for data categories with insufficient sample sizes. This balances the dataset to avoid or reduce biased training results for the ML agent. Counterfactual data augmentation can be applied to multiple datasets. The corresponding datasets may include the cell's state within a time period, the bandwidth of the cell's state within the time period, and the reward representing the energy consumption of the cell under that state and bandwidth.

[0128] Multiple datasets can be obtained from configuration management data for the first period and performance management data for the second period for multiple base stations in a communication network within a certain time period.

[0129] Configuration management data may include one or more of the following configuration management parameters for each cell of each of multiple base stations: (i) downlink bandwidth, (ii) radio unit type, (iii) power consumption parameters, and (iv) number of antennas.

[0130] Performance management data may include one or more of the following: the status of the communication network, including traffic information represented as the number of PRBs in the uplink and downlink of each cell of each base station in multiple base stations; and the power consumption of each base station radio unit in each cell.

[0131] 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 the energy consumption of the cell; and obtaining (802) multiple datasets.

[0132] In some embodiments, the first ML agent is trained using augmented data within the framework of the offline context robber algorithm to select the bandwidth that minimizes energy consumption for the cell state. The cell state includes the static energy level of the bandwidth and the variation of the bandwidth energy level.

[0133] Furthermore, in some embodiments, the second ML agent learns from historical data multiple observed PRB values ​​over a time period supported by the corresponding bandwidth in the configured bandwidth to handle traffic during peak periods.

[0134] The corresponding bandwidth can correspond to the corresponding range of supported PRB values.

[0135] In some embodiments, the second ML agent (i) determines which range the cell state falls into, (ii) identifies candidate bandwidth options from that range, and (iii) assigns a corresponding recommendation score to the candidate bandwidth options.

[0136] In some embodiments, the corresponding recommendation score has a value within a first range.

[0137] Furthermore, in some embodiments, a higher recommendation score within this range indicates a stronger recommendation for the corresponding candidate bandwidth option.

[0138] Determining (812) bandwidth allocation may include: allocating multiple bandwidths to multiple cells in the communication network and minimizing the total energy consumption of multiple cells in the communication network.

[0139] In some embodiments, for multiple bandwidth allocations that minimize total energy consumption, the quality of service of the communication network satisfies multiple KPIs of the multiple cells.

[0140] In some embodiments, the specified weight is adjustable within a second value range between a first end and a second end of the second range. In some embodiments, (i) the first end of the range causes the bandwidth allocation of the cell to maximize energy savings for the cell state, and (ii) the second end of the range causes the bandwidth allocation of the cell to maximize the cell's KPIs.

[0141] In some embodiments, the method further includes evaluating (810) bandwidth allocation based on one or more of the following: (i) the amount of energy saved; (ii) whether the KPI is affected by the energy saved, and if so, the amount of decrease in the KPI; and (iii) assessing the trade-off between the energy saved and the degradation of the KPI.

[0142] In some embodiments, the energy consumption of a cell is associated with the serving radio elements of the cell and the PRB utilization rate of the cell, and the method further includes: sorting multiple serving radio elements of each cell in a plurality of cells based on the corresponding energy consumption given the same PRB value (806).

[0143] Furthermore, in some embodiments, determining multiple corresponding thresholds for each KPI among multiple KPIs (808) is based on the maximum average utilized PRB supported by the corresponding bandwidth within the period.

[0144] like Figure 9 As shown, computing device 900 includes a processor 903 operatively coupled to a memory 905, a network interface 907, a first ML agent 112, a second ML agent 110, and / or any other component or any combination thereof. Some computing devices may utilize... Figure 9 The diagram shows all components or subsets of components. The degree of integration between components can vary depending on the computing device. Furthermore, some computing devices can contain multiple instances of a component, such as multiple processors, memory, etc.

[0145] Processor 903 is configured to process instructions and data and can be configured to implement any sequential state machine operable to execute instructions, which are stored as a machine-readable computer program in memory 905, a first ML agent 112, and / or a second ML agent 110. Processor 903 can be implemented as: one or more hardware-implemented state machines (e.g., implemented with discrete logic, a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), etc.); programmable logic together with appropriate firmware; one or more stored computer programs, a general-purpose processor (e.g., a microprocessor or a digital signal processor (DSP)) together with appropriate software; or any combination of the foregoing. For example, processor 903 may include multiple central processing units (CPUs).

[0146] In this example, network interface 907 can be configured to provide one or more interfaces to input devices, output devices, or one or more input and / or output devices. Examples of output devices include displays, monitors, printers, other output devices, or any combination thereof. Input devices can allow a user to capture information into computing device 900. Examples of input devices include touch-sensitive or presence-sensitive displays, cameras (e.g., digital cameras, digital camcorders, webcams, etc.), microphones, sensors, mice, trackballs, directional keyboards, touchpads, scroll wheels, smart cards, etc. Presence-sensitive displays may include capacitive or resistive touch sensors to sense input from the user. Sensors may be force sensors, optical sensors, proximity sensors, biometric sensors, etc., or any combination thereof. Output devices can use the same type of interface port as input devices. For example, a Universal Serial Bus (USB) port can be used to provide both input and output devices.

[0147] Memory 905, the first ML agent 112, and / or the 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), disk, optical disk, hard disk, removable magnetic tape, flash drive, etc. For example, memory 905, the first ML agent 112, and / or the second ML agent 110 may include one or more applications, such as an operating system, web browser application, widgets, utility engines, or other applications, and corresponding data. Memory 905, the first ML agent 112, and / or the second ML agent 110 may store any one or a combination of various operating systems used by computing device 900.

[0148] Memory 905, the first ML agent 112, and / or the second ML agent 110 can be configured to include multiple physical drive units, such as a redundant array of independent disks (RAID), flash memory, a USB flash drive, an external hard drive, a thumb drive, a pen drive, a key drive, a high-density digital versatile optical disc (HD-DVD) drive, an internal hard drive, a Blu-ray disc drive, a holographic digital data storage (HDDS) disc drive, an external mini dual in-line memory module (DIMM), synchronous dynamic random access memory (SDRAM), external micro DIMM SDRAM, smart card memory (e.g., a tamper-proof module in the form of a universal integrated circuit card (UICC) including one or more subscriber identification modules (SIMs), such as USIM and / or ISIM), other memory, or any combination thereof. The UICC can be, for example, an embedded UICC (eUICC), an integrated UICC (iUICC), or a removable UICC commonly referred to as a "SIM card". The memory 905, the first ML agent 112, and / or the second ML agent 110 may allow the computing device 900 to access instructions, applications, etc., stored on a transient or non-transient memory medium to unload or upload data. An article of art (e.g., an article of art utilizing a communication system) may be tangibly embodied in, or embodied in, the memory 905, the first ML agent 112, and / or the second ML agent 110, which may be or include a device-readable storage medium.

[0149] Processor 903 can be configured to communicate with a network using network interface 907. Network interface 907 may include one or more communication subsystems. Network interface 907 may include one or more transceivers for communication (e.g., communication with one or more remote transceivers capable of wireless communication, such as another computing device or local computing device, edge node, cloud node, etc.). Each transceiver may include a transmitter and / or receiver suitable for providing network communication (e.g., optical, electrical, etc.).

[0150] In the illustrated embodiment, the communication functions of network interface 907 may include cellular communication, Wi-Fi communication, LPWAN communication, data communication, voice communication, multimedia communication, short-range communication such as Bluetooth, near-field communication, location-based communication (e.g., using a Global Positioning System (GPS) to determine location), another type of communication function, or any combination thereof. Communication may be implemented according to one or more communication protocols and / or standards (e.g., IEEE 802.11, Code Division Multiple 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 Network (SONET), Asynchronous Transfer Mode (ATM), QUIC, Hypertext Transfer Protocol (HTTP), etc.).

[0151] Functionality implemented by some embodiments can be virtualized. In this context, virtualization means creating a virtual version of a device or computing device that may include a virtualized hardware platform, storage devices, and networking resources. As used herein, virtualization can be applied to any device or component thereof described herein and involves implementing at least a portion of the functionality as one or more virtual components. Some or all of the functionality described herein can be implemented as virtual components executed by one or more virtual machines (VMs) in one or more virtual environments hosted by one or more hardware nodes (e.g., hardware computing devices operating as edge nodes or cloud nodes). Furthermore, in embodiments, virtual nodes can be fully virtualized.

[0152] Applications (which may alternatively be referred to as software instances, virtual devices, network functions, virtual nodes, virtual network functions, etc.) can run in a virtualized environment to implement some of the features, functions, and / or benefits of some of the embodiments disclosed herein.

[0153] While the computing devices described herein may include combinations of the hardware components shown, other embodiments may include computing devices with different combinations of components. It should be understood that these computing devices may include any suitable combination of hardware and / or software required to perform the tasks, features, functions, and methods disclosed herein. The determination, calculation, acquisition, or similar operations described herein may be performed by processing circuitry that processes information in ways such as: converting acquired information into other information, comparing the acquired or converted information with information stored in the computing device, and / or performing one or more actions based on the acquired or converted information, and making determinations based on the results of said processing. Furthermore, although components are depicted as single boxes located within larger boxes or nested within multiple boxes, in practice, a computing device may include multiple different physical components constituting a single illustrated component, and functionality may be partitioned between individual components. For example, a communication interface may be configured to include any of the components described herein, and / or the functionality of a component may be partitioned between a processor and a network interface. In another example, the non-computationally intensive functions of any such component may be implemented in software or firmware, and the computationally intensive functions may be implemented in hardware.

[0154] In some embodiments, some or all of the functions described herein may be provided by processing circuitry that executes instructions stored in memory, which in some 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 functions may be provided by the processing circuitry, for example, in a hard-wired manner, without executing instructions stored on a separate or discrete device-readable storage medium. In any of these particular embodiments, the processing circuitry may be configured to perform the described functions regardless of whether instructions stored on a non-transitory computer-readable storage medium are executed. The benefits provided by such functions are not limited to the individual processing circuitry or other components of the computing device, but are enjoyed holistically by the computing device and / or generally by the end user and wireless network.

[0155] In some 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 executed by the at least one processor to perform operations. These operations include performing some or all of the functions described herein.

[0156] In some embodiments, the computer program includes program code to be executed by a processor (903, 11302) of a computing device (900, 1018, 1010), which includes 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. These operations include performing some or all of the functions described herein.

[0157] In some embodiments, a computer program product includes a non-transitory storage medium (905, 11304) comprising 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. These operations include performing some or all of the functions described herein.

[0158] Figure 10 An example of a communication system 1000 according to some embodiments is shown.

[0159] In this example, communication system 1000 (also referred to herein as a communication network) includes telecommunications network 1002, which includes access network 1004 (e.g., RAN) and core network 1006, which includes one or more core network nodes 1008. Access network 1004 includes one or more access network nodes, such as network nodes 1010a and 1010b (one or more of which may generally be referred to as network node 1010), or any other similar 3GPP access node or non-3GPP access point. Network node 1010 facilitates direct or indirect connections of user equipment (UE) (also referred to herein as “user equipment”), for example, connecting UE 1012a, UE 1012b, UE 1012c, and UE 1012d (one or more of which may generally be referred to as UE 1012) to core network 1006 via one or more wireless connections. The computing device (e.g., computing device 900) can be a cloud-based computing device 1018, a network node 1010, or another node that includes a computing device in a communication system 1000.

[0160] Examples of wireless communication via wireless connection include transmitting and / or receiving wireless signals using electromagnetic waves, radio waves, infrared waves, and / or other types of signals suitable for transmitting information without using wiring, cables, or other conductors. Furthermore, in various embodiments, communication system 1000 may include any number of wired or wireless networks, network nodes, UEs, and / or any other components or systems that can facilitate or participate in the communication of data and / or signals (whether via wired or wireless connections). Communication system 1000 may include any type of communication, telecommunications, data, cellular, radio network, and / or other similar system, and / or interface with any type of communication, telecommunications, data, cellular, radio network, and / or other similar system.

[0161] UE 1012 can be any of a variety of communication devices, including wireless devices that are arranged, configured, and / or operable to communicate wirelessly with network node 1010 and other communication devices. Similarly, network node 1010 is arranged, capable, configured, and / or operable to communicate directly or indirectly with UE 1012 and / or with other network nodes or devices in telecommunication network 1002 to achieve and / or provide network access (e.g., wireless network access) and / or to perform other functions in telecommunication network 1002 (e.g., management).

[0162] In the depicted example, core network 1006 connects network node 1010 to one or more hosts (such as host 1016). These connections can be direct connections or indirect connections via one or more intermediate networks or devices. In other examples, network nodes may be directly coupled to hosts. Core network 1006 includes one or more core network nodes (e.g., core network node 1008) that are formed together with hardware and software components. The characteristics of these components may be substantially similar to those described with respect to UEs, network nodes, and / or hosts, such that the description is generally applicable to the corresponding components of core network node 1008. Example core network nodes include the functions of one or more of the following: 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 Unhiding Function (SIDF), Unified Data Management (UDM), Secure Edge Protection Agent (SEPP), Network Open Function (NEF), and / or User Plane Function (UPF).

[0163] Host 1016 may be owned or under the control of a service provider other than the operator or provider of access network 1004 and / or telecommunications network 1002, and may be operated by or on behalf of the service provider. Host 1016 may host a variety of applications to provide one or more services. Examples of such applications include real-time and pre-recorded audio / video content, data collection services (e.g., retrieving and compiling data about various environmental conditions detected by multiple UEs), analytics functions, social media, functions for controlling or otherwise interacting with remote devices, functions for alarms and monitoring centers, or any other such functions performed by a server.

[0164] As a whole, Figure 10 The communication system 1000 enables connections between the UE, network nodes, and hosts. In this sense, the communication system can be configured to operate according to predefined rules or procedures, such as specific standards, including but 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 standard (WiFi); and / or any other suitable wireless communication standards, such as Global Microwave Access Interoperability (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.

[0165] In some examples, telecommunications network 1002 is a cellular network implementing 3GPP standardized features. Therefore, telecommunications network 1002 can support network slicing to provide different logical networks to different devices connected to it. For example, telecommunications network 1002 can provide URLLC services to some UEs while providing eMBB services to other UEs, and / or mMTC / massive IoT services to yet another UE.

[0166] In some examples, UE 1012 is configured to send and / or receive information without direct human interaction. For example, the UE may be designed to send information to access network 1004 according to a predetermined schedule when triggered by internal or external events or in response to a request from access network 1004. Additionally, the UE may be configured to operate in single-RAT mode, multi-RAT mode, or multi-standard mode. For example, the UE may operate using any or a combination of Wi-Fi, NR, and LTE, i.e., configured for multiple radio dual connectivity (MR-DC), such as E-UTRAN (Evolved UMTS Terrestrial Radio Access Network) New Radio Dual Connectivity (EN-DC).

[0167] In this example, hub 1014 communicates with 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, hub 1014 may be a controller, router, content source and analyzer, or any other communication device described herein relating to the UE. For example, hub 1014 may be a broadband router that enables the UE to access core network 1006. As another example, hub 1014 may be a controller that sends commands or instructions to one or more actuators of the UE. Commands or instructions may be received from the UE, network node 1010, or via executable code, scripts, procedures, or other instructions in hub 1014. As another example, hub 1014 may be a data collector that acts as a temporary storage device for UE data, and in some embodiments, may perform data analysis or other processing. As another example, hub 1014 may be a content source. For example, for a UE acting as a VR headset, display, speaker, or other media delivery device, hub 1014 can retrieve VR assets, video, audio, or other media or data related to perceived information via a network node, and then provide them directly to the UE after performing local processing and / or adding additional local content. In yet another example, hub 1014 acts as a proxy server or orchestrator for the UE, particularly if one or more of the UEs are low-power IoT devices.

[0168] Hub 1014 may have a persistent / persistent or intermittent connection to network node 1010b. Hub 1014 may also allow different communication schemes and / or scheduling between hub 1014 and UEs (e.g., UEs 1012c and / or 1012d) and between hub 1014 and core network 1006. In other examples, hub 1014 is connected to core network 1006 and / or one or more UEs via a wired connection. Furthermore, hub 1014 may be configured to connect to an M2M service provider via access network 1004 and / or to another UE via a direct connection. In some scenarios, a UE may establish a wireless connection with network node 1010 while still being connected via hub 1014 via a wired or wireless connection. In some embodiments, hub 1014 may be a dedicated hub—that is, a hub whose primary function is to route communication from network node 1010b to UE / to network node 1010b. In other embodiments, the hub 1014 may be a non-dedicated hub—that is, a device capable of operating to route communication between the UE and network node 1010b, but additionally capable of operating as a communication start point and / or endpoint for certain data channels.

[0169] Figure 11 A network node 11300 according to some embodiments is illustrated. As used herein, a network node refers to a device that is capable of, configured, arranged, and / or operable to communicate directly or indirectly with a UE and / or with other network nodes, computing devices, or devices in a telecommunications 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, NodeBs, evolved NodeBs (eNBs), and NR NodeBs (gNBs)), computing devices, O-RAN nodes, or components of O-RAN nodes (e.g., O-RUs, O-DUs, O-CUs).

[0170] Base stations can be classified based on the coverage they provide (or, in other words, their transmission power levels); therefore, depending on the coverage provided, a base station can be called a femtobase, picobase, microbase, or macrobase. A base station can be a relay node or a relay donor for control relays. Network nodes can also include one or more (or all) portions of a distributed radio base station, such as centralized digital units, distributed units (e.g., in O-RAN access nodes), and / or remote radio units (RRUs), sometimes referred to as remote radio headends (RRHs). These remote radio units can be integrated with antennas to form an antenna-integrated radio, or they can be independent of antenna integration. A portion of a distributed radio base station can also be referred to as a node in a distributed antenna system (DAS).

[0171] Other examples of network nodes include multi-transmitter point (multi-TRP) 5G access nodes, multi-standard radio (MSR) devices (e.g., MSR BS), network controllers (e.g., radio network controllers (RNC) or base station controllers (BSC)), base transceiver stations (BTS), transmitter points, transmitter nodes, multi-cell / multicast coordination entities (MCE), operations and maintenance (O&M) nodes, operations support system (OSS) nodes, self-organizing network (SON) nodes, location nodes (e.g., evolved Serving Mobility Location Center (E-SMLC)) and / or minimized drive test (MDT).

[0172] Network node 11300 includes processing circuitry 11302, memory 11304, communication interface 11306, and power supply 11308. The memory 11304 includes a first ML agent 112 and a second ML agent 110. Network node 11300 may consist of multiple physically separate components (e.g., Node B components and RNC components, BTS components and BSC components, etc.), each with its own corresponding components. In some scenarios where network node 11300 includes multiple separate components (e.g., BTS and BSC components), one or more separate components may be shared among several network nodes. For example, a single RNC can control multiple NodeBs. In such scenarios, each unique “NodeB and RNC pair” can be considered a single, separate network node in some cases. In some embodiments, network node 11300 may be configured to support multiple Radio Access Technologies (RATs). In this embodiment, some components can be replicated (e.g., separate memory 11304 exists for different RATs) and some components can be reused (e.g., the same antenna 11310 can be shared by different RATs). Network node 11300 may also include multiple sets of various illustrated components integrated into network node 11300 for different wireless technologies (e.g., GSM, WCDMA, LTE, NR, WiFi, Zigbee, Z-wave, LoRaWAN, RFID, or Bluetooth wireless technologies). These wireless technologies can be integrated into the same or different chips or chipsets and other components within network node 11300.

[0173] The processing circuitry 11302 may include one or more of the following: a microprocessor, a controller, a central processing unit, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array, or any other suitable computing device, resource, or combination of hardware, software and / or coding logic, operable to provide network node 11300 functionality, either alone or in combination with other network node 11300 components (e.g., memory 11304).

[0174] 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 a radio frequency (RF) transceiver circuitry 11312 and a baseband processing circuitry 11314. In some embodiments, the RF transceiver circuitry 11312 and the baseband processing circuitry 11314 may be on separate chips (or chipsets), boards, or units (e.g., radio units and digital units). In alternative embodiments, some or all of the RF transceiver circuitry 11312 and the baseband processing circuitry 11314 may be on the same chip or chipset, board, or unit group.

[0175] The memory 11304, including the first ML agent 112 and the second ML agent 110, may include any form of volatile or non-volatile computer-readable memory, including but not limited to permanent storage devices, solid-state memory, remotely mounted memory, magnetic media, optical media, random access memory (RAM), read-only memory (ROM), mass storage media (e.g., hard disk), removable storage media (e.g., flash drives, optical discs (CDs), or digital video discs (DVDs)), and / or any other volatile or non-volatile, non-transitory device-readable and / or computer-executable memory device that stores information, data, and / or instructions that can be used by the processing circuitry 11302. The memory 11304, including the first ML agent 112 and the second ML agent 110, may store any suitable instructions, data, or information, including computer processes, software, applications including logic, rules, codes, tables, and / or other instructions that can be executed by the processing circuitry 11302 and used by the network node 11300. The memory 11304, including the first ML agent 112 and the second ML agent 110, can be used to store any calculations performed by the processing circuitry 11302 and / or any data received via the communication interface 11306. In some embodiments, the processing circuitry 11302 and the memory 11304, including the first ML agent 112 and the second ML agent 110, are integrated together.

[0176] Communication interface 11306 is used for wired or wireless communication of signaling and / or data between network nodes, access networks, and / or UEs. As shown, communication interface 11306 includes a port / terminal 11316 for transmitting and receiving data to and from the network, for example, via a wired connection. Communication interface 11306 also includes radio front-end circuitry 11318, which may be coupled to antenna 11310, or in some embodiments to a portion of antenna 11310. Radio front-end circuitry 11318 includes a filter 11320 and an amplifier 11322. Radio front-end circuitry 11318 may be connected to antenna 11310 and processing circuitry 11302. Radio front-end circuitry 11318 may be configured to modulate the signal transmitted between antenna 11310 and processing circuitry 11302. Radio front-end circuitry 11318 may receive digital data to be transmitted to other network nodes or UEs via a wireless connection. The radio front-end circuit 11318 can use a combination of filter 11320 and / or amplifier 11322 to convert digital data into radio signals with appropriate channel and bandwidth parameters. The radio signals can then be transmitted via antenna 11310. Similarly, when data is received, antenna 11310 can collect radio signals, which are then converted into digital data by the radio front-end circuit 11318. The digital data can then be passed to processing circuitry 11302. In other embodiments, the communication interface may include different components and / or different combinations of components.

[0177] In some alternative embodiments, network node 11300 does not include a separate radio front-end circuitry 11318; instead, processing circuitry 11302 includes radio front-end circuitry and is connected to antenna 11310. Similarly, in some embodiments, all or some of the RF transceiver circuitry 11312 is part of communication interface 11306. In yet another embodiment, communication interface 11306 includes one or more ports or terminals 11316, radio front-end circuitry 11318, and RF transceiver circuitry 11312 as part of a radio unit (not shown), and communication interface 11306 communicates with baseband processing circuitry 11314, which is part of a digital unit (not shown).

[0178] Antenna 11310 may include one or more antennas or antenna arrays configured to transmit and / or receive wireless signals. Antenna 11310 may be coupled to radio front-end circuitry 11318 and may be any type of antenna capable of wirelessly transmitting and receiving data and / or signals. In some embodiments, antenna 11310 is decoupled from network node 11300 and may be connected to network node 11300 via an interface or port.

[0179] Antenna 11310, communication interface 11306, and / or processing circuitry 11302 can be configured to perform any receive operation and / or certain acquire operation described herein by a network node. Any information, data, and / or signals can be received from the UE, another network node, and / or any other network device. Similarly, antenna 11310, communication interface 11306, and / or processing circuitry 11302 can be configured to perform any transmit operation described herein by a network node. Any information, data, and / or signals can be transmitted to the UE, another network node, and / or any other network device.

[0180] Power supply 11308 provides power to the various components of network node 11300 in a form suitable for the various components (e.g., at the voltage and current levels required by each respective component). Power supply 11308 may also include or be coupled to power management circuitry to supply power to the components of network node 11300 for performing the functions described herein. For example, network node 11300 may be connected to an external power source (e.g., mains, power outlet) via input circuitry or an interface (e.g., cable), thereby supplying power to the power circuitry of power supply 11308. As another example, power supply 11308 may include a power source in the form of a battery or battery pack, which is connected to or integrated into the power circuitry. The battery can provide backup power if the external power source fails.

[0181] Embodiments of network node 11300 may include more than Figure 11 Additional components shown are provided to offer certain aspects of the functionality of the network node, including any of the functions described herein and / or any functionality required to support the topics described herein. For example, network node 11300 may include a user interface device to allow information to be input into and output from network node 11300. This allows users to perform diagnostic, maintenance, repair, and other management functions on network node 11300.

[0182] Figure 12This diagram illustrates a virtualization environment 12500 in which functionality implemented by some embodiments can be virtualized. In this context, virtualization means creating a virtual version of an apparatus or device that may include a virtualization hardware platform, storage devices, and network resources. As used herein, virtualization can be applied to any device or component thereof described herein, and relates to implementations where at least a portion of functionality is implemented as one or more virtual components. Some or all of the functionality described herein can be implemented as virtual components executed by one or more VMs in one or more virtual environments 12500 hosted by one or more hardware nodes, such as hardware network nodes operating as computing devices, network nodes, UEs, core network nodes, or hosts. Furthermore, in embodiments where virtual nodes do not require radio connectivity (e.g., core network nodes or hosts), the nodes can be fully virtualized.

[0183] Application 12502 (which may alternatively be referred to as a software instance, virtual device, network function, virtual node, virtual network function, etc.) runs in virtualization environment 12500 to implement some of the features, functions and / or benefits of some embodiments disclosed herein.

[0184] Hardware 12504 includes processing circuitry, memory storing software and / or instructions executable by the hardware processing circuitry, and / or other hardware devices described herein (such as network interfaces, input / output interfaces, etc.). The software can be executed by the processing circuitry to instantiate one or more virtualization layers 12506 (also referred to as a hypervisor or virtual machine monitor (VMM)), provide VMs 12508a and 12508b (one or more of which may generally be referred to as VM 12508) and / or perform any functions, features, and / or benefits described in relation to some embodiments described herein. Virtualization layer 12506 can present a virtual operating platform to VM 12508, which appears as network hardware.

[0185] VM 12508 includes virtual processing, virtual memory, virtual network or interface, and virtual storage, and can be operated by a corresponding virtualization layer 12506. Different embodiments of instances of virtual device 12502 can be implemented on one or more VMs 12508, and these implementations can be carried out in different ways. In some contexts, hardware virtualization is referred to as Network Functions Virtualization (NFV). NFV can be used to unify numerous network device types into industry-standard high-capacity server hardware, physical switches, and physical storage devices that can reside in data centers and customer premises equipment (CPE).

[0186] In the context of NFV, VM 12508 can be a software implementation of a physical machine, whose operating procedures are executed as if on a physical, non-virtualized machine. Each VM 12508, along with the hardware portion of hardware 12504 that executes that VM (whether it is dedicated hardware for that VM and / or hardware shared by that VM with other VMs), forms a separate virtual network element. Still within the context of NFV, the virtual network function is responsible for handling the specific network functions operating within one or more VMs 12508 on top of hardware 12504 and corresponding to application 12502.

[0187] Hardware 12504 can be implemented in a standalone network node with general or specific components. Hardware 12504 can implement some functions via virtualization. Alternatively, hardware 12504 can be part of a larger hardware cluster (e.g., in a data center or CPE) where many hardware nodes work together and are managed by management and orchestration 12510, which in particular oversees the lifecycle management of application 12502. In some embodiments, hardware 12504 is coupled to one or more radio units, each radio unit including one or more transmitters and one or more receivers that can be coupled to one or more antennas. The radio units can communicate directly with other hardware nodes via one or more suitable network interfaces and can be used in conjunction with virtual components to provide radio capabilities to virtual nodes (e.g., radio access nodes, base stations, or computing devices). In some embodiments, some signaling can be provided by using a control system 12512, which can alternatively be used for communication between hardware nodes and radio units.

[0188] Further definitions and examples are discussed below.

[0189] In the above description of certain embodiments of this disclosure, it should be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. Unless otherwise defined, all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the concept of this disclosure pertains. It will be further understood that terms (e.g., terms as defined in a general dictionary) should be interpreted as having a meaning consistent with their meaning in the context of this specification and related art, and not as having an ideal or overly literal meaning, unless so expressly defined herein.

[0190] When an element is referred to as “connected to,” “coupled to,” “responsive to,” or a variation thereof, it may be directly connected to, coupled to, or responsive to the other element, or there may be intermediate elements present. Conversely, when an element is referred to as “directly connected to,” “directly coupled to,” “directly responsive to,” or a variation thereof, there are no intermediate elements present. Throughout the document, similar reference numerals are used to denote similar elements. Furthermore, the terms “coupled,” “connected,” “responsive,” or variations thereof as used herein may include wireless coupling, connection, or responsiveness. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein are intended to also include the plural forms. For the sake of brevity and / or clarity, well-known functions or structures may not be described in detail. The term “and / or” (abbreviated as “ / ”) includes any and all combinations of one or more of the associated listed items.

[0191] 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 used only to distinguish one element / operation from another. Thus, a first element / operation in some embodiments may be referred to as a second element / operation in other embodiments without departing from the teachings of this disclosure. Throughout this specification, the same reference numerals denote the same or similar elements.

[0192] As used herein, the terms “including,” “comprising,” “containing,” “covering,” “constituting,” “included,” “having,” “possessing,” “having,” or variations thereof are open-ended and include one or more of the stated features, integers, elements, steps, components, or functions, but do not preclude the presence or addition of one or more other features, integers, elements, steps, components, functions, or combinations thereof. Furthermore, as used herein, the common abbreviation “eg (for example)” derives from the Latin phrase “exempligratia,” which can be used to introduce or specify a general example of a previously mentioned item and is not intended as a limitation of that item. The common abbreviation “ie (i.e.)” derives from the Latin phrase “idest,” and can be used to specify a specific item in a more general reference.

[0193] This document describes exemplary embodiments with reference to block diagrams and / or flowcharts illustrating computer-implemented methods, apparatus (systems and / or devices), and / or computer program products. It should be understood that blocks in the block diagrams and / or flowcharts, as well as combinations of blocks in the block diagrams and / or flowcharts, can be implemented by computer program instructions executed by one or more computer circuits. These computer program instructions can be provided to processor circuitry of general-purpose computer circuitry, special-purpose computer circuitry, and / or other programmable data processing circuitry to produce a machine, such that instructions executed by the processor of a computer and / or other programmable data processing apparatus translate and control transistors, values ​​stored in memory locations, and other hardware components within such circuitry to implement the functions / actions specified in the block diagrams and / or flowcharts, thereby creating means (functional bodies) and / or structures for implementing the functions / actions specified in the block diagrams and / or flowcharts.

[0194] These computer program instructions may also be stored in a tangible computer-readable medium capable of directing a computer or other programmable data processing apparatus to function in a specific manner, causing the instructions stored in the computer-readable medium to produce an article of writing, which includes instructions for implementing functions / actions specified in the blocks of block diagrams and / or flowcharts. Therefore, embodiments of this disclosure can be implemented by hardware and / or software (including firmware, resident software, microcode, etc.) running on a processor (e.g., a digital signal processor), which may be collectively referred to as a "circuit," a "module," or variations thereof.

[0195] It should also be noted that in some alternative implementations, the functions / actions marked in the boxes may not occur in the order indicated in the flowchart. For example, depending on the functions / actions involved, two boxes shown consecutively may actually be executed substantially simultaneously, or the boxes may sometimes be executed in reverse order. Furthermore, the functionality of a given box in a flowchart and / or block diagram may be divided into multiple boxes, and / or the functionality of two or more boxes in a flowchart and / or block diagram may be at least partially integrated. Finally, without departing from the scope of this disclosure, other boxes may be added / inserted between the shown boxes, and / or boxes / actions may be omitted. Additionally, although some boxes include arrows indicating the main direction of communication regarding the communication path, it should be understood that communication may occur in the opposite direction to the indicated arrows.

[0196] Many variations and modifications can be made to the embodiments without substantially departing from the principles of this disclosure. All such variations and modifications are intended to be included within the scope of this disclosure herein. Therefore, the foregoing subject matter should be understood as exemplary rather than restrictive, and the examples of embodiments are intended to cover all such modifications, improvements, and other embodiments falling within the spirit and scope of this disclosure. Thus, to the fullest extent permitted by law, the scope of this disclosure should be determined by the widest permissible interpretation of this disclosure, including examples of embodiments and their equivalents, and should not be limited to or restricted to the specific implementations described above.

Claims

1. A computer-implemented method executed by a computing device, the method comprising: Receive (804) data, the data including the cell status of a cell in the communication network; as well as The bandwidth allocation of the cell is determined based on (i) a first machine learning ML agent and (ii) a second ML agent, the first ML agent identifying candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understanding the key performance indicators (KPIs) of the cell and providing recommended bandwidth selection based on the physical resource block (PRB) values ​​observed for the cell state.

2. The computer implementation method according to claim 1, wherein, The determination (812) uses a specified weight applied to the second ML agent to determine the bandwidth allocation.

3. The computer implementation method according to claim 2, wherein, The specified weights balance the trade-off between the recommended bandwidth selections of the first ML agent and the second ML agent.

4. The computer-implemented method according to any one of claims 1 to 3, further comprising: The receiving (804) and determining (812) are repeated at multiple points within a time period.

5. The computer-implemented method according to any one of claims 1 to 4, wherein, The data includes augmented data, which comprises multiple datasets for multiple bandwidths obtained through counterfactual data augmentation.

6. The computer implementation method according to claim 5, wherein, The counterfactual data augmentation is applied to the plurality of datasets, and each dataset includes the state of the cell over a period of time, the bandwidth of the cell in the state over the period of time, and a reward representing the energy consumption of the cell in relation to the state and the bandwidth.

7. The computer implementation method according to claim 6, wherein, The multiple datasets are obtained from configuration management data for a first period and performance management data for a second period from multiple base stations in the communication network over a time period.

8. The computer implementation method according to claim 7, wherein, The configuration management data includes one or more of the following configuration parameters for each cell of each of the plurality of base stations: (i) downlink bandwidth, (ii) radio unit type, (iii) power consumption parameters, and (iv) number of antennas.

9. The computer-implemented method according to any one of claims 7 to 8, wherein, The performance management data includes one or more of the following: the status of the communication network, including traffic volume information represented as the number of PRBs in the uplink and downlink of each cell of each of the plurality of base stations; And the energy consumption of each base station radio unit in each cell.

10. The computer-implemented method according to any one of claims 7 to 9, further comprising: Process (800) historical data to associate at least one configuration management parameter of the cell with the cell status to determine the energy consumption of the cell; as well as Obtain the multiple datasets described in (802).

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

12. The computer-implemented method according to any one of claims 1 to 11, wherein, The second ML agent learns from historical data multiple observed PRB values ​​within a time period supported by the corresponding bandwidth in the configured bandwidth to handle traffic during peak periods.

13. The computer implementation method according to claim 12, wherein, The corresponding bandwidth corresponds to the corresponding range of supported PRB values.

14. The computer implementation method according to claim 13, wherein, The second ML agent (i) determines which range the cell state falls into, (ii) identifies candidate bandwidth options from the range, and (iii) assigns a corresponding recommendation score to the candidate bandwidth option.

15. The computer implementation method according to claim 14, wherein, The corresponding recommended score has a value within a first range.

16. The computer implementation method according to claim 15, wherein, A higher recommendation score within the range indicates a stronger recommendation for the corresponding candidate bandwidth option.

17. The computer-implemented method according to any one of claims 1 to 16, wherein, Determining (812) the bandwidth allocation includes: multiple bandwidth allocations for multiple cells in the communication network that minimize the total energy consumption of the multiple cells in the communication network.

18. The computer implementation method according to claim 17, wherein, For the plurality of bandwidth allocations that minimize the total energy consumption, the quality of service of the communication network satisfies a plurality of KPIs for the plurality of cells.

19. The computer-implemented method according to any one of claims 2 to 18, wherein, The specified weight can be adjusted within a second value range between the first and second ends of the second range.

20. The computer implementation method according to claim 19, wherein, (i) the first end of the range causes the bandwidth allocation of the cell to maximize energy saving for the cell state, and (ii) the second end of the range causes the bandwidth allocation to maximize the KPI of the cell.

21. The computer-implemented method according to any one of claims 1 to 20, further comprising: The bandwidth allocation described in (810) is evaluated based on one or more of the following: (i) the amount of energy saved; (ii) Whether the KPI is affected by the energy saved, and if so, the amount of degradation of the KPI; (iii) Assess the trade-off between the amount of energy saved and the degradation of the KPI.

22. The computer-implemented method according to any one of claims 17 to 21, wherein, The energy consumption of the cell is related to the serving radio unit of the cell and the PRB utilization rate of the cell. The method further includes: Based on the corresponding energy consumption under the same PRB value, the multiple serving radio elements of each cell in the plurality of cells are sorted (806).

23. The computer-implemented method according to any one of claims 17 to 22, further comprising: The multiple thresholds for each KPI in (808) are determined based on the maximum average utilized PRB supported by the corresponding bandwidth within the period.

24. A computing device (900, 1018, 1010), comprising: The first machine learning ML agent (112) and the second ML agent (110) in the communication network (1000); At least one processor (903, 11302); At least one memory (905, 11304) is connected to the at least one processor (903, 11302) and stores program code, which is executed by the at least one processor to perform operations including: Received data, the data including the cell status of cells in the communication network; and The bandwidth allocation for the cell is determined based on (i) a first ML agent and (ii) a second ML agent, wherein the first ML agent identifies candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understands the cell's key performance indicators (KPIs) and provides recommended bandwidth selection based on the physical resource block (PRB) values ​​observed for the cell state.

25. The computing device according to claim 24, wherein, The at least one memory (905, 11304) is connected to the at least one processor (903, 11302) and stores program code, which is executed by the at least one processor to perform the operation 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), the computing device (900, 1018, 1010) including a first machine learning (ML) agent (112) and a second ML agent (110) in a communication network (1000), wherein execution of the program code causes the computing device to perform operations, the operations including: Receive data, the data including the cell status of the cells in the communication network; as well as The bandwidth allocation for the cell is determined based on (i) a first ML agent and (ii) a second ML agent, wherein the first ML agent identifies candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understands the cell's key performance indicators (KPIs) and provides recommended bandwidth selection based on the physical resource block (PRB) values ​​observed for the cell state.

27. The computer program according to claim 26, wherein, The execution of the program code causes the computing device (900, 1018, 1010) to perform the operation according to any one of claims 2 to 23.

28. A computer program product comprising a non-transitory storage medium (905, 11304), said 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), said computing device (900, 1018, 1010) including a first machine learning ML agent (112) and a second ML agent (110) in a communication network (1000), wherein execution of said program code causes the computing device to perform operations including: Receive data, the data including the cell status of the cells in the communication network; as well as The bandwidth allocation for the cell is determined based on (i) a first ML agent and (ii) a second ML agent, wherein the first ML agent identifies candidate bandwidths for the cell that minimize energy consumption for the cell state, and the second ML agent understands the cell's key performance indicators (KPIs) and provides recommended bandwidth selection based on the physical resource block (PRB) values ​​observed for the cell state.

29. The computer program product according to claim 28, wherein, The execution of the program code causes the computing device to perform the operation according to any one of claims 2 to 23.

Images