WO2021244730A1

WO2021244730A1 - Measurement reporting and configuration in communication networks

Info

Publication number: WO2021244730A1
Application number: PCT/EP2020/065127
Authority: WO
Inventors: Jalil TAGHIA; Johan Eker; Andreas Johnsson; Robert MARKLUND; Hannes LARSSON
Original assignee: Telefonaktiebolaget Lm Ericsson (Publ)
Priority date: 2020-06-01
Filing date: 2020-06-01
Publication date: 2021-12-09

Abstract

A method performed by an apparatus for a communication network, for configuring metrics to be monitored in the communication network. The method comprises: receiving a measurement policy comprising an indication of one or more mandatory metrics; accessing a training dataset comprising data for a plurality of possible metrics from the communication network; and selecting, based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored, wherein the plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics, and wherein the one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics.

Description

C MEASUREMENT REPORTING AND CONFIGURATION IN COMMUNICATION

NETWORKS

Technical field Embodiments of the present disclosure relate to communication networks, and particularly provide methods, apparatus and machine-readable mediums for configuring measurement reporting in communication networks.

Background Communication networks are becoming increasingly complex. The architecture of modern wireless telecommunication networks, such as those standardized by the 3^rd Generation Partnership Project (3GPP), may comprise radio access networks, core networks, backhaul networks, transport networks, edge and cloud computing environments, applications and more. It therefore becomes increasingly complex to manage such networks to ensure that faults are identified and fixed, and that performance is acceptable. For example, telecommunication operators typically deliver services under strict service level agreements (SLAs), defining the level of service by reference to one or more performance metrics which are to be met. Measurements and monitoring functions operate within and throughout the telecom infrastructure to measure and report system performance, and thus enable data-driven management, whereby key performance indicators are calculated and tracked, faults are identified, and network and services proactively managed. So-called “zero-touch management” is one approach to network management in which the network becomes self-managing.

Summary

Extensive measurements and data collection are usually required for data-driven management. Typically measurement data is reported to servers via an agent, and then transferred to a database of measurement data which may be centralized or distributed within the network. However, the data collection process is associated with significant overheads, both in the time taken to acquire the data and the network resources utilized in reporting the data to a management function or network node. The overhead associated with measurements, data collection, and data transfer may even adversely affect network performance itself and potentially co-located services as well. For example, a radio access network node such as a Next Generation NodeB (gNB) has the capability of measuring a huge number of different metrics, monitoring both its own performance and the performance of connected devices. However, measuring all of these possible metrics will consume significant processing resources and lead to degraded services for those connected devices. Further, transferring the measurement data from the radio access network node to the management function also utilizes bandwidth in the backhaul network and thus adversely affects throughput of other data.

In some cases, metrics may be measured and reported within the network for reasons other than network management. In this case, the metrics can be passively sampled rather than actively measured, and thus no additional resources are used to measure or report such metrics for the purposes of network management. However, the processing of such metrics for network management does consume resources. Many metrics may be correlated, such that there is no or limited additional information from processing multiple correlated metrics. Thus the use of correlated metrics for network management, even if passively sampled, wastes resources that may be utilized more efficiently within the network.

A paper by Li et al (“Feature Selection: A Data Perspective”, ACM Computing Surveys. December 2017) contains an overview of recent advances in the field of feature selection, whereby data (especially high-dimensional data) is prepared for various data- mining and machine-learning problems.

Embodiments of the disclosure propose an approach for understanding and selecting the key telecom infrastructure metrics (or data features) to monitor, in order to reduce measurement overhead while still upholding requirements stemming from service-level agreements, data-driven applications, regulatory requirements, and policies from applications.

According to a first aspect of the disclosure, there is provided a method performed by an apparatus for a communication network, for configuring metrics to be monitored in the communication network. The method comprises: receiving a measurement policy comprising an indication of one or more mandatory metrics; accessing a training dataset comprising data for a plurality of possible metrics from the communication network. The method further comprises selecting, based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored, wherein the plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics, and wherein the one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics.

Further aspects provide apparatus for performing the method set out above. For example, in one aspect there is provided an apparatus for configuring metrics to be monitored in a communication network. The apparatus comprises processing circuitry and a non-transitory machine-readable medium storing instructions which, when executed by the processing circuitry, cause the apparatus to: receive a measurement policy comprising an indication of one or more mandatory metrics; access a training dataset comprising data for a plurality of possible metrics from the communication network; and select, based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored, wherein the plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics, and wherein the one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics.

A further aspect provides a non-transitory machine-readable medium. The non-transitory machine-readable medium stores instructions. The instructions, when executed by processing circuitry of an apparatus, cause the apparatus to: receive a measurement policy comprising an indication of one or more mandatory metrics; access a training dataset comprising data for a plurality of possible metrics from the communication network. The instructions, when executed by processing circuitry of an apparatus, further cause the apparatus to select, based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored, wherein the plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics, and wherein the one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics.

According to embodiments of the disclosure, therefore, one or more additional metrics are selected to be monitored in the communication network based on their statistical interaction with the one or more mandatory metrics to be monitored. In this way, the selection of metrics to be monitored takes into account one or more mandatory metrics, and thus embodiments of the disclosure have the technical advantage of enabling a reduction in or minimization of information loss when require to monitor one or more metrics.

Brief description of the drawings

For a better understanding of examples of the present disclosure, and to show more clearly how the examples may be carried into effect, reference will now be made, byway of example only, to the following drawings in which:

Figure 1 shows a system according to embodiments of the disclosure;

Figure 2 shows an apparatus according to embodiments of the disclosure;

Figure 3 is a flowchart of a method according to embodiments of the disclosure;

Figure 4 shows a neural network according to embodiments of the disclosure;

Figure 5 is a graph comparing the performance of the neural network shown in Figure 4 to other methods; and

Figures 6 and 7 are schematic diagrams of an apparatus according to embodiments of the disclosure.

Detailed description

Figure 1 shows a system 100 according to embodiments of the disclosure. The system 100 comprises a management node 102, a network architecture 104 and a client node 116.

The network architecture 104 comprises one or more network nodes and/or network functions which may be defined in hardware, software or a combination of hardware and software. In one embodiment, the network architecture 104 relates to a wireless communication network, e.g., a cellular network such as those standardized by the 3^rd Generation Partnership Project (3GPP). Thus in the illustrated embodiment, the network architecture 104 comprises a radio access network 114, which itself may comprise a plurality of radio access network nodes, such as base stations (e.g., evolved NodeBs (eNBs), next generation NodeBs (gNBs), etc) and wireless devices or user equipments (UEs). The network architecture may comprise or be defined within a virtual computing environment. For example, a core network such as the 5^th generation core (5GC) or the Evolved Packet Core (EPC) may be defined within such a virtual computing environment. Therefore in the illustrated embodiment, the network architecture comprises a virtual orchestration system 106 such as Kubernetes, a cloud computing platform 108 such as OpenStack, one or more operating systems 110 and one or more data centres 112.

One or more nodes or network functions within the network architecture 104 may be configured to monitor one or more metrics relating to operation of the network. In the context of the current disclosure, a metric is any parameter which can be measured in the network. A node may monitor a metric through active measurement (e.g., requesting another node to measure and report values for the metric) or through passive measurement (e.g., measuring the values for the metric directly). Those skilled in the art will appreciate that the range of metrics which could possibly be monitored is vast. For example, some of the possible metrics may relate to data throughput, such as an amount of network traffic flowing through a particular network node, or the latency of such network traffic, etc. The data may be categorized (e.g., according to a service the data relates to, according to a source or destination of the data, according to a quality of service associated with the data, etc) such that the amount of data within a particular category is itself a metric. In another example, the metrics may relate to the operational status of the network nodes, such as the transmit power, the total power consumption, number of active connections, etc.

Metrics may be distinguished by the node which measures them. That is, the same type of data measured in different nodes may be considered as different metrics for the purposes of the present disclosure. For example, the aggregate traffic throughput for a network node A may be considered a different metric to the aggregate traffic throughput for a network node B. Metrics may be further distinguished by the frequency with which the metrics are measured. For example, a first quantity measured at a first frequency may be considered a different metric to the same first quantity measured at a second, different frequency; similarly, average quantities may be considered different metrics to instantaneous quantities.

In the system of 100, the management node 102 is operable to configure the monitoring of metrics within the network architecture 104, and particularly to select the metrics which should be monitored so as to reduce the amount of metrics that might otherwise be monitored (i.e., so that not all possible metrics are monitored). One objective of the disclosure is to select metrics to be monitored while reducing or minimizing the amount of information which is lost by that selection. That is, as noted above, one objective of embodiments of the present disclosure is to reduce the number of monitored metrics so as to reduce the amount of network resources required to monitor performance of the network. However, some information may be lost by that selection. For example, the management node 102 may select metrics to be monitored so that redundant metrics are not selected and thus information loss is reduced or minimized. In this context, a metric may be considered redundant if it comprises information which is contained within, or which can be derived from, at least one other metric. For example, a metric may be entirely redundant if its information is comprised entirely within, or which can be entirely derived from, at least one other metric. Similarly, a metric may be considered partially redundant if its information is comprised partially within, or which can be partially derived from, at least one other metric. Further information on the selection process is set out below.

The management node 102 may be established within the network architecture 104 itself, or outside the network (as illustrated). For example, in the former case the management node 102 may be implemented within a standardized network node or network function, such as an operations support system (OSS) agent implemented as a user plane function (UPF).

It may be mandatory to monitor one or more metrics. For example, service-level agreements (SLAs) may specify a level of performance in terms of one or more metrics, which should therefore be monitored within the network architecture to check for compliance with the SLA. In another example, a network operator may wish to monitor certain metrics, e.g., to check for likely faults in the network architecture. A regulator may require one or more metrics to be monitored to check for compliance with regulations. For example, regulators set maximum transmit powers for radio access nodes, typically measured as an average over a short time window.

In the illustrated embodiment, these mandatory metrics, also known as a measurement policy, are provided to the management node 102 by the client node 116. The mandatory metrics may additionally or alternatively be provided directly as an input to the management node 102 (e.g., by human input).

According to embodiments of the disclosure, the management node 102 is operable to select a plurality of metrics to be measured, including the one or more mandatory metrics dictated by the measurement policy and one or more additional metrics. The additional metrics are selected based on statistical interaction with the mandatory metrics, so that information loss is reduced and redundant metrics (e.g., those metrics with redundant information compared to the mandatory metrics and other additional metrics) are not selected.

The selection of metrics may further be based on a training dataset comprising data samples for a plurality of possible metrics. The plurality of possible metrics may comprise all metrics which it is possible to measure in the network architecture 104, or only a subset of those metrics. Thus the training dataset may comprise values for the plurality of possible metrics at a plurality of time instances. The training dataset may be obtained from the network architecture 104, e.g., through a data collection agent.

Figure 1 shows an embodiment in which a management apparatus 102 configures other nodes of the network architecture 104 to monitor selected metrics. In alternative embodiments, the selection of metrics to be monitored may be performed by the monitoring node itself. Figure 2 shows an apparatus 200 according to these embodiments. The apparatus 200 may comprise any network node. For example, the apparatus may be implemented in a radio access network node, e.g., a base station such as a gNB, eNB etc; in a core network node or function, e.g., access and mobility management function (AMF), session management function (SMF), user plane function (UPF), etc; or in a wireless device, e.g., a UE.

The apparatus comprises a selector module 202 and a reporting or processing module 206. In the illustrated embodiment, the apparatus also comprises a database 204 of training data; however, in alternative embodiments the database 204 may be external to the apparatus 200 and merely accessible by it.

The selector module 202 receives one or more mandatory metrics, for example, from a client node or through human input, and interacts with the database 204 to select a plurality of metrics to be monitored substantially as described above. In the embodiment of Figure 2, however, rather than configuring another node to monitor the metrics, the selector module 202 provides a control output to the reporting or processing module 206. The control output comprises an indication of the selected metrics to be monitored, such that the reporting or processing module 206 is enabled to select those metrics from the database 204. The selected metrics may then be reported to another network node (e.g., to a management node or any node monitoring the performance of the network), or used for processing within the apparatus 200. In the latter case, the selected metrics may be used as inputs to a machine-learning model, for example, to select an action to be performed by the apparatus 200.

Figure 3 is a flowchart of a method for configuring metrics to be monitored according to embodiments of the disclosure. The method may be performed by the management node 102 or the apparatus 200 described above, or the apparatuses 600 and 700 described below.

The apparatus performing the method is implemented in a communication network. In one embodiment, the communication network comprises a wireless communication network, e.g., a cellular network such as those standardized by 3GPP. In other embodiments, the network may be a wired network such as a public switched telephone network or a computer network. The communication network has a plurality of metrics which it is possible to monitor. For further discussion of the metrics, see the description above with respect to Figure 1. A goal of the method is to select a subset of those metrics to be monitored, so as to utilize network resources more efficiently while reducing or minimizing information loss as a result of the selection.

The method begins in step 300, in which the apparatus receives a measurement policy comprising an indication of one or more metrics which it is mandatory to monitor. The measurement policy may be input by a human operator, or received from another node of the network (e.g., a client node or server) for example. Optionally, the measurement policy may also include an indication of the frequency of measurements, or other metadata surrounding the monitoring of the mandatory metrics.

In step 302, the apparatus accesses a training dataset comprising data samples for a plurality of possible metrics (including the mandatory metrics). The plurality of possible metrics may comprise all possible metrics measurable in the network, or a subset of those metrics. The dataset may be stored within the apparatus, particularly for embodiments in which the apparatus is the monitoring node itself (see Figure 2), or external to the apparatus and accessible by it. For example, the dataset may be collected by a data-collection agent in communication with the network. The dataset may be stored in any suitable database, such as TimescaleDB, Prometheus, etc. The database may be periodically updated with new data to enable re-calculation of selected metrics in step 304 below. jn step 304, the apparatus selects a plurality of metrics to be monitored, based on the training dataset and the one or more mandatory metrics. The selected plurality of metrics to be monitored is a subset of the plurality of possible metrics (i.e., those metrics for which data is available in the dataset), and comprises the one or more mandatory metrics specified in step 300 and one or more additional metrics of the plurality of possible metrics in the dataset. The one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics. That is, the selection determines the additional metrics to be monitored, given the knowledge that the mandatory metrics must be monitored. In one embodiment, the additional metrics are selected so as to reduce or minimize information loss when monitoring the selected plurality of metrics instead of the plurality of possible metrics. For example, metrics which comprise redundant information already present in the mandatory metrics may not be selected as additional metrics.

In one embodiment, step 304 comprises providing the one or more mandatory metrics as statically defined values in a machine-learning model, and then training the machinelearning model based on the training dataset. For example, the machine-learning model may comprise a neural network, and the mandatory metrics provided as statically defined nodes in that neural network.

Figure 4 shows a neural network 400 according to embodiments of the disclosure, comprising a single-layer selector network 402 and a decoder network 404. The selector network 402 is a single-layer neural network with a number of nodes V₁ ... Vj related to the mandatory metrics (where J is the number of mandatory metrics), and a predetermined number of nodes U₁ ... U_K related to the features that should be learned from data in a close interaction with the induced J policies (where K is the number of additional metrics to be learned). K may be set by the designer of the neural network 400 so as to achieve a desired level of complexity.

The input training data X₁ ...X_D comprises time-series data for D possible metrics. The nodes representing the mandatory metrics are represented by discrete one-hot variables U₁ ... U_J and the nodes representing the K induced features are represented by CONCRETE random variables V₁ ... V_K. A one-hot variable is a vector in which one value is equal to 1 and all other values are equal to 0. Thus each one-hot variable selects a single one of the metrics and represents one mandatory metric. A CONCRETE random variable is a function that acts as a hybrid between continuous and discrete values; the term “CONCRETE” is a portmanteau of “continuous” and “discrete”. The CONCRETE random variable has a distribution defined by a temperature parameter τ ∈ (0,∞) and a scale parameter

, where

denotes the D-dimensional target feature space on the real continuous space

. At high values of the temperature parameter, the values of the CONCRETE variable vary continuously; as the temperature parameter approaches zero, the CONCRETE variable converges towards a discrete one-hot variable. As will be described in more detail below, the temperature variable is thus reduced while the neural network is trained, such that each CONCRETE variable V₁ ... V_K selects one additional metric from the input data.

The goal of the decoder network 404 is to reconstruct the input data X₁ 1..X_D from the output of the selector layer 402. In orderto achieve this goal, a loss function is generated based on the difference between the input data X₁ ... X_D and the reconstructed data output by the decoder network 404. The calculated value of the loss function is

used to update the scale parameter α and the weights of the decoder network 404 and the process is repeated until a stopping criterion is reached. For example, the stopping criterion may comprise the value of the loss function between consecutive iterations falling below a threshold, or a maximum number of iterations being performed.

The architecture of the decoder network 404 can be set by the user based on the complexity and availability of data, and is not defined further herein. Figure 4 shows a single-layer architecture, but there may be any number of layers with any architecture. There are no strict requirements on the decoder architecture.

In more detail, let

of size

be a subset of the target feature space which includes the mandatory features chosen according to the measurement policy. The term “feature” is used interchangeably herein with “metric”. The goal of the neural network 400 is to find the subset

of size

and to learn a reconstruction function

such that the expected loss

between X in the target space and the reconstructed

is minimized, that is: where p(X ) is the data distribution. The variable θ

represents the weights of the decoder network 404. The output of the decoder

is equal to

of course. In practice the true data distribution p(X ) is unknown. Given N independent measurements collected in (the N measurements may relate to data

collected at N different time instances, for example) and use of the central limit theorem, instead we seek to minimize the empirical loss

The network 400 is first initialized, and then repeatedly optimized based on the loss function until a stopping criterion is reached. At that point, the metrics to be monitored are selected based on the contents of the selector layer 402.

Step 1 : Initialization

1 .a. Set

of size

given the must-have policies.

1 .b. Set K which is the size of the subset

.

- Choose the architecture of the decoder network

which serves as the reconstruction function.

- Construct U

according to .

Define the initial temperature parameter τ₀ > 0 and a scheduling strategy for reducing τ during training.

- Choose a learning rate λ.

Step 2: Optimization

- Construct and initialize the CONCRETE variable V by taking random samples from a CONCRETE distribution with the initial temperature parameter τ₀ > 0 and α₀.

- Construct according to:

. This is

the output of the selector network 402. Here ʘ refers to the inner product.

- Attempt to reconstruct the original data X from the output of the selector network using the decoder network 404. The output of the decoder network 404 is

·

- Compute the gradient of the loss between the reconstructed data and

the original data X with respect to θ and with respect to α, that is

Update the decoder parameter θ according to:

, and the scale parameter of the CONCRETE distribution according to

.

Step 3: Repeat Step 2 until the stopping criterion is reached, e.g., convergence of the loss, a maximum number of iterations has been performed, etc. The input training data may be provided repeatedly as an input to step 2. Each repetition of the training data is referred to as an “epoch”.

The temperature parameter t is updated according to the choice of scheduling strategy in step 1 , and is gradually reduced throughout the training. For example, the temperature parameter may be reduced such that at the last epoch (e.g., the last planned iteration through the training data), the temperature approaches zero. The temperature parameter may be reduced per epoch, for example, with a linear step size or in a nonlinear fashion (e.g., using exponential decay or cosine annealing).

The selector layer 402 now defines the metrics which are to be selected for monitoring. The mandatory policies U₁ ... U_J are statically defined one-hot variables and thus have not changed. The temperature parameter has been reduced such that the CONCRETE random variables V₁ ... V_K also approach discrete one-hot variables. Each CONCRETE random variable thus points to a single metric in the dataset which should be selected as an additional metric to be monitored. The values of the CONCRETE random variables may be compared to a threshold (which will typically be defined very close to 1) to determine whether the particular metric should be selected.

After step 304, the apparatus may perform different actions. In one embodiment, particularly where the apparatus is implemented in a node which measures the metrics itself (e.g., as shown above in Figure 2), the method proceeds to step 306 in which the apparatus monitors the plurality of metrics output from step 304. The monitored metrics may then be processed locally or reported to another node as described above with respect to Figure 2.

In another embodiment, particularly where the apparatus is implemented in a management node (such as the management node 102), the method proceeds to step 308 in which the apparatus configures one or more nodes of the network to monitor the selected metrics output by step 304. For example, the apparatus may transmit one or more configuration messages to one or more nodes comprising indications of the metrics to be measured. Optionally, the configuration messages may comprise indications of the frequency with which the metrics should be measured.

Thus Figure 3 describes a method by which metrics can be selected to be monitored in a communication network.

Figure 5 is a graph showing the performance of the neural network 400 against other methods. The performance was tested using the following experimental set-up:

Dataset: The data X consists of 500 feature attributes (or metrics) and 19380 samples. The data is collected from a small testbed (described in Yanggratoke, Rerngvit, et al. "Predicting service metrics for cluster-based services using real-time analytics." 2015 11th International Conference on Network and Service Management (CNSM). IEEE, 2015), and the data X corresponds to system dynamics captured using Systems Activity Report (e.g. CPU utilization, memory utilization, etc). On this DC we execute a key-value store database. Corresponding to each sample x there is a response vector y which measures the time to read from the database (mean read time, 95^th percentile, and 99^th percentile). Data samples are divided into a train set and a test set. The test set consists of 5814 samples and the train set consists of 13566 samples.

Mandatory features: For illustration of the method, two attributes out of 500 feature attributes are selected as the mandatory features based on domain knowledge, namely: 'X1_cpu21_.idle' and ‘X0_kbcommit'.

Experiment design: The objective is to evaluate the effectiveness of the selected features (metrics) in minimizing information loss and reducing utilization of network resources. For this illustration, a standard multilayer perceptron (MLP) neural network is used as a regressor in the four scenarios listed below.

(i) Full feature space. The entire feature space is fed directly into an MLP regressor.

(ii) Feature selection considering the mandatory features. A subset of features is selected as described above with respect to Figures 3 and 4, taking into consideration the mandatory features. The selected feature set is fed into the MLP regressor.

(iii) Feature selection without considering mandatory policies. A subset of features is selected using a neural network as shown in Figure 4, but without including any mandatory features in the selector layer 402. The neural network uses the same initialization as that for scenario (ii). The selected feature set is fed into the MLP regressor.

(iv) Representation learning using principal component analysis (PCA). The feature sub-space containing the principal components is fed into the MLP regressor.

The regression performance is evaluated in terms of mean absolute error between the true responses and their predictions. It can be seen from Figure 5 that scenario (ii) achieves similar performance to scenario (iii) while allowing for incorporation of domain knowledge and other factors in the selection of the mandatory features. In fact, scenario (ii) achieves the best performance overall when 18 features are selected. Further, scenarios (ii) and (iii) reduce overheads for monitoring the metrics by 96%, as compared to scenarios (i) and (iv), which monitor all 500 metrics. Scenario (i) achieves the poorest performance in terms of mean absolute error, possibly because the large feature may lead to poor machine learning training.

Figure 6 is a schematic diagram of an apparatus 600 according to embodiments of the disclosure. The apparatus 600 may perform the method described above with respect to Figure 3. The apparatus 600 may correspond to or be implemented within the management node 102 or the apparatus 200 described above.

The apparatus 600 comprises processing circuitry 602 (such as one or more processors, digital signal processors, general purpose processing units, etc), a computer-readable medium (e.g., memory such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, etc) 604 and one or more interfaces 606. The components are illustrated coupled together in series; however, those skilled in the art will appreciate that the components may be coupled together in any suitable manner (e.g., via a system bus or suchlike).

According to embodiments of the disclosure, the computer-readable medium 604 stores instructions which, when executed by the processing circuitry 602, cause the apparatus 600 to: receive a measurement policy comprising an indication of one or more mandatory metrics; access a training dataset comprising data for a plurality of possible metrics from the communication network; and select, based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored. The plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics. The one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics.

In further embodiments of the disclosure, the apparatus 600 may comprise power circuitry (not illustrated). The power circuitry may comprise, or be coupled to, power management circuitry and is configured to supply the components of apparatus 600 with power for performing the functionality described herein. Power circuitry may receive power from a power source. The power source and/or power circuitry may be configured to provide power to the various components of apparatus 600 in a form suitable for the respective components (e.g., at a voltage and current level needed for each respective component). The power source may either be included in, or external to, the power circuitry and/or the apparatus 600. For example, the apparatus 600 may be connectable to an external power source (e.g., an electricity outlet) via an input circuitry or interface such as an electrical cable, whereby the external power source supplies power to the power circuitry. As a further example, the power source may comprise a source of power in the form of a battery or battery pack which is connected to, or integrated in, the power circuitry. The battery may provide backup power should the external power source fail. Other types of power sources, such as photovoltaic devices, may also be used.

Figure 7 is a schematic diagram of an apparatus 700 according to further embodiments of the disclosure. The apparatus 700 may perform the method described above with respect to Figure 3. The apparatus 700 may correspond to or be implemented within the management node 102 or the apparatus 200 described above.

The apparatus 700 comprises a receiving unit 702, an accessing unit 704 and a selecting unit 706. The receiving unit 702 is configured to receive a measurement policy comprising an indication of one or more mandatory metrics. The accessing unit 704 is configured to access a training dataset comprising data for a plurality of possible metrics from the communication network. The selecting unit 706 is configured to select, based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored. The plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics. The one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics. The term “unit” may have conventional meaning in the field of electronics, electrical devices and/or electronic devices and may include, for example, electrical and/or electronic circuitry, devices, modules, processors, memories, logic solid state and/or discrete devices, computer programs or instructions for carrying out respective tasks, procedures, computations, outputs, and/or displaying functions, and so on, as such as those that are described herein.

The disclosure thus provides methods, apparatus and computer-readable media for selecting metrics to be monitored in a communication network. In particular, the method comprises receiving an indication of one or more mandatory metrics to be monitored, and then selecting additional metrics to be monitored taking into account the mandatory metrics.

References in the present disclosure to “one embodiment”, “an embodiment” and so on, indicate that the embodiment described may include a particular feature, structure, or characteristic, but it is not necessary that every embodiment includes the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to implement such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.

It should be understood that, although the terms “first”, “second” and so on may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could be termed a second element, and similarly, a second element could be termed a first element, without departing from the scope of the disclosure. As used herein, the term “and/or” includes any and all combinations of one or more of the associated listed terms.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the present disclosure. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises”, “comprising”, “has”, “having”, “includes” and/or “including”, when used herein, specify the presence of stated features, elements, and/or components, but do not preclude the presence or addition of one or more other features, elements, components and/ or combinations thereof. The terms “connect”, “connects”, “connecting” and/or “connected” used herein cover the direct and/or indirect connection between two elements. The present disclosure includes any novel feature or combination of features disclosed herein either explicitly or any generalization thereof. Various modifications and adaptations to the foregoing exemplary embodiments of this disclosure may become apparent to those skilled in the relevant arts in view of the foregoing description, when read in conjunction with the accompanying drawings. However, any and all modifications will still fall within the scope of the non-limiting and exemplary embodiments of this disclosure.

Claims

1. A method performed by an apparatus for a communication network, for configuring metrics to be monitored in the communication network, the method comprising: receiving (300) a measurement policy comprising an indication of one or more mandatory metrics; accessing (302) a training dataset comprising data for a plurality of possible metrics from the communication network; and selecting (304), based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored, wherein the plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics, and wherein the one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics.

2. The method according to claim 1 , wherein the one or more additional metrics are selected so as to reduce information loss when monitoring the plurality of metrics to be monitored instead of the plurality of possible metrics.

3. The method according to claim 1 or claim 2, wherein selecting a plurality of metrics to be monitored comprises providing the one or more mandatory metrics as statically defined values in a machine-learning model (400), and training the machine-learning model based on the training dataset.

4. The method according to claim 3, wherein the machine-learning model comprises a neural network.

5. The method according to claim 4, wherein the one or more mandatory metrics are defined by respective one-hot variables in a selector layer (402) of the neural network, and wherein the selector layer further comprises a variable vector representing the one or more additional metrics of the plurality of possible metrics.

6. The method according to claim 5, wherein the neural network (400) is configured to reconstruct a subset of data samples from the training dataset output by the selector layer, and wherein weights of the neural network are updated iteratively based on a loss function between original values of the subset of data samples and reconstructed values of the subset of data samples until a stop criterion is reached.

7. The method according to claim 6, wherein the stop criterion comprises convergence of the loss function to less than a predefined value between successive iterative updates to the weights of the neural network.

8. The method according to any one of claims 5 to 7, wherein values of the variable vector are updated iteratively according to a temperature parameter which controls a level of discreteness of the values of the variable vector.

9. The method according to claim 8, wherein a value of the temperature parameter is reduced in successive iterations.

10. The method according to claim 8 or 9, wherein values of the variable vector are compared to a threshold, and wherein additional metrics are selected for the plurality of metrics to be monitored based on the comparison to the threshold.

11. The method according to any one of the preceding claims, further comprising configuring (308) one or more nodes of the communication network to monitor the plurality of metrics to be monitored.

12. The method according to any one of the preceding claims, further comprising monitoring (306) the plurality of metrics to be monitored.

13. The method according to any one of the preceding claims, wherein a number of the plurality of metrics to be monitored is predetermined.

14. An apparatus (102, 200, 600, 700) for configuring metrics to be monitored in a communication network, the apparatus comprising processing circuitry (602) and a non-transitory machine-readable medium (604) storing instructions which, when executed by the processing circuitry, cause the apparatus to: receive (300) a measurement policy comprising an indication of one or more mandatory metrics; access (302) a training dataset comprising data for a plurality of possible metrics from the communication network; and select (304), based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored, wherein the plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics, and wherein the one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics.

15. The apparatus according to claim 14, wherein the one or more additional metrics are selected so as to reduce information loss when monitoring the plurality of metrics to be monitored instead of the plurality of possible metrics.

16. The apparatus according to claim 14 or claim 15, wherein the apparatus is caused to select a plurality of metrics to be monitored by providing the one or more mandatory metrics as statically defined values in a machine-learning model, and training the machine-learning model based on the training dataset.

17. The apparatus according to claim 16, wherein the machine-learning model comprises a neural network.

18. The apparatus according to claim 17, wherein the one or more mandatory metrics are defined by respective one-hot variables in a selector layer of the neural network, and wherein the selector layer further comprises a variable vector representing the one or more additional metrics of the plurality of possible metrics.

19. The apparatus according to claim 18, wherein the neural network is configured to reconstruct a subset of data samples from the training dataset output by the selector layer, and wherein weights of the neural network are updated iteratively based on a loss function between original values of the subset of data samples and reconstructed values of the subset of data samples until a stop criterion is reached.

20. The apparatus according to claim 19, wherein the stop criterion comprises convergence of the loss function to less than a predefined value between successive iterative updates to the weights of the neural network.

21. The apparatus according to any one of claims 18 to 20, wherein values of the variable vector are updated iteratively according to a temperature parameter which controls a level of discreteness of the values of the variable vector.

22. The apparatus according to claim 21 , wherein a value of the temperature parameter is reduced in successive iterations.

23. The apparatus according to claim 21 or 22, wherein values of the variable vector are compared to a threshold, and wherein additional metrics are selected for the plurality of metrics to be monitored based on the comparison to the threshold.

24. The apparatus according to any one of claims 14 to 23, wherein the apparatus is further caused to configure one or more nodes of the communication network to monitor the plurality of metrics to be monitored.

25. The apparatus according to any one of claims 14 to 24, wherein the apparatus is further caused to monitor the plurality of metrics to be monitored.

26. The apparatus according to any one of claims 14 to 25, wherein a number of the plurality of metrics to be monitored is predetermined.

27. A non-transitory machine-readable medium storing instructions which, when executed by processing circuitry of an apparatus, cause the apparatus to: receive a measurement policy comprising an indication of one or more mandatory metrics; access a training dataset comprising data for a plurality of possible metrics from the communication network; and select, based on the training dataset and the one or more mandatory metrics, a plurality of metrics to be monitored, wherein the plurality of metrics to be monitored is a subset of the plurality of possible metrics and comprises the one or more mandatory metrics and one or more additional metrics of the plurality of possible metrics, and wherein the one or more additional metrics are selected based on statistical interaction between the one or more mandatory metrics and the one or more additional metrics.