Optimization in communication networks

Abstract

A meta-optimizer node, optimizer node, and methods of operating thereof are disclosed. A meta-optimizer node for use in a communication network, the network comprising a plurality of optimizer nodes, wherein respective nodes of the plurality of optimizer nodes are configured to perform, using a model, an optimization task for the communication network, the meta-optimizer node configured to: receive, from the plurality of optimizer nodes, respective optimization task metadata; send, to an optimizer node of the plurality of optimizer nodes for input to the model, a representation of the optimization tasks of the plurality of optimizer nodes, wherein the representation is generated based on the received task metadata.

Description

[0001] OPTIMIZATION IN COMMUNICATION NETWORKS

[0002] Technical Field

[0003] This disclosure relates to methods, nodes and communication networks. More particularly but non- exclusively, the disclosure relates to optimization in communication networks.

[0004] Background

[0005] Optimization of communication network parameters has a potential to improve quality of experience (QoE) of the users of the network, make it more robust, flexible and resource efficient. There are challenges however, because the choice of an optimal parameter may depend on a number of varying characteristics of the network environment, for example the traffic distribution or the type of propagation environment. Optimization techniques using artificial intelligence (Al) are promising to solve this problem.

[0006] Almasan, Paul, et al. "Digital twin network: Opportunities and challenges." arXiv preprint arXiv:2201.01144 (2022) describes use of digital twin (DT) and optimizer pairs to find optimal network parameters. An Al optimizer is interacting with the digital twin, using reinforcement learning or other optimization techniques, such as for example zeroth-order block coordinate decent, to try parameter combinations and evaluate their performance. Once iterations with the digital twin are complete, the best parameter configuration identified is sent to the real network for deployment.

[0007] Chen, Fei, et al. "Federated meta-learning with fast convergence and efficient communication." arXiv preprint arXiv:1802.07876 (2018) teaches a federated meta-learning framework, where a parameterized algorithm (or meta-learner) is shared, instead of a global model in previous approaches.

[0008] The interaction between the Al optimizer and digital twin is performed every time a new parameter needs to be provided to the network, for example, on a daily basis. This interaction can consist of many iterations between the Al optimizer and digital twin which requires a significant amount of computing resources power and could lead to increasing operating expenditures (OPEX). We refer to this interaction as the inner loop. Another challenge with existing technology is how to address the fact that multiple types of network parameters need to be changed. These parameters may include, for example antenna tilt, downlink power, load balancing between frequencies, antenna azimuth, cell individual offset, uplink power etc. Deploying one large digital twin - optimizer pair to optimize all those parameters jointly would be difficult to scale and would make the number of iterations between optimizer and digital twin even larger. Additionally, each of these parameters may require different update frequencies.

[0009] Scaling difficulties also arise when considering the number of nodes in the network that should be optimized.

[0010] Techniques, such as warm starting the model in the optimizer, can be used to further increase efficiency of the inner loops. However, in the case of Al-based optimization, it has been shown that warm starting can cause catastrophic forgetting and could lead to worse performance than without warm starting - a problem also known as a negative transfer gain.

[0011] Pre-training the optimization model is another alternative technique which would not require any inner loop cost at deployment time. However, it might cause the optimizer to be less adaptable to changes in the environment. All the environment conditions would need to be anticipated at training time.

[0012] There is therefore a need for improved solutions that address at least some of the aforementioned issues. Thus, according to a first aspect herein there is provided a meta-optimizer node for use in a communication network. The network comprises a plurality of optimizer nodes, wherein respective nodes of the plurality of optimizer nodes are configured to perform, using a model, an optimization task for the communication network. The meta-optimizer node is configured to receive, from the plurality of optimizer nodes, respective optimization task metadata; send, to an optimizer node of the plurality of optimizer nodes for input to the model, a representation of the optimization tasks of the plurality of optimizer nodes, wherein the representation is generated based on the received task metadata.

[0013] According to a second aspect herein there is provided an optimizer node for a communication network, the optimizer node configured to perform an optimization task using a model. The optimizer node is configured to send, to a meta-optimizer node, optimization task metadata; receive, from the meta-optimizer node, a representation of the optimization task and of at least a further optimization task performed by another optimizer node; input the received representation into the model. According to a third aspect herein there is provided a communication network. The communication network comprises a meta-optimizer node according to the first aspect and a plurality of optimizer nodes according to the second aspect.

[0014] According to a fourth aspect herein there is provided a method of operating a meta-optimizer node for use in a communication network. The network comprises a plurality of optimizer nodes, wherein respective nodes of the plurality of optimizer nodes are configured to perform, using a model, an optimization task for the communication network. The method comprises the steps of receiving, from the plurality of optimizer nodes, respective optimization task metadata; sending, to an optimizer node of the plurality of optimizer nodes for input to the model, a representation of the optimization tasks of the plurality of optimizer nodes, wherein the representation is generated based on the received task metadata.

[0015] According to a fifth aspect herein there is provided a method of operating an optimizer node for a communication network. The optimizer node is configured to perform an optimization task using a model. The method comprises the steps of sending, to a meta-optimizer node, optimization task metadata; receiving, from the meta-optimizer node, a representation of the optimization task and of at least a further optimization task performed by another optimizer node; inputting the received representation into the model.

[0016] According to a sixth aspect herein there is provided a computer program comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out a method according to any one of the fourth or fifth aspect.

[0017] According to a seventh aspect herein there is provided a computer program product comprising non transitory computer readable media having stored thereon a computer program according to the sixth aspect.

[0018] Brief iption of the

[0019] 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, by way of example only, to the following drawings in which:

[0020] Figure 1 shows an example of a communication network.

[0021] Figure 2 is a block diagram illustrating a communication network according to some embodiments. Figure 3 shows a network node in accordance with some embodiments.

[0022] Figure 4 is a block diagram illustrating a virtualization environment in which functions implemented by some embodiments may be virtualized.

[0023] Figure 5 is a signalling diagram illustrating another example communication network according to some embodiments.

[0024] Figure 6 is a signalling diagram illustrating another example communication network in the O-RAN context, according to some embodiments.

[0025] Figure 7 is a block diagram illustrating steps of the method of operating a meta-optimizer node according to some embodiments.

[0026] Figure 8 is a block diagram illustrating steps of the method of operating an optimizer node according to some embodiments.

[0027] Detailed Description

[0028] The following sets forth specific details, such as particular embodiments or examples for purposes of explanation and not limitation. It will be appreciated by one skilled in the art that other examples may be employed apart from these specific details. In some instances, detailed descriptions of well-known methods, nodes, interfaces, circuits, and devices are omitted so as not obscure the description with unnecessary detail. Those skilled in the art will appreciate that the functions described may be implemented in one or more nodes using hardware circuitry (e.g., analog and / or discrete logic gates interconnected to perform a specialized function, ASICs, PLAs, etc.) and / or using software programs and data in conjunction with one or more digital microprocessors or general-purpose computers. Nodes that communicate using the air interface also have suitable radio communications circuitry. Moreover, where appropriate the technology can additionally be considered to be embodied entirely within any form of computer-readable memory, such as solid-state memory, magnetic disk, or optical disk containing an appropriate set of computer instructions that would cause a processor to carry out the techniques described herein.

[0029] Hardware implementation may include or encompass, without limitation, digital signal processor, DSP hardware, a reduced instruction set processor, hardware (e.g., digital or analogue) circuitry including but not limited to application specific integrated circuit(s), ASIC and / or field programmable gate array(s), FPGA(s), and (where appropriate) state machines capable of performing such functions. Figure 1 shows an example of a communication network 100 in accordance with some embodiments. The communication or telecommunication network 102 may include an access network 104, such as a radio access network (RAN), and a core network 106, which includes one or more core network nodes 108. The access network 104 includes one or more access network nodes, such as network nodes 110a and 110b (one or more of which may be generally referred to as network nodes 110), or any other similar 3rd Generation Partnership Project (3GPP) access nodes or non-3GPP access points. Moreover, as will be appreciated by those of skill in the art, a network node is not necessarily limited to an implementation in which a radio portion and a baseband portion are supplied and integrated by a single vendor. Thus, it will be understood that network nodes include disaggregated implementations or portions thereof. For example, in some embodiments, the telecommunication network 102 includes one or more Open-RAN (ORAN) network nodes. An ORAN network node is a node in the telecommunication network 102 that supports an ORAN specification (e.g., a specification published by the O-RAN Alliance, or any similar organization) and may operate alone or together with other nodes to implement one or more functionalities of any node in the telecommunication network 102, including one or more network nodes 110 and / or core network nodes 108.

[0030] Examples of an ORAN network node include an open radio unit (O-RU), an open distributed unit (O- DU), an open central unit (O-CU), including an O-CU control plane (O-CU-CP) or an O-CU user plane (O-CU-UP), a RAN intelligent controller (near-real time or non-real time) hosting software or software plug-ins, such as a near-real time control application (e.g., xApp) or a non-real time control application (e.g., rApp), or any combination thereof (the adjective "open" designating support of an ORAN specification). The network node may support a specification by, for example, supporting an interface defined by the ORAN specification, such as an Al, Fl, Wl, El, E2, X2, Xn interface, an open fronthaul user plane interface, or an open fronthaul management plane interface. Moreover, an ORAN access node may be a logical node in a physical node. Furthermore, an ORAN network node may be implemented in a virtualization environment (described further below) in which one or more network functions are virtualized. For example, the virtualization environment may include an O-Cloud computing platform orchestrated by a Service Management and Orchestration Framework via an O-2 interface defined by the O-RAN Alliance or comparable technologies. The network nodes 110 facilitate direct or indirect connection of user equipment ( U E ), such as by connecting UEs 112a, 112b, 112c, and 112d (one or more of which may be generally referred to as UEs 112) to the core network 106 over one or more wireless connections.

[0031] Example wireless communications over a wireless connection include transmitting and / or receiving wireless signals using electromagnetic waves, radio waves, infrared waves, and / or other types of signals suitable for conveying information without the use of wires, cables, or other material conductors. Moreover, in different embodiments, the communication network 102 may include any number of wired or wireless networks, network nodes, UEs, and / or any other components or systems that may facilitate or participate in the communication of data and / or signals whether via wired or wireless connections. The communication network 102 may include and / or interface with any type of communication, telecommunication, data, cellular, radio network, and / or other similar type of system.

[0032] The UEs 112 may be any of a wide variety of communication devices, including wireless devices arranged, configured, and / or operable to communicate wirelessly with the network nodes 110 and other communication devices. Similarly, the network nodes 110 are arranged, capable, configured, and / or operable to communicate directly or indirectly with the UEs 112 and / or with other network nodes or equipment in the telecommunication network 102 to enable and / or provide network access, such as wireless network access, and / or to perform other functions, such as administration in the telecommunication network 102.

[0033] In the depicted example, the core network 106 connects the network nodes 110 to one or more hosts, such as host 116. These connections may be direct or indirect via one or more intermediary networks or devices. In other examples, network nodes may be directly coupled to hosts. The core network 106 includes one more core network nodes (e.g., core network node 108) that are structured with hardware and software components. Features of these components may be substantially similar to those described with respect to the UEs, network nodes, and / or hosts, such that the descriptions thereof are generally applicable to the corresponding components of the core network node 108. Example core network nodes include functions of one or more of a Mobile Switching Center (MSC), Mobility Management Entity (MME), Home Subscriber Server (HSS), Access and Mobility Management Function (AMF), Session Management Function (SMF), Authentication Server Function (AUSF), Subscription Identifier De-concealing function (SIDF), Unified Data Management (UDM), Security Edge Protection Proxy (SEPP), Network Exposure Function (NEF), and / or a User Plane Function (UPF).

[0034] The host 116 may be under the ownership or control of a service provider other than an operator or provider of the access network 104 and / or the telecommunication network 102, and may be operated by the service provider or on behalf of the service provider. The host 116 may host a variety of applications to provide one or more service. Examples of such applications include live and prerecorded audio / video content, data collection services such as retrieving and compiling data on various ambient conditions detected by a plurality of UEs, analytics functionality, social media, functions for controlling or otherwise interacting with remote devices, functions for an alarm and surveillance center, or any other such function performed by a server.

[0035] As a whole, the communication network 102 of Figure 1 enables connectivity between the UEs, network nodes, and hosts. In that sense, the communication network may be configured to operate according to predefined rules or procedures, such as specific standards that include, but are not limited to: Global System for Mobile Communications (GSM); Universal Mobile Telecommunications System (UMTS); Long Term Evolution (LTE), and / or other suitable 2G, 3G, 4G, 5G standards, or any applicable future generation standard (e.g., 6G); wireless local area network (WLAN) standards, such as the Institute of Electrical and Electronics Engineers (IEEE) 802.11 standards (WiFi); and / or any other appropriate wireless communication standard, such as the Worldwide Interoperability for Microwave Access (WIMax), Bluetooth, Z-Wave, Near Field Communication (NFC) ZigBee, LIFi, and / or any low- power wide-area network (LPWAN) standards such as LoRa and Sigfox.

[0036] In some examples, the telecommunication network 102 is a cellular network that implements 3GPP standardized features. Accordingly, the telecommunications network 102 may support network slicing to provide different logical networks to different devices that are connected to the telecommunication network 102. For example, the telecommunications network 102 may provide Ultra Reliable Low Latency Communication (URLLC) services to some UEs, while providing Enhanced Mobile Broadband (eMBB) services to other UEs, and / or Massive Machine Type Communication (mMTC) / Massive loT services to yet further UEs.

[0037] In some examples, the UEs 112 are configured to transmit and / or receive information without direct human interaction. For instance, a UE may be designed to transmit information to the access network 104 on a predetermined schedule, when triggered by an internal or external event, or in response to requests from the access network 104. Additionally, a UE may be configured for operating in single- or multi-RAT or multi-standard mode. For example, a UE may operate with any one or combination of Wi-Fi, NR (New Radio) and LTE, i.e. being configured for multi-radio dual connectivity (MR-DC), such as E-UTRAN (Evolved-UMTSTerrestrial Radio Access Network) New Radio - Dual Connectivity (EN-DC).

[0038] In the example, the hub 114 communicates with the access network 104 to facilitate indirect communication between one or more UEs (e.g., UE 112c and / or 112d) and network nodes (e.g., network node 110b). In some examples, the hub 114 may be a controller, router, content source and analytics, or any of the other communication devices described herein regarding UEs. For example, the hub 114 may be a broadband router enabling access to the core network 106 for the UEs. As another example, the hub 114 may be a controller that sends commands or instructions to one or more actuators in the UEs. Commands or instructions may be received from the UEs, network nodes 110, or by executable code, script, process, or other instructions in the hub 114. As another example, the hub 114 may be a data collector that acts as temporary storage for UE data and, in some embodiments, may perform analysis or other processing of the data. As another example, the hub 114 may be a content source. For example, for a UE that is a VR headset, display, loudspeaker or other media delivery device, the hub 114 may retrieve VR assets, video, audio, or other media or data related to sensory information via a network node, which the hub 114 then provides to the UE either directly, after performing local processing, and / or after adding additional local content. In still another example, the hub 114 acts as a proxy server or orchestrator for the UEs, in particular if one or more of the UEs are low energy loT devices.

[0039] The hub 114 may have a constant / persistent or intermittent connection to the network node 110b. The hub 114 may also allow for a different communication scheme and / or schedule between the hub 114 and UEs (e.g., UE 112c and / or 112d), and between the hub 114 and the core network 106. In other examples, the hub 114 is connected to the core network 106 and / or one or more UEs via a wired connection. Moreover, the hub 114 may be configured to connect to an M2M service provider over the access network 104 and / or to another UE over a direct connection. In some scenarios, UEs may establish a wireless connection with the network nodes 110 while still connected via the hub 114 via a wired or wireless connection. In some embodiments, the hub 114 may be a dedicated hub - that is, a hub whose primary function is to route communications to / from the UEs from / to the network node 110b. In other embodiments, the hub 114 may be a non-dedicated hub - that is, a device which is capable of operating to route communications between the UEs and network node 110b, but which is additionally capable of operating as a communication start and / or end point for certain data channels.

[0040] Figure 2 shows a communication network 200 according to some embodiments. The network 200 may comprise any of the example networks 102, 104, 106, 450, 460 described in respect of figure 1 and 2. Said network 200 further comprises a plurality of optimizer nodes 212, 222, 232, 242. The respective nodes of the plurality of optimizer nodes 212, 222, 232, 242 are configured to perform, using a model 213, 223, 233, 243, an optimization task for the communication network 200.

[0041] The optimizer node 212, 222, 232, 242 is in charge of searching for the best network configuration that fulfils the specified requirements e.g., by a network operator, such as minimizing the maximum link utilization. If the performance metrics from the network indicate that the solution is not good enough, the network optimizer continues the search until the stopping condition is met. The best solution found so far may be applied directly to the live network 250, 260.

[0042] The optimization task may comprise updating of a network parameter A, B, C, D for said network 200, such as live network 250, 260. The update may be performed as a direct command, such as sending a desired value of the network parameter A, B, C, D directly to the network 200, 250, 260. The update may also comprise an incremental command, such as instructing to increase or decrease a value of the network parameter A, B, C, D. The network parameters to be updated may depend on the type of the communication network 200. For radio-type networks, the network parameter may comprise, for example, any of the configuration management (CM) parameters, a maximum downlink transmission power, remote electrical antenna tilt, half-power beam width, azimuth, cell individual offset or interfrequency load balancing parameter.

[0043] An example illustrating scenario involving two optimization tasks A and B is the following. Optimization task A comprises optimizing a remote electrical tilt (RET) of base station antennas in a cluster of base stations (cluster A). The network parameter consists of changing the CM parameters corresponding to RET for all the antennas in the cluster. The optimizer node for this task uses a reinforcement learning agent. The reinforcement learning agent controls the tilt incrementally in the digital twin. After a few iterations, the agent has steered the digital twin towards an optimal tilt configuration. This configuration is then implemented in the real network. The reward function is an average of coverage, quality and throughput. Input features comprise aggregated coverage, quality, and throughput KPIs, current tilt value and current power value, and deployment-related features such as antenna heights, urban / rural deployment, intersite distance, size of the RET increments, mechanical tilt.

[0044] Optimization task B comprises optimizing the maximum downlink power of base station antennas, in another cluster of base stations (cluster B). The network parameter comprises changing the CM parameters corresponding to the maximum downlink power for all the antennas in the cluster. The optimizer node also uses a reinforcement learning agent. The reward function is to minimize the average downlink power and contains a penalty term for coverage degradation. Input features may be the same as for task A.

[0045] The optimizer nodes 212, 222, 232, 242 use their respective models 213, 223, 233, 243 to perform the optimization task. The model 213, 223, 233, 243 may comprise a trainable machine learning model. This machine learning model may be initialized with weights or can be conditioned by a representation of the optimization tasks, for example in the form of a context vector. In one embodiment, the optimizer node may comprise a gradient-based optimizer, such as one using reinforcement learning or gradient descent. The optimizer node aims to minimize a loss function by computing gradients with respect to that loss. The computed loss function or the model may be non-differentiable.

[0046] The model 213, 223, 233, 243 takes as input a list of network key performance indicators (KPI) that are observed by the optimizer node and describes the state of the network 200, 250, 260. These input features are generated by network and may comprise Performance Management (PM) counters and CM counters. In the context of a radio-type network, examples of such counters include: loads of each cells, Reference Signal Received Power (RSRP) of each user, interference levels, angle of arrival, aggregated coverage, Quality of Service (QoS) information, throughput data. The input features may also comprise static information that is deployment related, for example inter-site distance, height of the base station antenna.

[0047] The optimizer node 212, 222, 232, 242 is configured to send, to a meta-optimizer node 270, optimization task metadata. The metadata may comprise all the characteristics of an optimization task. The optimization task metadata may comprise an indicator of a type of the network parameter. The metadata may further comprise at least one indicator of a type of network operation data used by the corresponding optimizer node 212, 222, 232, 242 to evaluate an effect of updating the network parameter. At least one indicator of a type of network operation data may comprise a network performance metric (e.g., PM counters) or network configuration information (e.g. CM counters). The metadata may encode the characteristics of the optimization task into a vector representation. The vector representation may comprise a one-hot encoded vector that has a value '1' for all the features used by the tasks among all the possible features used by the optimizer node. For example, one optimizer may use information about an angle of arrival while another may not use that feature in the input features. These two tasks will have different task metadata.

[0048] The optimization task metadata may also comprise network deployment information, such as location of the cells, which cell cluster is being optimized, antenna heights, location of deployment (e.g., urban or rural), summary of the traffic distribution in the cluster of cells being optimized (e.g. load, variations in the load, type of traffic). This information may be fixed over several rounds of interactions between the optimizer and the network.

[0049] The network operation data may be obtained directly from the subnetwork 251, 252, 261, or may be generated by the digital twin 211, 221, 231, 241. The respective optimizer node 212, 222, 232, 242 may be paired with a digital twin 211, 221, 231, 241 of a subnetwork 251, 252, 261 of the communication network 250, 260, and the pair of the digital twin 211, 221, 231, 241 and optimizer node 212, 222, 232, 242 may be configured to jointly perform the optimization task for the subnetwork 251, 252, 261 by iteratively updating the network parameter for said subnetwork 251, 252, 261 until a condition is satisfied.

[0050] Providing distributed digital twins and optimizers where one pair of digital twin-optimizer is associated with a given optimization task addresses challenges related to maintaining a single centralized digital twin for the entire network. The optimization task may include optimizing one or more network parameters for a cluster of nodes. This setup is more scalable. To further improve this setup and enable pairs to benefit from learnings of each other, the meta-optimizer 270 is introduced. In this way, the lifecycle management of each digital twin - optimizer pair is facilitated, and independent management of a large number of models is no longer required. This in turn saves resources.

[0051] The digital twin 211, 221, 231, 241 may comprise a digital representation or a faithful copy of the real- world network, such as subnetwork 251, 252, 261. The digital twin includes live network measurement data such as users' radio frequency (RF) characteristics, traffic profiles, mobility patterns, network metrics such as cell utilization, energy consumptions, configuration settings etc. The digital twin may take tuneable parameters value as input such as RET, DL power, UL power, handover related threshold parameters, run a sequence of behaviour models to mimic how the real system would behave, and output network-related performance metrics, such as utilization, coverage, inference, energy consumptions. The behaviour models may comprise analytic-based models (such as mathematical operations, if-else logics, join operations) and ML-based models.

[0052] The optimizer node 212, 222, 232, 242 may be interacting with the digital twin 211, 221, 231, 241 using reinforcement learning, or other optimization techniques (e.g., zeroth-order block coordinate decent) to try parameter combinations and evaluate their performance. After iterating with the digital twin 211, 221, 231, 241, the best parameter configuration found may be sent to the subnetwork 251, 252, 261 for deployment. The interaction between the optimizer nodes 212, 222, 232, 242 and the digital twins 211, 221, 231, 241 may be performed every time a new parameter needs to be sent to the network, for example on a daily basis.

[0053] Optimizing one node affects the performance of its neighboring node. Attempting to optimize an entire network jointly would be unscalable. Clustering the nodes and optimizing different clusters is more favorable. Pairing the digital twin with an optimizer node for a given optimization task, which may comprise optimizing one or more network parameters, and for each subnetwork 251, 252, 261, such as a cluster 251, 252, 261 of nodes, is more scalable and considers the fact that performance of one node may affect the performance of a neighboring node at the scale of the subnetwork. The subnetwork 251, 252, 261 may comprise a plurality of network entities grouped according to a similarity criterion. In one example, the subnetwork comprises a radio access network, RAN. The network parameter may comprise a RAN parameter. In one example, the subnetwork 251, 252, 261 may comprise a cluster of cells 251, 252, 261 of RAN. The similarity criterion in this scenario may be the neighboring relationship between the nodes in the cluster.

[0054] The optimizer node 212, 222, 232, 242 may be configured to send, to the meta-optimizer node 270, a model performance value and receive, from the meta-optimizer node, a model parameter value for updating the model 213. The model parameter value may be generated based on the optimization task metadata and model performance value and based on further optimization task metadata and a further model performance value associated with another optimizer node. The optimization task and the further optimization task may be the same.

[0055] The optimizer node 212, 222, 232, 242 may be further configured to send, to the meta-optimizer node 270, dynamic data that represents state trajectories of the inner loop. The state trajectories may be represented by the list of input features values, applied parameter changes and cost functions in the digital twin.

[0056] The meta-optimizer node 270 is to be used in a communication network 200. Said network 200 further comprises a plurality of optimizer nodes 212, 222, 232, 242. The respective nodes 212, 222, 232, 242 of the plurality of optimizer nodes are configured to perform, using a model 213, 223, 233, 243 an optimization task for the communication network 200, such as respective cell clusters 251, 252, 261.

[0057] The meta-optimizer node 270 is configured to receive, from the plurality of optimizer nodes 612, 622, 632, 642, respective optimization task metadata. The meta-optimizer node 270 is further configured to send, to an optimizer node of the plurality of optimizer nodes 212, 222, 232, 242 for input to the model, a representation of the optimization tasks of the plurality of optimizer nodes 212, 222, 232, 242, wherein the representation is generated based on the received task metadata.

[0058] The optimization task may comprise updating of a network parameter for the communication network 200. The optimization task metadata comprises an indicator of a type of the network parameter. The optimization task metadata may further comprise at least one indicator of a type of network operation data used by the corresponding optimizer node 212, 222, 232, 242 to evaluate an effect of updating the network parameter. The at least one indicator may comprise a network performance metric or network configuration information. The representation of the optimization tasks or context representation may be generated by a context or task embedding generator, which may comprise a machine learning model trained to output the embedding, for example in the form of a vector. The representation may comprise an embedding generated using an attention neural network. As such it may accept a variable number of inputs for the optimization task metadata and would make it easy to use the meta-optimizer 270 asynchronously. The generator may be trained to map the optimization task metadata to a compact vector representation. The representation or context vector is then used as input by the models (e.g., Al models) in the optimizer node to condition its internal model 213, 223, 233, 243 (e.g., condition its internal policy). This approach is more flexible as it adapts well to tasks with different design spaces. The weights of the context generator may be updated by minimizing the aggregated loss of all the optimizer nodes 212, 222, 232, 242 that sent data to the meta-optimizer 270. All the optimizer nodes

[0059] 212, 222, 232, 242 may use the same policy weights, and may specialize just by receiving different task context from the meta-optimizer 270. The task context vector may be added as input to the Al model

[0060] 213, 223, 233, 243 of the optimizer node 212, 222, 232, 242 along with its other input features to help it specialize. It acts as a conditioning vector that helps the models 213, 223, 233, 243 adapt quickly thus requiring less iterations with the network or digital twins.

[0061] The meta-optimizer node 270 may be further configured to receive, from the plurality of optimizer nodes 212, 222, 232, 242 respective model performance values. The model 213, 223, 233, 243 may comprise a gradient-based optimization model and the model performance values may comprise a loss value or a gradient of a loss value. The model performance values may comprise loss values with respect to interactions between an optimizer node 212, 222, 232, 242 and the digital twin 211, 221,

[0062] 231, 241 or the network 200, 250, 260, 251, 252, 261. The model performance values may comprise one value per interaction or one value per episode, where the episode corresponds to a series of interactions after the digital twin 211, 221, 231, 241 is updated. The model performance values may further comprise gradients of the loss with respect to the task context vector or optimizer model parameter value (e.g. weight), or meta-gradients. The communication of these meta-gradients may be the triggered dynamically. The meta-gradients may be computed by the optimizer node 212, 222,

[0063] 232, 242 before being sent. These meta-gradients are not the gradients with respect to the policy weights of the optimizer node. They contain the information to update the optimization model parameters and / or context vector in such a way that a few iterations of optimizer node policy weights update will lead to a good performance of the optimizer node.

[0064] The meta-optimizer node 270 may be further configured to send, to the optimizer node of the plurality of optimizer nodes 212, 222, 232, 242, a model parameter value, such as a meta-weight, for updating the model 213, 223, 233, 243 of the respective optimizer node 212, 222, 232, 242. The model parameter value may be generated based on the received optimization task metadata and model performance values. The meta-optimizer node 270 may comprise a machine learning model 271, such as a neural network 271. The model parameter value may comprise an output of said neural network 271. The model parameter value may be further based on respective task priority values determined for the respective optimization tasks. The determination of the task priority values may comprise learning the respective task priority values using a priority machine learning model based on the received optimization task metadata.

[0065] The meta-weights 0 may be updated using an update rule that utilizes the gradients VLl ncommunicated by the optimizer nodes 213, 223, 233, 243. These gradients may be weighted depending on the priority of the optimization tasks. The general form of the update rule may be:

[0066] 6 «- MetaWeightUpdate(doi(i L,n, dn, w-l n')

[0067] In one example, task priorities w,nmay be determined using the optimization task metadata d± nor by passing them explicitly to the update rule. This may be done for example using hand-engineered logics. In one example scenario, a task optimizing downlink (DL) power may be prioritized over a remote electrical tilt (RET), or certain cell clusters may be prioritized over others. In the simplest case, the optimization task metadata, d.l n, may not be used for the meta-weight update and the task priorities would just be one for all tasks. However, in some scenarios, they may be used to filter out certain tasks or increase the priority of other ones. Finally, the meta-weights 6 may be updated using the weighted Model Agnostic Meta Learning (MAM L)update rules (as described for example in Finn, Chelsea, Pieter Abbeel, and Sergey Levine. "Model-agnostic meta-learning for fast adaptation of deep networks." International conference on machine learning. PMLR, 2017):

[0068] 6 «- 90id~ a i W^eLi, where a is the learning rate, a hyperparameter.

[0069] Other techniques, such as federated learning or transfer learning as a special case of federated learning, differ in objective. Federated learning or transfer learning teach back-and-forth exchange of information, such as model parameter gradients, with a server entity. The objective of the meta- optimizer 270 is to optimize the meta-weight for fast adaptation, rather than directly optimizing model parameters for improved performance as in federated learning. More specifically, the gradients received by the meta-optimizer 270 are not the gradient of the optimizer policy weights but rather the gradients with respect to the meta-weights. In other words, the meta-optimizer 270 is not calculating the policy weights, it is calculating meta-weights or context vectors that can then be used by optimizers 213, 223, 233, 243 to compute the policy weights in a more efficient way (with less gradient updates on the policy weights).

[0070] As used herein, network node refers to equipment capable, configured, arranged and / or operable to communicate directly or indirectly with a UE and / or with other network nodes or equipment, in a telecommunication network. Examples of network nodes include, but are not limited to, access points (APs) (e.g., radio access points), base stations (BSs) (e.g., radio base stations, Node Bs, evolved Node Bs (eNBs) and NR NodeBs (gNBs)), O-RAN nodes or components of an O-RAN node (e.g., O-RU, O-DU, O-CU).

[0071] Base stations may be categorized based on the amount of coverage they provide (or, stated differently, their transmit power level) and so, depending on the provided amount of coverage, may be referred to as femto base stations, pico base stations, micro base stations, or macro base stations. A base station may be a relay node or a relay donor node controlling a relay. A network node may also include one or more (or all) parts of a distributed radio base station such as centralized digital units, distributed units (e.g., in an O-RAN access node) and / or remote radio units (RRUs), sometimes referred to as Remote Radio Heads (RRHs). Such remote radio units may or may not be integrated with an antenna as an antenna integrated radio. Parts of a distributed radio base station may also be referred to as nodes in a distributed antenna system (DAS).

[0072] Other examples of network nodes include multiple transmission point (multi-TRP) 5G access nodes, multi-standard radio (MSR) equipment such as MSR BSs, network controllers such as radio network controllers (RNCs) or base station controllers (BSCs), base transceiver stations (BTSs), transmission points, transmission nodes, multi-cell / multicast coordination entities (MCEs), Operation and Maintenance (O&M) nodes, Operations Support System (OSS) nodes, Self-Organizing Network (SON) nodes, positioning nodes (e.g., Evolved Serving Mobile Location Centers (E-SMLCs)), and / or Minimization of Drive Tests (MDTs).

[0073] The network node 300, which may implement an optimizer node or meta-optimizer node according to embodiments, includes a processing circuitry 302, a memory 304, a communication interface 306, and a power source 308. The network node 300 may be composed of multiple physically separate components (e.g., a NodeB component and a RNC component, or a BTS component and a BSC component, etc.), which may each have their own respective components. In certain scenarios in which the network node 300 comprises multiple separate components (e.g., BTS and BSC components), one or more of the separate components may be shared among several network nodes. For example, a single RNC may control multiple NodeBs. In such a scenario, each unique NodeB and RNC pair, may in some instances be considered a single separate network node. In some embodiments, the network node 300 may be configured to support multiple radio access technologies (RATs). In such embodiments, some components may be duplicated (e.g., separate memory 304 for different RATs) and some components may be reused (e.g., a same antenna 310 may be shared by different RATs). The network node 300 may also include multiple sets of the various illustrated components for different wireless technologies integrated into network node 300, for example GSM, WCDMA, LTE, NR, WiFi, Zigbee, Z-wave, LoRaWAN, Radio Frequency Identification (RFID) or Bluetooth wireless technologies. These wireless technologies may be integrated into the same or different chip or set of chips and other components within network node 300.

[0074] The processing circuitry 302 may comprise a combination of one or more of a microprocessor, controller, microcontroller, central processing unit, digital signal processor, application-specific integrated circuit, field programmable gate array, or any other suitable computing device, resource, or combination of hardware, software and / or encoded logic operable to provide, either alone or in conjunction with other network node 300 components, such as the memory 304, to provide network node 300 functionality.

[0075] In some embodiments, the processing circuitry 302 includes a system on a chip (SOC). In some embodiments, the processing circuitry 302 includes one or more of radio frequency (RF) transceiver circuitry 312 and baseband processing circuitry 314. In some embodiments, the radio frequency (RF) transceiver circuitry 312 and the baseband processing circuitry 314 may be on separate chips (or sets of chips), boards, or units, such as radio units and digital units. In alternative embodiments, part or all of RF transceiver circuitry 312 and baseband processing circuitry 314 may be on the same chip or set of chips, boards, or units.

[0076] The memory 304 may comprise any form of volatile or non-volatile computer-readable memory including, without limitation, persistent storage, solid-state memory, remotely mounted memory, magnetic media, optical media, random access memory (RAM), read-only memory (ROM), mass storage media (for example, a hard disk), removable storage media (for example, a flash drive, a Compact Disk (CD) or a Digital Video Disk (DVD)), and / or any other volatile or non-volatile, non- transitory device-readable and / or computer-executable memory devices that store information, data, and / or instructions that may be used by the processing circuitry 302. The memory 304 may store any suitable instructions, data, or information, including a computer program, software, an application including one or more of logic, rules, code, tables, and / or other instructions capable of being executed by the processing circuitry 302 and utilized by the network node 300. The memory 304 may be used to store any calculations made by the processing circuitry 302 and / or any data received via the communication interface 306. In some embodiments, the processing circuitry 302 and memory 304 is integrated.

[0077] The communication interface 306 is used in wired or wireless communication of signalling and / or data between a network node, access network, and / or UE. As illustrated, the communication interface 306 comprises port(s) / terminal(s) 316 to send and receive data, for example to and from a network over a wired connection. The communication interface 306 also includes radio front-end circuitry 318 that may be coupled to, or in certain embodiments a part of, the antenna 310. Radio front-end circuitry 318 comprises filters 320 and amplifiers 322. The radio front-end circuitry 318 may be connected to an antenna 310 and processing circuitry 302. The radio front-end circuitry may be configured to condition signals communicated between antenna 310 and processing circuitry 302. The radio frontend circuitry 318 may receive digital data that is to be sent out to other network nodes or UEs via a wireless connection. The radio front-end circuitry 318 may convert the digital data into a radio signal having the appropriate channel and bandwidth parameters using a combination of filters 320 and / or amplifiers 322. The radio signal may then be transmitted via the antenna 310. Similarly, when receiving data, the antenna 310 may collect radio signals which are then converted into digital data by the radio front-end circuitry 318. The digital data may be passed to the processing circuitry 302. In other embodiments, the communication interface may comprise different components and / or different combinations of components.

[0078] In certain alternative embodiments, the network node 300 does not include separate radio front-end circuitry 318, instead, the processing circuitry 302 includes radio front-end circuitry and is connected to the antenna 310. Similarly, in some embodiments, all or some of the RF transceiver circuitry 312 is part of the communication interface 306. In still other embodiments, the communication interface 306 includes one or more ports or terminals 316, the radio front-end circuitry 318, and the RF transceiver circuitry 312, as part of a radio unit (not shown), and the communication interface 306 communicates with the baseband processing circuitry 314, which is part of a digital unit (not shown). The antenna 310 may include one or more antennas, or antenna arrays, configured to send and / or receive wireless signals. The antenna 310 may be coupled to the radio front-end circuitry 318 and may be any type of antenna capable of transmitting and receiving data and / or signals wirelessly. In certain embodiments, the antenna 310 is separate from the network node 300 and connectable to the network node 300 through an interface or port.

[0079] The antenna 310, communication interface 306, and / or the processing circuitry 302 may be configured to perform any receiving operations and / or certain obtaining operations described herein as being performed by the network node. Any information, data and / or signals may be received from a UE, another network node and / or any other network equipment. Similarly, the antenna 310, the communication interface 306, and / or the processing circuitry 302 may be configured to perform any transmitting operations described herein as being performed by the network node. Any information, data and / or signals may be transmitted to a UE, another network node and / or any other network equipment.

[0080] The power source 308 provides power to the various components of network node 300 in a form suitable for the respective components (e.g., at a voltage and current level needed for each respective component). The power source 308 may further comprise, or be coupled to, power management circuitry to supply the components of the network node 300 with power for performing the functionality described herein. For example, the network node 300 may be connectable to an external power source (e.g., the power grid, an electricity outlet) via an input circuitry or interface such as an electrical cable, whereby the external power source supplies power to power circuitry of the power source 308. As a further example, the power source 308 may comprise a source of power in the form of a battery or battery pack which is connected to, or integrated in, power circuitry. The battery may provide backup power should the external power source fail.

[0081] Embodiments of the network node 300 may include additional components beyond those shown in Figure 3 for providing certain aspects of the network node's functionality, including any of the functionality described herein and / or any functionality necessary to support the subject matter described herein. For example, the network node 300 may include user interface equipment to allow input of information into the network node 300 and to allow output of information from the network node 300. This may allow a user to perform diagnostic, maintenance, repair, and other administrative functions for the network node 300. Figure 4 is a block diagram illustrating a virtualization environment 500 in which functions implemented by some embodiments may be virtualized. In the present context, virtualizing means creating virtual versions of apparatuses or devices which may include virtualizing hardware platforms, storage devices and networking resources. As used herein, virtualization can be applied to any device described herein, or components thereof, and relates to an implementation in which at least a portion of the functionality is implemented as one or more virtual components. Some or all of the functions described herein may be implemented as virtual components executed by one or more virtual machines (VMs) implemented in one or more virtual environments 500 hosted by one or more of hardware nodes, such as a hardware computing device that operates as a network node, UE, core network node, or host. Further, in embodiments in which the virtual node does not require radio connectivity (e.g., a core network node or host), then the node may be entirely virtualized. In some embodiments, the virtualization environment 500 includes components defined by the O-RAN Alliance, such as an O-Cloud environment orchestrated by a Service Management and Orchestration Framework via an 0-2 interface.

[0082] Applications 502 (which may alternatively be called software instances, virtual appliances, network functions, virtual nodes, virtual network functions, etc.) are run in the virtualization environment 400 to implement some of the features, functions, and / or benefits of some of the embodiments disclosed herein.

[0083] Hardware 504 includes processing circuitry, memory that stores software and / or instructions executable by hardware processing circuitry, and / or other hardware devices as described herein, such as a network interface, input / output interface, and so forth. Software may be executed by the processing circuitry to instantiate one or more virtualization layers 506 (also referred to as hypervisors or virtual machine monitors (VMMs)), provide VMs 508a and 508b (one or more of which may be generally referred to as VMs 508), and / or perform any of the functions, features and / or benefits described in relation with some embodiments described herein. The virtualization layer 506 may present a virtual operating platform that appears like networking hardware to the VMs 508.

[0084] The VMs 508 comprise virtual processing, virtual memory, virtual networking or interface and virtual storage, and may be run by a corresponding virtualization layer 506. Different embodiments of the instance of a virtual appliance 502 may be implemented on one or more of VMs 508, and the implementations may be made in different ways. Virtualization of the hardware is in some contexts referred to as network function virtualization (N FV). NFV may be used to consolidate many network equipment types onto industry standard high volume server hardware, physical switches, and physical storage, which can be located in data centers, and customer premise equipment.

[0085] In the context of NFV, a VM 508 may be a software implementation of a physical machine that runs programs as if they were executing on a physical, non-virtualized machine. Each of the VMs 508, and that part of hardware 504 that executes that VM, be it hardware dedicated to that VM and / or hardware shared by that VM with others of the VMs, forms separate virtual network elements. Still in the context of NFV, a virtual network function is responsible for handling specific network functions that run in one or more VMs 508 on top of the hardware 504 and corresponds to the application 502.

[0086] Hardware 504 may be implemented in a standalone network node with generic or specific components. Hardware 504 may implement some functions via virtualization. Alternatively, hardware 504 may be part of a larger cluster of hardware (e.g., such as in a data center or CPE) where many hardware nodes work together and are managed via management and orchestration 510, which, among others, oversees lifecycle management of applications 502. In some embodiments, hardware 504 is coupled to one or more radio units that each include one or more transmitters and one or more receivers that may be coupled to one or more antennas. Radio units may communicate directly with other hardware nodes via one or more appropriate network interfaces and may be used in combination with the virtual components to provide a virtual node with radio capabilities, such as a radio access node or a base station. In some embodiments, some signalling can be provided with the use of a control system 512 which may alternatively be used for communication between hardware nodes and radio units.

[0087] The interaction between elements of the network according to some embodiments will now be illustrated in the context of a flow diagram of figure 7 that illustrates a method 800 of operating a meta-optimizer node, figure 8 that illustrates a method 900 of operating an optimizer node, and signalling diagram of figure 5.

[0088] The example network 600 is a third-generation partnership project (3GPP) network. The meta- optimizer 690 may be introduced as a new network function (NF). The optimizer nodes may be hosted in existing network functions for training machine learning model, such as a Model Training logical function (MTLF). The digital twins may be hosted using data streaming functions, such as Data Collection Coordination Function (DCCF).

[0089] In this example, we consider two clusters of cells, gNodeB cluster A 610 and gNodeB cluster B 620, and two optimization tasks, optimization task A 630 and optimization task B, one for each cluster. The inner-loops for each task are implemented using separate DCCF and MTLF nodes, respectively DCCF A 650 and MTLF A 660, and DCCF B 670 and MTLF B 680, and but each loop could also be implemented in the same node, for example in an MTLF node. Most of the components of this invention can be implemented in a distributed manner. Precisely, all the digital twins, all the optimizers, and the metaoptimizers could be virtualized and executed as separate nodes.

[0090] In general, the frequency of interaction between the optimizers and the digital twins may be much higher than between the meta-optimizer and optimizer nodes. As a consequence, in some cases a preferred design would be to have digital twins and their respective optimizers be part of the same node (different node for each pair), so as to avoid excessive consumption of resources such as network bandwidth and overhead processing.

[0091] The meta-optimizer 690 is getting the data for meta-optimization from both the DCCF and MTLF, and sends the meta-weights and the context vectors to the respective MTLF nodes. In this example scenario, DCCF A 650, MTLF A 660, DCCF B 670, MTLF B 680 and meta-optimizer 690 are all part of a network data and analytics function (NWDAF) node 601.

[0092] Optimization task A 630 comprises RET optimization in gNodeB cluster A 610 and optimization task B comprises power optimization in gNodeB cluster B 620.

[0093] The gNodeB in cluster A 610 sends, at step 611, network data to DCCF A 650. The network data may comprise optimization task metadata, for example meta-features for the optimization task A 630, such as inter-site distance or antenna heights. The digital twin model is updated, at step 651, such that its state reflects the meta-feature.

[0094] The reinforcement learning (RL) loop 631 is then initiated. The RL agent at the MTLF A 660 receives, at step 651, the state and reward from the DCCF A 650. The state corresponds to input features for the optimization task A 630. The reward corresponds to the optimizer node reward (e.g., sum of coverage, quality, and throughput KPIs). The RL agent sends incremental tilt commands at step 661, to the DCCF A 650. The RL agent is then updated according to the RL algorithm being used (e.g., Deep Q Learning, DQ.N) at step 662. This inner loop 631 repeats N times. After iterating with the digital twin, the best tilt configuration found is sent 665 to the gNodeB cluster A 610 for deployment.

[0095] The optimizer node is configured to send 902 and the meta-optimizer 690 is configured to receive 802, from the plurality of optimizer nodes, respective optimization task metadata. After N repetitions of the inner loop 631, the MTLF A 660 communicates 663 to the meta-optimizer 690, the meta-gradients, that is, the gradient of the inner loop loss (e.g. a Bellman error for DQ.N) with respect to the metaweights. The DCCF A 650 communicates 652, to the meta-optimizer 690, meta-features that can be determined to compute the priority of optimization task A 640 according to a rule-based strategy.

[0096] Steps 621, 671, 672, 681, 682 execute in a similar manner for optimization task B 640. The MTLF B communicates 683 the meta-gradients, the DCCF B communicates 673 the meta-features. One of the meta-features for task B 640 indicates that cluster B 620 is a rural area.

[0097] Upon processing the meta-features from task B 640, the meta-optimizer node 690 sets the power optimization priority to 0.5 (half) and the RET optimization priority to 1. A rule-based policy has been implemented to down-prioritize downlink power reduction in the rural area as to not risk any coverage degradation. The meta-optimizer 690 then updates 691, 692 its internal models for generating a representation and model parameter values.

[0098] The meta-optimizer 690 is configured to send 804, and an optimizer node of the plurality of optimizer nodes is configured to receive 904, for input to the model, a representation of the optimization tasks of the plurality of optimizer nodes, wherein the representation is generated based on the received task metadata. The received representation is then input 906 into the model. New meta-weights are generated and sent 693, 694 to MTLF A 660, and MTLF B 680. Similarly, new context vector A and context vector B are sent 695, 696 to MTLF A 660 and MTLF B 680. MTLF A 660 re-initializes 664 the RL agent to use the new meta-weights a starting new inner loop. MTLF B 680 proceeds similarly.

[0099] Figure 6 shows an example embodiment implemented using one or more Open-RAN (ORAN) network nodes. An ORAN network node is a node in the communication network 700 that supports an ORAN specification (e.g., a specification published by the O-RAN Alliance, or any similar organization) and may operate alone or together with other nodes to implement one or more functionalities of any node in the communication network 700, including one or more network nodes 300 and / or core network nodes.

[0100] Examples of an ORAN network node include an open radio unit (O-RU), an open distributed unit (O- DU), an open central unit (O-CU), including an O-CU control plane (O-CU-CP) or an O-CU user plane (O-CU-UP), a RAN intelligent controller (near-real time or non-real time) hosting software or software plug-ins, such as a near-real time control application (e.g., xApp) or a non-real time control application (e.g., rApp), or any combination thereof (the adjective "open" designating support of an ORAN specification). The network node may support a specification by, for example, supporting an interface defined by the ORAN specification, such as an Al, Fl, Wl, El, E2, X2, Xn interface, an open fronthaul user plane interface, or an open fronthaul management plane interface. Moreover, an ORAN access node may be a logical node in a physical node. Furthermore, an ORAN network node may be implemented in a virtualization environment (described further below) in which one or more network functions are virtualized. For example, the virtualization environment may include an O-Cloud computing platform orchestrated by a Service Management and Orchestration (SMO) Framework via an 0-2 interface defined by the O-RAN Alliance or comparable technologies.

[0101] Each inner loop may be implemented as a separate r-app, hosted in the non-real time RIC 701, possibly from multiple vendors.

[0102] The meta-optimizer 710 may be implemented as yet another r-app. The different r-apps may communicate within the SMO and non-real time RIC part of O-RAN using the publish / subscribe model supported in the SMO and the R1 interface. Each inner loop may publish their inner loop data and meta-features and meta-gradients, and may subscribe to the meta optimizer 710 to receive the metaweights and context vectors. It is the other way around for the meta-optimizer 710.

[0103] The meta-optimizer 710 may be used to meta-optimize optimization tasks from different vendors as long as they subscribe and expose the information required by the meta optimizer 710.

[0104] The main O-RAN interfaces involved may be 01 and Al to communicate with the near-RT RIC 702.

[0105] In the example illustrated in figure 6, there are two optimization tasks 703, 704. The digital twin may be implemented using collector nodes, e.g. DT r-app A 720, DT r-app B 730. The optimizer nodes may be part of the non-RT RIC 701 as two r-apps for example, Optimizer A r-app 740 and Optimizer B r-app 750. The meta-optimizer 710 is illustrated as a separate node in the SMO for clarity, but it may also be implemented as an r-app within the non-RT RIC 701. The operations and signalling included in the optimization task A 703 and optimization task B 704 correspond to respective operations explained in the context of optimization tasks A and B 630, 640 of figure 5. Similarly, the remaining signalling messages illustrated in fig. 6 correspond to respective signalling in fig. 5, mutatis mutandis, therefore the detailed description of these steps will not be repeated here as it will be understood by the skilled person.

[0106] Advantageously, some embodiments allow for more efficient, flexible and adaptable optimization in the communication network that requires less compute power, less energy and less capital and operating expenditures. Advantageously, some embodiments enable avoiding expensive re-training of the models, particularly in the inner loops which are executed on a daily basis. Some embodiments reduce a number of iterations between the optimizer and the digital twin. Some embodiments allow for avoidance of catastrophic forgetting associated with prior warm starting the models. Some embodiments make the optimizer more adaptable to changes to the environment. Some embodiments allow reduction of the resource cost of running the optimization, particularly of inner loop and thus maximize OPEX savings when using the optimizers. Some embodiments allow improved lifecycle management that can be carried for the models of the meta-optimizer, which does not need to be done individually for each optimizer or digital twin-optimizer pair. Some embodiments improve performance of the optimizers or optimizer-digital twin pairs.

Claims

Claims1. A meta-optimizer node (270) for use in a communication network (250), the network comprising a plurality of optimizer nodes (212, 222, 232, 242), wherein respective nodes of the plurality of optimizer nodes are configured to perform, using a model (213, 223, 233, 243), an optimization task for the communication network (250), the meta-optimizer node (270) configured to: receive, from the plurality of optimizer nodes (212, 222, 232, 242), respective optimization task metadata; send, to an optimizer node of the plurality of optimizer nodes (212, 222, 232, 242) for input to the model (213, 223, 233, 243), a representation of the optimization tasks of the plurality of optimizer nodes, wherein the representation is generated based on the received task metadata.

2. A meta-optimizer node (270) according to claim 1, wherein the optimization task comprises updating of a network parameter for the communication network, and wherein the optimization task metadata comprises an indicator of a type of the network parameter.

3. A meta-optimizer node (270) according to claim 2, wherein each optimizer node of the plurality of optimizer nodes (212, 222, 232, 242) is paired with a respective digital twin (211, 221, 231, 241) of a subnetwork (251, 252) of the communication network (250), and wherein each pair of optimizer node and digital twin is configured to jointly perform the respective optimization task for the subnetwork by iteratively updating the network parameter for said subnetwork until a condition is satisfied.

4. A meta-optimizer node (270) according to claims 2 to 3, wherein the optimization task metadata comprises at least one indicator of a type of network operation data used by the corresponding optimizer node to evaluate an effect of updating the network parameter.

5. A meta-optimizer node (270) according to claim 4, wherein the at least one indicator comprises a network performance metric or network configuration information.

6. A meta-optimizer node (270) according to any preceding claim, wherein the representation comprises an embedding generated using an attention neural network.

7. A meta-optimizer node (270) according to any preceding claim, further configured to: receive, from the plurality of optimizer nodes (212, 222, 232, 242), respective model performance values; send, to the optimizer node of the plurality of optimizer nodes (212, 222, 232, 242), a model parameter value for updating the model of the respective optimizer node, wherein the model parameter value is generated based on the received optimization task metadata and model performance values.

8. A meta-optimizer node (270) according to claim 7, further comprising a neural network, wherein the model parameter value comprises an output of the neural network.

9. A meta-optimizer node (270) according to any of the claims 7 to 8 when dependent on claim 3, wherein the model (213, 223, 233, 243) comprises a gradient-based optimization model and the model performance values comprise a loss value or a gradient of a loss value.

10. A meta-optimizer node (270) according to any of the preceding claims 7 to 9, wherein the model parameter value is further based on respective task priority values determined for the respective optimization tasks.

11. A meta-optimizer node (270) according to claim 10, wherein the determination of the task priority values comprises learning the respective task priority values using a machine learning model based on the received optimization task metadata.

12. A meta-optimizer node (270) according to any preceding claim, wherein the optimization task of each of the plurality of optimizer nodes is the same.

13. A meta-optimizer node (270) according to any of the preceding claims 3 to 12, wherein the subnetwork (251, 252) comprises a radio access network, RAN, and wherein the network parameter comprises a RAN parameter.

14. A meta-optimizer node (270) according to any of the preceding claims 3 to 13, wherein the subnetwork (251, 252) comprises a plurality of network entities grouped according to a similarity criterion.

15. A meta-optimizer node (270) according to claim 14 when dependent on claim 13, wherein the plurality of network entities are cells of the RAN.

16. An optimizer node (212, 222, 232, 242) for a communication network (250), the optimizer node (212, 222, 232, 242) configured to perform an optimization task using a model (213, 223, 233, 243), the optimizer node (212, 222, 232, 242) configured to: send, to a meta-optimizer node (270), optimization task metadata; receive, from the meta-optimizer node (270), a representation of the optimization task and of at least a further optimization task performed by another optimizer node (212, 222, 232, 242); input the received representation into the model (213, 223, 233, 243).

17. An optimizer node (212, 222, 232, 242) according to claim 16, wherein the optimization task comprises updating of a network parameter for the communication network (250), and wherein the optimization task metadata comprises an indicator of a type of the network parameter.

18. An optimizer node (212, 222, 232, 242) according to claim 17, wherein the optimizer node (212, 222, 232, 242) is paired with a digital twin (211, 221, 231, 241) of a subnetwork (251, 252) of the communication network (250), and wherein the pair of the digital twin and optimizer node is configured to jointly perform the optimization task for the subnetwork by iteratively updating the network parameter for said subnetwork until a condition is satisfied.

19. An optimizer node (212, 222, 232, 242) according to any of the claims 17 to 18, wherein the optimization task metadata comprises at least one indicator of a type of network operation data used by the optimizer node to evaluate an effect of updating the network parameter.

20. An optimizer node (212, 222, 232, 242) according to claim 19, wherein the at least one indicator comprises a network performance metric or network configuration information.

21. An optimizer node (212, 222, 232, 242) according to claim 19 to 20 when dependent on claim 18, wherein the network operation data is generated by the digital twin (211, 221, 231, 241).

22. An optimizer node (212, 222, 232, 242) according to claims 19 to 21 when dependent on claim 18, wherein the network operation data is obtained directly from the subnetwork (251, 252).

23. An optimizer node (212, 222, 232, 242) according to any of the preceding claims 16 to 22, further configured to: send, to the meta-optimizer node (270), a model performance value; receive, from the meta-optimizer node (270), a model parameter value for updating the model (213, 223, 233, 243), wherein the model parameter value is generated based on the optimization task metadata and model performance value and based on further optimization task metadata and a further model performance value associated with another optimizer node (212, 222, 232, 242).

24. An optimizer node (212, 222, 232, 242) according to any of the preceding claims 16 to 23, wherein the optimization task and the further optimization task are the same.

25. An optimizer node (212, 222, 232, 242) according to claim 18, wherein the subnetwork (251, 252) comprises a radio access network, RAN, and wherein the network parameter comprises a RAN parameter.

26. An optimizer node (212, 222, 232, 242) according to claims 18 or 25, wherein the subnetwork comprises a plurality of network entities grouped according to a similarity criterion.

27. An optimizer node (212, 222, 232, 242) according to claim 26 when dependent on claim 25, wherein the plurality of network entities are cells of the RAN.

28. A communication network (200, 600, 700) comprising: a meta-optimizer node (271, 690, 710) according to any of the claims 1 to 15;a plurality of optimizer nodes (212, 222, 232, 242, 660, 680, 740, 750) according to any of the claims 16 to 27.

29. A method (800) of operating a meta-optimizer node for use in a communication network, the network comprising a plurality of optimizer nodes, wherein respective nodes of the plurality of optimizer nodes are configured to perform, using a model, an optimization task for the communication network, the method comprising the steps of: receiving (802), from the plurality of optimizer nodes, respective optimization task metadata; sending (804), to an optimizer node of the plurality of optimizer nodes for input to the model, a representation of the optimization tasks of the plurality of optimizer nodes, wherein the representation is generated based on the received task metadata.

30. A method of claim 29, further comprising steps of operating a meta-optimizer node according to any of the claims 2 to 15.

31. A method (900) of operating an optimizer node for a communication network, the optimizer node being configured to perform an optimization task using a model, the method comprising the steps of: sending (902), to a meta-optimizer node, optimization task metadata; receiving (904), from the meta-optimizer node, a representation of the optimization task and of at least a further optimization task performed by another optimizer node; inputting (906) the received representation into the model.

32. A method of claim 31, further comprising steps of operating an optimizer node according to any of the claims 16 to 27.

33. A computer program comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out a method according to any one of claims 29 to 32.

34. A computer program product comprising non transitory computer readable media having stored thereon a computer program according to claim 33.

Images