An intent decomposition method includes sending, by a first network device, a first sub-intent target value to a second network device. The method also includes receiving, by the first network device, a first message from the second network device. The first message is used to indicate that the first sub-intent target value is not achieved by the second device. The first message includes a first measurement value of the second network device. The method further includes re-decomposing, by the first network device, an intent based on the first measurement value, and sending first sub-intent target values obtained through re-decomposition to the second network device and one or more other network devices different from the first network device and the second network device until all the first sub-intent target values obtained through re-decomposition are achieved or none of the first sub-intent target values obtained through re-decomposition are achieved.

Embodiments are directed towards embodiments are directed toward systems and methods for user plane function (UPF) and network slice load balancing within a 5G network. Example embodiments include systems and methods for load balancing based on current UPF load and thresholds that depend on UPF capacity; UPF load balancing using predicted throughput of new UE on the network based on network data analytics; UPF load balancing based on special considerations for low latency traffic; UPF load balancing supporting multiple slices, maintaining several load-thresholds for each UPF and each slice depending on the UPF and network slice capacity; and UPF load balancing using predicted central processing unit (CPU) utilization and/or predicted memory utilization of new UE on the network based on network data analytics.

This disclosure describes methods and systems for exchanging AI computing information for load prediction model between network elements of a wireless communication network. The methods include: sending, by a first network element of a wireless communication network, a first message for load prediction to a second network element of the wireless communication network, wherein the first message comprises at least one of an input to a machine learning model for load prediction of the first network element or model configuration information of the machine learning model.

Embodiments are directed towards systems and methods for user plane function (UPF) and network slice load balancing within a 5G network. Example embodiments include systems and methods for load balancing based on current UPF load and thresholds that depend on UPF capacity; UPF load balancing using predicted throughput of new UE on the network based on network data analytics; UPF load balancing based on special considerations for low latency traffic; UPF load balancing supporting multiple slices, maintaining several load-thresholds for each UPF and each slice depending on the UPF and network slice capacity; and UPF load balancing using predicted central processing unit (CPU) utilization and/or predicted memory utilization of new UE on the network based on network data analytics.

Software-defined networking (SDN) can be utilized with a wireless network platform to increase efficiencies and mitigate service lapses. Within an SDN enabled on-demand dynamic 5G network slice management architecture, the SDN can be utilized for on the-fly deployment of network slicing. For example, an SDN network slice broker (SNSB), can facilitate an on-demand allocation of network resources performing admission control, resource negotiation, and charging. Additionally, the system can comprise an SDN-enabled edge slice mobile edge computing (MEC) coordinator and a local slice MEC agent. Thus, the SDN facilitate on-demand alternate paths, by utilizing the SDN-enabled edge slice MEC coordinator and the local slice MEC agents at various slices.

A method performed by a method performed by a network node for performing admission control in a wireless communications network is provided. The network node serves a first and one or more second UEs. The network node receives (201) from the first UE, an access request for a radio resource for communication between the first UE and the network node. The network node further estimates (203) a first prediction of a requirement of the radio resource related to the access request, based on a measured initial data traffic between the network node and the first UE. The network node determines (205) a first threshold related to the first prediction, as a function of a measured data traffic load between the network node and the one or more second UEs. The network node then performs (206) admission control by deciding whether or not to admit the radio resource to the first UE, based on whether or not the first prediction exceeds the first threshold.

Aspects of the present disclosure relate to allocating RAN resources among RAN slices using a machine learning model. In examples, the machine learning model may determine an optimal RAN resource configuration based on compute power needs. As a result, RAN resource allocation generation and compute power requirements may improve, even in instances with changing or unknown network conditions. In examples, a prediction engine may receive communication parameters and/or requirements associated with service-level agreements (SLAs) for applications executing at least partially at a device in communication with the RAN. The RAN may generate one or more RAN resource configuration for implementation among RAN slices. Upon a change in network conditions or SLA requirements, an optimal RAN configuration may be determined in terms of required compute power.

A method for managing a first node in a communication network is disclosed, wherein the first node is operable to exchange traffic flows with other nodes in the communication network. The method includes using a Variational Autoencoder (VAE) to generate a predicted traffic distribution for the first node, wherein the VAE has been trained using information about historical data flows exchanged by the first node with at least one other node in the communication network, and configuring at least one radio resource parameter of the first node based on the obtained predicted traffic distribution for the first node. Also disclosed are a method including training a VAE and nodes and a computer program product suitable for carrying out such methods.

Selecting data packages received by a host device is provided. The host device receives a number of N data packages. For each data package of the N data packages: a prioritized data package parameter is calculated. The data packages are sorted according to a predetermined sorting-scheme considering at least prioritized data package parameters. A number of M data packages are down-selected from the N data packages, wherein M < N. The one or more of the down-selected M data packages are processed. Accordingly, hardware and software requirements of the host device may be reduced due to a reduced computational complexity.

Embodiments are directed towards systems and methods for selecting, in a Fifth Generation (5G) cellular telecommunication network, a User Plane Function (UPF) of a plurality of UPFs on which to anchor a Protocol Data Unit (PDU) session of a new user equipment (UE) newly appearing on the cellular telecommunication network. The selection is based on: a location of the new UE; a plurality of current loads for each UPF of the plurality of UPFs; a predicted UE load of the new UE based on network data analytics; and predicted UPF loads of the plurality of UPFs as a function of time considering the predicted UE load based on network data analytics from the Network Data Analytics Function. In the UPF selection, the Session Management Function (SMF) gives higher priority to shorter term predicted loads than longer term predicted loads. Also, in the UPF selection, the PDU session of the UE is preferred to attach on the UPF in the current serving area in which the UE is located.