Conditional generative model recommendation for radio network

Patent metadata

Field	Value
Publication number	US-12574303-B2
Application number	US-202118029237-A
Country	US
Kind code	B2
Filing date	Oct 5, 2021
Priority date	Oct 6, 2020
Publication date	Mar 10, 2026
Grant date	Mar 10, 2026

Abstract

A method performed by a computing device for a radio network for configuration of a network device on which network or data energy can be collected while preserving specified conditions in the radio network is provided. The method includes receiving inputs to a conditional generative model. The inputs include the specified conditions in the radio network including a value for a predicted key performance indicator, KPI, constraint parameter for a time period, a target value for a optimization parameter, and a latent variable. The method further includes outputting from the conditional generative model a configuration data for a future time period for the network node or the cell of the radio network. The configuration data is bounded by the specified conditions including the predicted KPI constraint parameter, the target value for the optimization parameter, and the latent variable.

First claim

The invention claimed is:

1 . A method performed by a computing device for a radio network for configuration of a network device on which network or data energy can be collected while preserving specified conditions in the radio network, the method comprising:
receiving inputs to a conditional generative model, the inputs comprising the specified conditions in the radio network comprising a value for a predicted key performance indicator, KPI, constraint parameter for a time period, a target value for a optimization parameter, and a latent variable; and outputting from the conditional generative model a configuration data for a future time period for the network node or the cell of the radio network, wherein the configuration data is bounded by the specified conditions comprising the predicted KPI constraint parameter, the target value for the optimization parameter, and the latent variable.

2 . The method of claim 1 , wherein the latent variable comprises an encoded representation of a value corresponding to the optimization parameter and a value corresponding to the predicted KPI constraint parameter.

3 . The method of claim 1 , wherein the target value for the optimization parameter has an adjustable value from a plurality of variable values, wherein each variable value in the plurality of variable values represents a combination of a value of a category of a KPI from a plurality of the predicted KPI constraint parameters and a value of a optimization parameter associated with the category of the KPI from the plurality of the predicted KPI constraint parameters.

4 . The method of claim 1 , wherein the configuration data comprises a plurality of settings for each of a plurality of configuration management attributes and hardware, respectively, for the network node or the cell.

5 . The method of claim 1 , wherein the value for the predicted KPI constraint parameter and the target value for the optimization parameter are each received from a prediction model and wherein the prediction model comprises a model trained with a performance management dataset.

6 . The method of claim 5 , wherein the performance management dataset comprises a multivariate timeseries form of periodic performance management counters in the radio network.

7 . The method of claim 1 , further comprising:
training a configuration management model, wherein the training comprises inputting a set of configuration management attributes to the configuration management model and outputting from the configuration management model a value for the predicted KPI constraint parameter and a value for the optimization parameter; and training a performance management model, wherein the training comprises inputting a set of performance management data to the performance management model and outputting from the performance management model a prediction for the predicted KPI constraint parameter for the time period and a prediction for the optimization parameter for the time period.

8 . The method of claim 1 , further comprising:
training the conditional generative model with a configuration management dataset comprising a plurality of configuration management attributes and a corresponding plurality of conditional variables, wherein each of the plurality of configuration management attributes is associated with a corresponding conditional variable comprising the configuration management attributes and a corresponding quantized form of each of the predicted KPI constraint parameter and the target value for the optimization parameter, and wherein the training the conditional generative model comprises receiving as an input to the conditional generative model all configuration management attributes associated with the corresponding plurality of conditional variables, and outputting a synthetic configuration file for the network node based on the input.

9 . The method of claim 8 , wherein the configuration management dataset comprises data for a first granularity and wherein the set of performance management data comprises data for a second granularity, and further comprising:
applying inverse transformation on the configuration management dataset to distribute the configuration management dataset on a greater granularity level compared to the first granularity prior to applying the inverse transformation.

10 . The method of claim 1 , further comprising:
automatically tuning the conditional generative model, wherein the automatically tuning comprises: predicting from a configuration management model the target value using an original configuration data, predicting from the conditional generative model the target value using a generated configuration data, wherein the generated configuration data corresponds to the an original state of the radio network via the predicted KPI constraint parameter and the optimization parameter, obtaining a distance between the original configuration data and the generated configuration data, and reducing the distance.

11 . The method of claim 10 , wherein the reducing the distance comprises at least one of:
retuning the configuration management model and the conditional generative model, respectively; increasing a granularity of a discretization process for obtaining the quantized value of the predicted KPI constraint parameter and the quantized value for the optimization parameter; adjusting a standard deviation in the conditional generative model; and using a distance evaluation metric, wherein the distance evaluation metric comprises at least one of a KL divergence, a Euclidean distance, a coefficient of determination, and a cosine similarity.

12 . The method of claim 1 , further comprising:
obtaining the target value for the optimization parameter based on perturbating on the optimization parameter while maintaining the value for at least one predicted KPI constraint parameter.

13 . The method of claim 12 , wherein the perturbating comprises perturbate on the optimization input to a decoder of the conditional generative model parameter while maintaining the value for the at least one predicted KPI constraint to obtain generated configuration management data and predictions of corresponding KPI and energy constraints.

14 . The method of claim 1 , further comprising:
quantifying a performance result for the configuration data, wherein the quantifying comprises inputting the configuration data to a configuration management model, outputting from the configuration management model a value for the predicted KPI constraint parameter and a value for the optimization parameter.

15 . The method of claim 1 , wherein the optimization parameter is an energy consumption of the network node or the cell, wherein the target value for the optimization parameter comprises a value for the energy consumption of the network node or the cell that is lower than a value for the energy consumption from an embedded representation of configuration data comprising different values of the predicted KPI constraint parameter and different values of the energy consumption, and wherein the predicted KPI constraint parameter comprises at least one or more of a radio resource control connection setup success rate, a call setup success ratio per data, a cell throughput in uplink and downlink, a cell utilization, a traffic data volume in uplink and downlink, an intra and/or inter frequency handover success rate.

16 . The method of claim 1 , wherein the network device comprises a base station or a device for a cell.

17 . A computing device comprising a conditional generative model for c

Assignees

• Ericsson Telefon Ab L M

Inventors

• Ickin Selim
• Gorbatov Oleg
• Vandikas Konstantinos
• Eleftheriadis Lackis
• Nyström Cecilia
• Hallberg Helene

Classifications

• H04L41/16 Primary

  using machine learning or artificial intelligence · CPC title

• H04W24/04

  Arrangements for maintaining operational condition · CPC title

• G06N3/0475

  Generative networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/091

  Active learning · CPC title