Automating configuration management in cellular networks

Patent metadata

Field	Value
Publication number	US-12477353-B2
Application number	US-202218148365-A
Country	US
Kind code	B2
Filing date	Dec 29, 2022
Priority date	Dec 29, 2022
Publication date	Nov 18, 2025
Grant date	Nov 18, 2025

Abstract

Methods and apparatuses for automating configuration management in cellular networks. A method of a UE comprises: training, based on historical samples, a regression model y using samples obtained from a set of parameters, wherein the regression model y comprises a function of a first term X and a second term h; and predicting, based on the regression model y, a target KPI to capture parameter impacts corresponding to the second term h.

First claim

What is claimed is:

1 . A network entity in a communication system, the network entity comprising:
memory; and a processor operably connected to the memory, the processor configured to:
train, based on historical samples, a regression model y using samples obtained from a set of parameters, wherein the regression model y comprises a function of a first term X and a second term h, and
predict, based on the regression model y, a target key performance index (KPI) to capture parameter impacts corresponding to the second term h.

2 . The network entity of claim 1 , wherein the processor is further configured to:
determine whether each of samples is balanced among the set of parameters; and perform an oversampling operation based on a determination that each of samples is imbalanced.

3 . The network entity of claim 1 , wherein the first term X is a regressor that is a selected factor from a performance management (PM) counter and the second term h is a proxy variable, the first term X being a non-tunable variable.

4 . The network entity of claim 1 , wherein the processor is further configured to predict an output of the regression model y based on a long-term cell operation.

5 . The network entity of claim 1 , wherein the processor is further configured to:
identify a set of distinct parameters P i each of which corresponding to the second term h; identify a set of regression models y ij corresponding to each of the set of distinct parameters P i ; and compute a median based on the set of regression models y ij .

6 . The network entity of claim 1 , wherein the processor is further configured to:
select, based on a machine learning (ML) operation, a set of KPIs from data of performance management (PM); and remove, from the set of KPIs, a KPI including causality with a configuration management (CM) setting for training the regression model y.

7 . The network entity of claim 6 , wherein the processor is further configured to:
compute a Pearson correlation coefficient for the target KPI; and generate, based on the Pearson correlation coefficient, a ranking list including factors that are related to the target KPI.

8 . The network entity of claim 1 , wherein the processor is further configured to:
construct a classification model using a machine learning (ML) algorithm; determine whether configuration management (CM) is changed; determine whether a KPI is degraded based on a determination that the CM is changed; and perform, based on the classification model, a diagnosis to determine whether a parameter misconfiguration causes a degradation of the KPI.

9 . The network entity of claim 8 , wherein the processor is further configured to:
compute first anomaly samples after changing of the CM and second anomaly samples before changing of the CM; and generate, based on the first anomaly samples and the second anomaly samples, a relative metrics and absolute metrics.

10 . The network entity of claim 9 , wherein the processor is further configured to:
compute, based on the relative metrics and the absolute metrics, average metrics; generate, based on the average metrics, a mapping function; and generate, based on the mapping function, output labels using a threshold, the output labels comprising an unlikely, an undecided, or a likely.

11 . A method of a network entity in a communication system, the method comprising:
training, based on historical samples, a regression model y using samples obtained from a set of parameters, wherein the regression model y comprises a function of a first term X and a second term h; and predicting, based on the regression model y, a target key performance index (KPI) to capture parameter impacts corresponding to the second term h.

12 . The method of claim 11 , further comprising:
determining whether each of samples is balanced among the set of parameters; and performing an oversampling operation based on a determination that each of samples is imbalanced.

13 . The method of claim 11 , further comprising predicting an output of the regression model y based on a long-term cell operation, wherein the first term X is a regressor that is a selected factor from a performance management (PM) counter and the second term h is a proxy variable, the first term X being a non-tunable variable.

14 . The method of claim 11 , further comprising:
identifying a set of distinct parameters P i each of which corresponding to the second term h; identifying a set of regression models y ij corresponding to each of the set of distinct parameters P i ; and computing a median based on the set of regression models y ij .

15 . The method of claim 11 , further comprising:
selecting, based on a machine learning (ML) operation, a set of KPIs from data of performance management (PM); and removing, from the set of KPIs, a KPI including causality with a configuration management (CM) setting for training the regression model y.

16 . The method of claim 15 , further comprising:
computing a Pearson correlation coefficient for the target KPI; and generating, based on the Pearson correlation coefficient, a ranking list including factors that are related to the target KPI.

17 . The method of claim 11 , further comprising:
constructing a classification model using a machine learning (ML) algorithm; determining whether configuration management (CM) is changed; determining whether a KPI is degraded based on a determination that the CM is changed; and performing, based on the classification model, a diagnosis to determine whether a parameter misconfiguration causes a degradation of the KPI.

18 . The method of claim 17 , further comprising:
computing first anomaly samples after changing of the CM and second anomaly samples before changing of the CM; and generating, based on the first anomaly samples and the second anomaly samples, a relative metrics and absolute metrics.

19 . The method of claim 18 , further comprising:
computing, based on the relative metrics and the absolute metrics, average metrics; generating, based on the average metrics, a mapping function; and generating, based on the mapping function, output labels using a threshold, the output labels comprising an unlikely, an undecided, or a likely.

20 . A non-transitory computer-readable medium comprising program code, that when executed by at least one processor, causes a network entity to:
train, based on historical samples, a regression model y using samples obtained from a set of parameters, wherein the regression model y comprises a function of a first term X and a second term h, and predict, based on the regression model y, a target key performance index (KPI) to capture parameter impacts corresponding to the second term h.

Assignees

• Samsung Electronics Co Ltd

Inventors

• Xin Yan
• Kurt Mehmet N
• Albert Samuel
• Ford Russell
• Yagli Semih
• Zhang Jianzhong

Classifications

• H04W24/02

  Arrangements for optimising operational condition · CPC title

• H04W16/22 Primary

  Traffic simulation tools or models · CPC title