Granted Patent B2

US 12,659,241 · App. 18/495,032 · Granted Jun 16, 2026

Inventors: Oliver Coudert (Arlington, VA); Durga Prasad Satapathy (Ashburn, VA); Javed Rahman (Leesburg, VA)

Assignee: T-MOBILE INNOVATIONS LLC

H04L41/5009H04L41/16

Abstract

Claims (68)

1 . A method of operating a wireless communication network to anonymize network Key Performance Indicators (KPIs) to train a machine learning model, the method comprising:

retrieving the network KPIs generated by network KPI sources;

filtering the network KPIs based on an intended function of the machine learning model;

identifying correlated ones of the filtered KPIs and grouping the correlated ones of the filtered KPIs into KPI groups based on at least one network condition;

for each KPI group, sorting the filtered KPIs into KPI ranges;

for each KPI range, tokenizing the KPI range by converting numeric values of the filtered KPIs that compose the KPI range into strings to generate tokenized KPIs; and

storing the tokenized KPIs in a KPI database accessible by the machine learning model for training.

2 . The method of claim 1 wherein retrieving the network KPIs generated by the network KPI sources comprises retrieving the network KPIs from Radio Access Networks (RANs), control plane network functions, and user plane network functions.

3 . The method of claim 1 wherein:

the intended function of the machine learning model is to generate synthetic KPI streams comprising a set of synthetic KPIs; and

filtering the network KPIs based on the intended function of the machine learning model comprises selecting the network KPIs that correspond to the set of synthetic KPIs.

4 . The method of claim 1 wherein identifying the correlated ones of the filtered KPIs comprises processing each of the filtered KPIs using a Dynamic Time Warping (DTW) algorithm to measure similarities between each of the filtered KPIs.

5 . The method of claim 1 wherein:

the at least one network condition comprises one or more of a Radio Access Network (RAN) configuration, a Radio Access Technology (RAT) type, a network loading condition, a geographic location type, a network behavior, or a time period; and

grouping the correlated ones of the filtered KPIs into the KPI groups based on at least one network condition comprises grouping the correlated ones of the filtered KPIs into the KPI groups based on at least one of the RAN configuration, the RAT type, the network loading condition, the geographic location type, the network behavior, or the time period.

6 . The method of claim 1 wherein sorting the filtered KPIs into the KPI ranges comprises:

determining a numeric distribution of the numeric values of the filtered KPIs,

selecting a range size for the filtered KPIs based on the numeric distribution;

selecting the KPI ranges based on the range size; and

grouping ones of the filtered KPIs based on the KPI ranges.

7 . The method of claim 1 wherein tokenizing the KPI range by converting the numeric values of the filtered KPIs that compose the KPI range into the strings to generate the tokenized KPIs comprises:

determining an amount of the filtered KPIs that compose the KPI range;

selecting a word to represent the KPI range; and

generating the tokenized KPIs by replacing the numeric values of the filtered KPIs in the KPI range with the word.

8 . The method of claim 1 wherein:

the machine learning model comprises a Large Language Model (LLM); and further comprising:

retrieving the tokenized KPIs from the KPI database and training the LLM using the tokenized KPIs to generate synthetic KPI streams comprising a set of synthetic KPIs.

9 . The method of claim 1 wherein the network KPIs comprise one or more of downlink traffic volume, call count, average downlink throughput, access failure rate, call drop rate, call drop count, Physical Resource Block (PRB) utilization, average number of Radio Resource Control (RRC) connected users, packet loss rate, average Received Signal Received Quality (RSRQ), average Received Signal Received Power (RSRP), Random Access Channel (RACH) success rate, or average Singal-to-Interference plus Noise Ratio (SINR).

10 . A wireless communication network to anonymize network Key Performance Indicators (KPIs) to train a machine learning model, the wireless communication network comprising:

a network analytics system configured to:

retrieve the network KPIs generated by network KPI sources;

filter the network KPIs based on an intended function of the machine learning model;

identify correlated ones of the filtered KPIs and group the correlated ones of the filtered KPIs into KPI groups based on at least one network condition;

for each KPI group, sort the filtered KPIs into KPI ranges; and

for each KPI range, tokenize the KPI range by converting numeric values of the filtered KPIs that compose the KPI range into strings to generate tokenized KPIs; and

a KPI store configured to:

store the tokenized KPIs in a KPI database accessible by the machine learning model for training.

11 . The wireless communication network of claim 10 wherein the network analytics system is configured to retrieve the network KPIs from Radio Access Networks (RANs), control plane network functions, and user plane network functions to retrieve the network KPIs generated by the network KPI sources.

12 . The wireless communication network of claim 10 wherein:

the intended function of the machine learning model is to generate synthetic KPI streams comprising a set of synthetic KPIs; and

the network analytics system is configured to select the network KPIs that correspond to the set of synthetic KPIs to filter the network KPIs.

13 . The wireless communication network of claim 10 wherein the network analytics system is configured to process each of the filtered KPIs using a Dynamic Time Warping (DTW) algorithm to measure similarities between each of the filtered KPIs to identify the correlated ones of the filtered KPIs.

14 . The wireless communication network of claim 10 wherein:

the at least one network condition comprises one or more of a Radio Access Network (RAN) configuration, a Radio Access Technology (RAT) type, a network loading condition, a geographic location type, a network behavior, or a time period; and

the network analytics system is configured to group the correlated ones of the filtered KPIs into the KPI groups based on at least one of the RAN configuration, the RAT type, the network loading condition, the geographic location type, the network behavior, or the time period.

15 . The wireless communication network of claim 10 wherein the network analytics system is further configured to:

determine a numeric distribution of the numeric values of the filtered KPIs,

select a range size for the filtered KPIs based on the numeric distribution;

select the KPI ranges based on the range size; and

group ones of the filtered KPIs based on the KPI ranges to sort the filtered KPIs into the KPI ranges.

16 . The wireless communication network of claim 10 wherein the network analytics system is further configured to:

determine an amount of the filtered KPIs that compose the KPI range;

select a word to represent the KPI range; and

generate the tokenized KPIs by replacing the numeric values of the filtered KPIs in the KPI range with the word.

17 . The wireless communication network of claim 10 wherein:

the machine learning model comprises a Large Language Model (LLM); and

the LLM is configured to retrieve the tokenized KPIs from the KPI database and train its constituent machine learning algorithms using the tokenized KPIs to generate synthetic KPI streams comprising a set of synthetic KPIs.

18 . The wireless communication network of claim 10 wherein the network KPIs comprise one or more of downlink traffic volume, call count, average downlink throughput, access failure rate, call drop rate, call drop count, Physical Resource Block (PRB) utilization, average number of Radio Resource Control (RRC) connected users, packet loss rate, average Received Signal Received Quality (RSRQ), average Received Signal Received Power (RSRP), Random Access Channel (RACH) success rate, or average Singal-to-Interference plus Noise Ratio (SINR).

19 . A wireless communication network to anonymize network Key Performance Indicators (KPIs) to train a machine learning model, the wireless communication network comprising:

a Network Data Analytics Function (NWDAF) configured to:

retrieve the network KPIs generated by network KPI sources;

filter the network KPIs based on an intended function of the machine learning model;

identify correlated ones of the filtered KPIs and group the correlated ones of the filtered KPIs into KPI groups based on at least one network condition;

for each KPI group, sort the filtered KPIs into KPI ranges;

for each KPI range, tokenize the KPI range by converting numeric values of the filtered KPIs that compose the KPI range into strings to generate tokenized KPIs; and

an Analytics Data Repository Function (ADRF) configured to:

store the tokenized KPIs in a KPI database accessible by the machine learning model for training.

20 . The wireless communication network of claim 19 wherein the network KPIs comprise one or more of downlink traffic volume, call count, average downlink throughput, access failure rate, call drop rate, call drop count, Physical Resource Block (PRB) utilization, Radio Resource Control (RRC) connected users count, packet loss rate, average Received Signal Received Quality (RSRQ), average Received Signal Received Power (RSRP), Random Access Channel (RACH) success rate, or average Singal-to-Interference plus Noise Ratio (SINR).

Assignments (1)

Continuity (1)

References Cited (21)