Granted Patent B2

US 12,507,038 · App. 17/851,805 · Granted Dec 23, 2025

Inventors: Antoine T. Tran (Issaquah, WA); Alexander Anh Tran (San Diego, CA)

Assignee: T-Mobile USA, Inc.

H04W4/029G06N20/00H04W4/021H04W24/08H04W24/10

Abstract

Claims (76)

1 . A computer-implemented method for dynamically generating an estimate of a real-time distribution density of wireless devices in a geographic area, the method comprising:

receiving real-time network activity data collected at multiple cell sites in a geographic area,

wherein the real-time network activity data indicates a real-time count of network activities including voice, data, or messaging activities by multiple wireless devices currently being served by the multiple cell sites in the geographic area,

wherein each of the multiple cell sites has a coverage area having a radius of at least 5 kilometers, and

wherein a size of the geographic area is approximately same as a combined network coverage area of the multiple cell sites; and

processing the real-time network activity data with a machine learning (ML) model that is trained based on non-real-time network activity data to predict distributions of wireless devices in the geographic area,

wherein the non-real-time network activity data indicate counts of voice, data, or messaging by a portion of wireless devices previously served at different points in time by the multiple cell sites in the geographic area and includes location data of the portion of wireless devices,

wherein the location data is based on one or more location sensors of each wireless device of the portion of wireless devices, and

wherein the ML model is configured to predict a real-time distribution of wireless devices in the geographic area at different points in time;

predicting, based on the ML model, a real-time distribution of the multiple wireless devices in the geographic area,

wherein the real-time distribution is predicted based on a pattern of network activities and locations of wireless devices at prior points in time that are analogous to a current point in time; and

dynamically generating an estimate of a distribution density of the multiple wireless devices in the geographic area for a particular point in time based on the predicted real-time distribution, adjusted by the real-time count of network activities;

subsequent to generating the estimate, generating an actual distribution density of wireless devices for the particular point in time using diagnostic metrics received from wireless devices located in the geographic area during the particular point in time; and

adjusting the ML model based on a comparison between the estimate of the distribution density and the actual distribution density.

2 . The method of claim 1 , wherein the real-time network activity data is collected at the multiple cell sites in the geographic area for a time interval that is less than or equal to a preset time period.

3 . The method of claim 1 , wherein the method further comprises:

designating the geographical area into multiple distinct and adjacent subregions,

each of the multiple subregions have a polygonal shape that includes four or more corners, and

predicting the real-time distribution of the multiple wireless devices in the geographic area includes predicting a number of wireless devices located within the respective multiple subregions in real-time.

4 . The method of claim 1 , wherein the non-real-time network activity data is collected periodically from the portion of wireless devices via mobile application instances operating on the portion of wireless devices.

5 . The method of claim 1 , wherein the real-time distribution of the wireless devices for a particular time point during a particular day of a week is predicted based on a pattern of network activities and locations of wireless devices at a prior time point that matches the particular time point during the particular day of the week.

6 . The method of claim 1 , wherein the prior points in time that are analogous to the current point in time are time intervals ranging from 5 minutes to 60 minutes.

7 . The method of claim 1 , wherein:

the estimate of the distribution density of the multiple wireless devices in the geographic area includes estimates for amounts of wireless devices for multiple adjacent subregions of the geographic area, and

a diameter of a respective subregion of the multiple adjacent subregions ranges from five meters to 100 meters.

8 . The method of claim 1 , wherein:

the estimate of the distribution density of the multiple wireless devices in the geographic area includes estimates for amounts of wireless devices for multiple distinct and adjacent subregions of the geographic area,

the multiple subregions have different sizes, and

a size of a respective subregion of the multiple subregions is predefined based on an expected population density of the respective subregion.

9 . The method of claim 1 , further comprising:

estimating a non-real-time network activity data of all wireless devices previously served at the different points in time by the multiple cell sites in the geographic area based on the non-real-time network activity data of the portion of wireless devices.

10 . The method of claim 1 ,

wherein the non-real-time network activity data comprises historical diagnostics data collected by mobile application instances operating on the portion of wireless devices, and

wherein users of the portion of wireless devices have authorized the collection of the diagnostics data.

11 . A computer system for dynamically generating an estimate of a real-time distribution density of wireless devices in a geographic area, the system comprising:

at least one hardware processor, and

at least one non-transitory memory storing instructions, which, when executed by the at least one hardware processor, cause the system to:

receive real-time network activity data collected at multiple cell sites in a geographic area,

wherein each of the multiple cell sites has a coverage area having a radius of at least 5 kilometers, and

wherein a size of the geographic area is approximately same as a combined network coverage area of the multiple cell sites;

process the real-time network activity data with a machine learning (ML) model that is trained based on non-real-time network activity data to predict distributions of wireless devices on the geographic area,

wherein the non-real-time network activity data indicates counts of network activities by wireless devices previously served at different time intervals by the multiple cell sites in the geographic area and includes location data of the wireless devices,

wherein the location data is based on one or more location sensors of each of the wireless devices, and

wherein the ML model is configured to predict a real-time distribution of wireless devices on the geographic area at different time intervals;

predict, based on the ML model, a real-time distribution of the multiple wireless devices in the geographic area,

wherein the real-time distribution is predicted based on a pattern of network activities and locations of wireless devices at prior time intervals that are analogous to a current time interval;

dynamically generating an estimate of a distribution density of the multiple wireless devices on the geographic area for a particular point in time based on the predicted real-time distribution, adjusted by the real-time count of network activities;

subsequent to generating the estimate, generating an actual distribution density of wireless devices for the particular point in time using diagnostic metrics received from wireless devices located in the geographic area during the particular point in time; and

adjusting the ML model based on a comparison between the estimate of the distribution density and the actual distribution density.

12 . The system of claim 11 , wherein the network activity data include voice, data, or messaging activities.

13 . The system of claim 11 , wherein the prior time intervals that are analogous to the current time interval range from 5 minutes to 60 minutes.

14 . The system of claim 11 , wherein the real-time distribution of the wireless devices for a particular time point during a particular day of a week is predicted based on a pattern of network activities and locations of wireless devices at a prior time point that matches the particular time point during the particular day of the week.

15 . The system of claim 11 , wherein:

the estimate of the distribution density of the multiple wireless devices in the geographic area includes estimates for amounts of wireless devices for multiple adjacent subregions of the geographic area, and

a diameter of a respective subregion of the multiple adjacent subregions ranges from five meters to 100 meters.

16 . The system of claim 11 , wherein:

the estimate of the distribution density of the multiple wireless devices in the geographic area includes estimates for amounts of wireless devices for multiple distinct and adjacent subregions of the geographic area,

the multiple subregions have different sizes, and

a size of a respective subregion of the multiple subregions is predefined based on an expected population density of the respective subregion.

17 . A computer-implemented method for training a machine learning (ML) classifier for estimation of a real-time distribution density of wireless devices in a geographic area, the method comprising:

designating a geographic area into multiple subregions,

wherein the geographic area is included in a network coverage area of multiple cell sites,

wherein each of the multiple cell sites has a coverage area having a radius of at least 5 kilometers, and

wherein a size of the geographic area is approximately same as a combined network coverage area of the multiple cell sites;

collecting a training data set including non-real-time network activity data indicating counts of voice, data, or messaging activities by wireless devices previously served at different points in time by the multiple cell sites in the geographic area and includes location data of the wireless devices,

wherein the location data is based on one or more location sensors of each of the wireless devices;

training an untrained ML classifier with the training data set to associate the non-real-time network activity of the wireless devices at the different points in time with respective subregions of the geographic area based on the location data of the wireless devices thereby obtaining a trained ML classifier that estimates a real-time distribution of the multiple wireless devices across the subregions of the geographic area;

generating an estimate of a distribution density of the multiple wireless devices of the geographic area for a particular point in time using the trained ML;

subsequent to generating the estimate, generating an actual distribution density of wireless devices for the particular point in time using diagnostic metrics received from wireless devices located in the geographic area during the particular point in time; and

adjusting the trained ML model based on a comparison between the estimate of the distribution density and the actual distribution density.

18 . The method of claim 17 , wherein the multiple subregions are distinct from and adjacent to each other and the multiple subregions have a polygonal shape that includes four or more corners.

19 . The method of claim 17 , wherein:

the multiple subregions have different sizes, and

a size of a respective subregion of the multiple subregions is predefined based on an expected population density of the respective subregion.

20 . The method of claim 17 , wherein a diameter of a respective subregion of the multiple subregions ranges from five meters to 100 meters.

21 . The method of claim 17 , wherein collecting the training data set includes collecting the non-real-time network activity data periodically for a preset time interval during a particular time period.

Assignments (1)

Continuity (1)

References Cited (107)