Determining overshooting cells of base stations based on machine learning
A system can determine an overshooting cell of a base station in a mobile network. In some implementations, the system can determine a plurality of tiles to cover a geographic area. The system can determine, from a user device, a radio measurement associated with a tile of the plurality of tiles. The radio measurement may be determined based on a signal, associated with a cell of the base station, which is received by the user device when the user device is located in the tile. The system can generate a multidimensional matrix, e.g., a 2D or 3D matrix, from the plurality of tiles and the radio measurement. The multidimensional matrix may indicate a footprint of the cell. The system can invoke a machine learning model to determine, based on the footprint, an overshoot by the cell in the geographic area. Other aspects are also described and claimed.
1 . A method for determining an overshooting cell of a base station in a mobile network, the method comprising:
determining a plurality of tiles to cover a geographic area;
determining, from a user device, a radio measurement associated with a tile of the plurality of tiles, the radio measurement determined based on a signal, associated with a cell of the base station in the geographic area, which is received by the user device when the user device is located in the tile;
generating a multidimensional matrix from the plurality of tiles and the radio measurement, the multidimensional matrix indicating a footprint of the cell;
transforming the multidimensional matrix to an image of the footprint; and
invoking a machine learning model comprising a trained image classifier to determine, based on the image, an overshoot or a normal operation by the cell in the geographic area.
2 . The method of claim 1 , further comprising:
training the machine learning model based on images of footprints from a plurality of cells in a plurality of base stations, wherein at least one of the footprints indicates the normal operation and another of the footprints indicates the overshoot.
3 . The method of claim 1 , wherein determining the plurality of tiles includes dividing the geographic area into equally sized regions that are adjacent to one another and include a coverage area for the base station.
4 . The method of claim 1 , wherein the multidimensional matrix includes i) a position of each tile of the plurality of tiles in the geographic area, and ii) an aggregated value of radio measurements associated with the tile.
5 . The method of claim 1 , wherein the footprint represents a coverage area where user devices can connect to signals from the cell.
6 . The method of claim 1 , wherein transforming the multidimensional matrix includes transforming into a larger matrix utilized by the machine learning model to determine the overshoot.
7 . The method of claim 1 , wherein transforming the multidimensional matrix includes resizing or reshaping the multidimensional matrix for compatibility with training of the machine learning model.
8 . The method of claim 1 , wherein the multidimensional matrix represents a grayscale or color image.
9 . The method of claim 1 , wherein the multidimensional matrix is a 3D matrix that includes different types of radio measurements in the plurality of tiles.
10 . The method of claim 1 , wherein the machine learning model comprises a convolutional neural network (CNN).
11 . The method of claim 1 , further comprising:
changing a configuration of the base station based on determining the overshoot.
12 . The method of claim 1 , further comprising:
determining, from the user device, a plurality of measurements of a same type corresponding to the tile, wherein the radio measurement is an aggregate of the plurality of measurements.
13 . The method of claim 1 , further comprising:
determining, from the user device, a plurality of measurements of different types corresponding to the tile, wherein the multidimensional matrix includes a plurality of aggregated values of measurement types in a dimension as different channels.
14 . The method of claim 1 , wherein the radio measurement is a type that indicates a power level, a signal quality, or a propagation delay of the signal to the user device.
15 . The method of claim 1 , further comprising:
determining a size of the tile based on at least one of a size of the geographic area or a density of base stations in the geographic area.
16 . The method of claim 1 , wherein the multidimensional matrix includes null values corresponding to tiles of the plurality of tiles that do not have radio measurements satisfying a filtering criterion.
17 . A non-transitory computer readable medium storing instructions operable to cause one or more processors to perform operations comprising:
determining a plurality of polygons to cover a geographic area;
determining, from a user device, a radio measurement associated with a polygon of the plurality of polygons, wherein the radio measurement is determined based on a signal, associated with a cell of a base station in the geographic area, which is received by the user device when the user device is located in the polygon;
generating a 3D matrix from the plurality of polygons and the radio measurement, the 3D matrix indicating a footprint of the cell; and
invoking a machine learning model to determine, based on the footprint, an overshoot by the cell in the geographic area.
18 . The non-transitory computer readable medium storing instructions of claim 17 , the operations further comprising:
receiving global positioning system (GPS) information from the user device that indicates a time and a location corresponding to the radio measurement.
19 . The non-transitory computer readable medium storing instructions of claim 17 , the operations further comprising:
determining, from the user device, a plurality of measurements corresponding to the polygon, wherein the radio measurement is a median of the plurality of measurements.
20 . The non-transitory computer readable medium storing instructions of claim 17 , the operations further comprising:
receiving a plurality of measurements from a plurality of user devices in the geographic area; and
generating the 3D matrix based on the plurality of measurements.