Optimizing cellular networks using deep learning

1. A computer-implemented method for optimizing a cellular network having a plurality of cells, comprising:

selecting a group of cells comprising a cell identified as underperforming and one or more neighboring cells;

ranking the one or more neighboring cells based on one or more relationship parameters between the underperforming cell and the one or more neighboring cells;

generating a state tensor for the group of cells based on the ranking of the one or more neighboring cells, the state tensor reflecting operational communication conditions within the group of cells in the cellular network;

processing the state tensor to produce one or more recommendations for adjusting one or more parameters of the cellular network; and

adjusting the one or more parameters of the cellular network based on the one or more recommendations.

2. The computer-implemented method of claim 1 , wherein processing the state tensor to produce one or more recommendations for adjusting one or more parameters of the cellular network comprises:

submitting the state tensor as input into a deep neural network; and

producing one or more recommendations, using the deep neural network, for adjusting one or more parameters of the cellular network.

3. The computer-implemented method of claim 2 , wherein the deep neural network is a convolutional neural network.

4. The computer-implemented method of claim 2 , wherein the producing one or more recommendations, using the deep neural network, for adjusting one or more parameters of the cellular network comprises:

applying the solution to the cellular network; and

determining whether to accept the one or more recommendations based on an impact of the one or more recommendations on the cellular network, wherein the impact is measured by a change in the overall coverage and capacity of the cellular network.

5. The computer-implemented method of claim 4 , further comprising:

rejecting the one or more recommendations in response to determining that the impact of the one or more recommendations negatively affect the overall coverage and capacity of the cellular network;

accepting the one or more recommendations in response to determining that the impact of the one or more recommendations positively affect the overall coverage and capacity of the cellular network; and

updating the deep neural network based on whether the one or more recommendations were accepted.

6. The computer-implemented method of claim 1 , wherein the cell is identified as underperforming when one or more cell performance measurements exceed one or more thresholds.

7. The computer-implemented method of claim 6 , wherein the one or more cell performance measurements comprise an objective function value that is calculated based on a relationship between signal power and signal quality.

8. The computer-implemented method of claim 1 , wherein the one or more relationship parameters comprise an interference level between the underperforming cell and the one or more neighboring cells.

9. The computer-implemented method of claim 8 , wherein the interference level is a value that is averaged over a period of time.

10. The computer-implemented method of claim 1 , wherein the generating a state tensor for the group of cells based on the ranking of the one or more neighboring cells comprises:

selecting a size for a base grid;

constructing the base grid of the selected size, wherein cells in the group of cells are placed in a position on both a horizontal axis and a vertical axis of the base grid;

extracting multiple channels of information for one or more pairs of cells, each of the one or more pairs of cells comprising a cell from the horizontal axis and a cell from the vertical axis in the base grid; and

constructing the state tensor based on the base grid and the extracted multiple channels of information.

11. The computer-implemented method of claim 10 , wherein the constructing a base grid comprises:

placing the underperforming cell in the base grid at the center of the horizontal axis and the vertical axis; and

arranging the one or more neighboring cells in the base grid such that higher ranking neighboring cells are placed closer to the underperforming cell on the horizontal axis and the vertical axis.

12. The computer-implemented method of claim 1 , wherein the group of cells and one or more additional groups of cells are each assigned a priority.

13. The computer-implemented method of claim 12 , wherein the priorities of the group of cells and one or more additional group of cells is calculated based on one or more of the severity, preference, weight, and neighbor relationships associated with the underperforming cell in the group of cells.

14. The computer-implemented method of claim 12 , wherein the group of cells and the one or more additional groups of cells are optimized in an order of a highest to a lowest priority.

15. The computer-implemented method of claim 12 , wherein the group of cells and at least one of the one or more additional groups of cells having the highest priority are optimized in parallel.

16. A cellular network optimization device, comprising:

a non-transitory memory storage comprising instructions; and

one or more processors in communication with the memory, wherein the one or more processors are configured to execute the instructions to:

select a group of cells comprising a cell identified as underperforming and one or more neighboring cells;

rank the one or more neighboring cells based on one or more relationship parameters between the underperforming cell and the one or more neighboring cells;

generate a state tensor for the group of cells based on the ranking of the one or more neighboring cells, the state tensor reflecting operational communication conditions within the group of cells in the cellular network;

process the state tensor to produce one or more recommendations for adjusting one or more parameters of the cellular network; and

adjust the one or more parameters of the cellular network based on the one or more recommendations.

17. The cellular network optimization device of claim 16 wherein the one or more processors are further configured to execute the instructions to:

submit the state tensor as input into a deep neural network; and

produce one or more recommendations, using the deep neural network, for adjusting one or more parameters of the cellular network.

18. The cellular network optimization device of claim 16 , wherein the step to generate a state tensor for the group of cells based on the ranking of the one or more neighboring cells comprises:

select a size for a base grid;

construct the base grid of the selected size, wherein cells in the group of cells are placed in a position on both a horizontal axis and a vertical axis of the base grid;

extract multiple channels of information for one or more pairs of cells, each of the one or more pairs of cells comprising a cell from the horizontal axis and a cell from the vertical axis in the base grid; and

construct the state tensor based on the base grid and the extracted multiple channels of information.

19. The cellular network optimization device of claim 18 , wherein the construct a base grid comprises:

place the underperforming cell in the base grid at the center of the horizontal axis and the vertical axis; and

arrange the one or more neighboring cells in the base grid such that higher ranking neighboring cells are placed closer to the underperforming cell on the horizontal axis and the vertical axis.

20. A non-transitory computer-readable medium storing computer instructions for optimizing a cellular network having one or more cells that when executed by one or more processors, cause the one or more processors to perform the steps of:

select a group of cells comprising a cell identified as underperforming and one or more neighboring cells;

rank the one or more neighboring cells based on one or more relationship parameters between the underperforming cell and the one or more neighboring cells;

generate a state tensor for the group of cells based on the ranking of the one or more neighboring cells, the state tensor reflecting operational communication conditions within the group of cells in the cellular network;

process the state tensor to produce one or more recommendations for adjusting one or more parameters of the cellular network; and

adjust the one or more parameters of the cellular network based on the one or more recommendations.