Machine-Learning Cellular Coverage Planning From Radiation Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for computing cellular antenna coverage, particularly in 5G networks, are computationally demanding and time-consuming, making it challenging to efficiently plan and optimize large-scale cellular network deployments due to the complex interaction of antenna parameters and geospatial obstacles.
Innovation Solution
Employing machine learning models, such as neural networks, to estimate cellular antenna coverage and interference by transforming radiation patterns into signal strength arrays and augmenting them with antenna and environmental parameters, significantly reducing computation time from seconds to milliseconds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional methods are used to compute cellular antenna coverage, then coverage computation accuracy is maintained, but computation time becomes excessively long (seconds to minutes)
Solution Approach 1:
The patent creates a machine learning model that learns from pre-computed coverage data and environmental factors to predict coverage patterns. The model copies the essential coverage characteristics without performing exhaustive traditional computations, reducing time from seconds to milliseconds while maintaining accuracy through trained predictions rather than raw calculation.
Solution Approach 2:
The patent transforms the coverage computation problem by changing the input parameters from traditional physical propagation models to machine learning model inputs including radiation patterns, antenna parameters, and environmental features. This parameter transformation enables rapid prediction through trained neural networks instead of time-consuming physical simulations.
2Productivity
If traditional coverage computation methods are used, then computational accuracy is maintained, but the complexity and time required for network planning increases
Solution Approach 1:
The patent replaces traditional mechanical computation systems with machine learning inference systems. Instead of executing complex propagation models and ray-tracing algorithms that demand significant computational resources, the system uses trained neural networks that perform rapid pattern recognition and prediction, substituting heavy computational mechanics with lightweight AI inference.
Solution Approach 2:
The patent performs preliminary training of the machine learning model using extensive pre-computed data and environmental models. Once trained, the model can rapidly predict coverage for new scenarios without re-executing the complex computational processes, effectively performing the heavy lifting in advance during the training phase rather than during actual network planning.
Data Source
AI summary
In one example, a method performed by a processing system including at least one processor includes creating a geospatial model of an environment in which a cellular network is to be deployed, transforming, for each cellular antenna of a proposed antenna layout of the cellular network, a radiation pattern of the each cellular antenna into a signal strength array, to create a plurality of signal strength arrays, augmenting, for each signal strength array of the plurality of signal strength arrays, the each signal strength array with at least one parameter of a corresponding cellular antenna of the proposed antenna layout and at least one value describing the environment in which the cellular network is to be deployed, and estimating a coverage of the proposed antenna layout based on the signal strength array, as augmented, using a machine learning model.