Cellular Throughput Prediction via KPI Quantile Summarization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cellular networks face challenges in predicting throughput due to fluctuating radio channel conditions and varying cell load, especially with increasing mobile device traffic for video and interactive applications, which affects network performance and user experience.
Innovation Solution
A system using machine learning algorithms that leverage Key Performance Indicators (KPIs) from both device and network environments to predict future throughput by summarizing historical data, improving prediction accuracy through statistical methods and increasing the history length and prediction horizon.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional throughput prediction methods are used, then the system is simple to implement, but prediction accuracy is low due to fluctuating radio channel conditions and varying cell load
Solution Approach 1:
The patent introduces Key Performance Indicators (KPIs) as intermediary variables that mediate between the complex fluctuating radio channel conditions and the throughput prediction. These KPIs serve as simplified representations of network state that can be processed by machine learning algorithms to improve prediction accuracy without requiring direct analysis of all underlying physical layer parameters.
Solution Approach 2:
The system performs preliminary summarization of historical KPI data before feeding it to prediction algorithms. By pre-processing and aggregating historical performance data into meaningful statistical features, the system prepares optimized input data that enhances prediction accuracy while reducing the computational complexity during the actual prediction phase.
2Measurement precision
If historical data length is increased to improve prediction accuracy, then prediction errors decrease, but data processing time and computational resources increase
Solution Approach 1:
The patent extracts only the most relevant features from historical KPI data rather than processing all available data points. By identifying and extracting key statistical features and patterns from historical data, the system maintains high prediction accuracy while significantly reducing the computational burden and processing time required for analysis.
3Measurement precision
If quantile summarization is used to reduce prediction errors, then the 90th percentile of absolute error decreases from 51% to 22%, but the computational complexity of data processing increases
Solution Approach 1:
The patent transforms the distribution of historical KPI data by applying quantile summarization, which changes the parameter representation from raw values to quantile-based statistics. This transformation reduces prediction errors by better capturing the distribution characteristics of network performance data, while the computational complexity increase is managed through efficient algorithms.
4Reliability
If network-based KPIs are incorporated to improve accuracy by 21%, then prediction reliability increases, but the system complexity and data collection requirements increase
Solution Approach 1:
The patent merges device-based KPIs with network-based KPIs to create a comprehensive prediction system. By combining data from both device and network perspectives, the system achieves improved reliability and accuracy. The merging is accomplished through a unified machine learning framework that processes both data sources together, managing system complexity through integrated architecture rather than separate processing systems.
Data Source
AI summary
Aspects of the subject disclosure may include, for example, a method in which a processing system identifies a plurality of performance indicators comprising device performance indicators for a plurality of communication devices on a cellular network and network performance indicators for the cellular network. The method also includes obtaining historical data regarding the plurality of performance indicators for each of a series of time points during a past time period; the historical data for each of the plurality of performance indicators form an array of values for that performance indicator. The method further includes generating from each array a set of inputs to an algorithm for predicting a throughput of the cellular network during a future time period; the set of inputs comprises quantiles of the array, and the algorithm comprises a machine learning algorithm. Other embodiments are disclosed.