Wireless Environment Impact Detection for Application Performance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wireless communications applications struggle to detect and mitigate performance issues such as stuttering audio or reduced quality, which are difficult to identify despite active network connections, as these issues often stem from multivariate factors in the wireless environment.
Innovation Solution
A computing device monitors wireless data elements, extracts feature values, and uses a trained classifier, such as a neural network, to predict performance impacts on applications like voice calls and video streaming, enabling detection and mitigation of these issues independently of the specific application being run.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If application-specific performance monitoring is implemented, then detection accuracy for application-level issues is improved, but system complexity and implementation cost increase
Solution Approach 1:
The patent introduces a network-side performance predictor as an intermediary component that monitors wireless environment parameters and predicts application performance impacts without requiring direct integration with applications. This mediator approach allows accurate detection of performance issues like audio stuttering and video quality degradation while maintaining system simplicity, as the predictor operates independently at the network level using extracted feature values from wireless data elements.
Solution Approach 2:
The performance predictor is designed as a universal system that can monitor and predict performance impacts across multiple different applications (voice calls, video streaming, etc.) using a single unified architecture. By extracting common feature values from wireless data elements and using a general classification model, the system achieves multi-functional capability without requiring application-specific implementations, thus reducing overall system complexity.
2Loss of time
If real-time performance monitoring is implemented, then response time for mitigation is improved, but energy consumption increases
Solution Approach 1:
The system performs preliminary monitoring and prediction of wireless environment parameters continuously, building up feature values and training models in advance. By pre-extracting features from wireless data elements and pre-training classification models on historical data, the system can quickly predict performance impacts when they occur without requiring intensive real-time processing, thus reducing energy consumption while maintaining fast response capability.
Solution Approach 2:
The performance predictor operates autonomously at the network level, self-monitoring wireless environment parameters and self-predicting application performance impacts without requiring active participation or energy consumption from the applications themselves. The system serves applications by providing performance predictions on demand, eliminating the need for applications to continuously monitor their own performance, thereby significantly reducing overall energy consumption.
3Measurement precision
If application data is processed for performance detection, then detection accuracy is improved, but data processing load increases
Solution Approach 1:
The patent extracts only the necessary feature values from wireless data elements, separating the essential monitoring function from the application data processing. By taking out and processing only the critical features (signal strength, packet loss, latency, etc.) rather than processing entire application data streams, the system achieves accurate performance detection with minimal processing load, avoiding the need to analyze all application-level data in detail.
Data Source
AI summary
A method in a computing device includes: monitoring, via a communications interface of the computing device, wireless data elements including application data corresponding to a communications application, and non-application data; automatically generating labels based on the application data of the wireless data elements, the labels indicating performance impacts observable in the communications application; selecting a portion of the wireless data elements corresponding to the non-application data; extracting, from each wireless data element of the non-application data, a set of feature values; generating a plurality of samples from the non-application data, each sample including (i) a number of the sets of feature values based on a sample size, and (ii) one of the automatically generated labels; training a classifier based on the plurality of samples, the classifier configured to receive further non-application data and generate a predicted impact indicator selected from the labels; and deploying the classifier.