AI Hierarchical Service Awareness Engine for Encrypted Traffic Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing number of mobile applications and encrypted network traffic make it challenging for service operators to accurately recognize specific applications in real-time, especially when applications share common content delivery network servers, hindering effective service treatment and billing processes.
Innovation Solution
An AI-based hierarchical service awareness engine uses decode equivalent classes (DECs) to recognize applications by training AI models, which generate inference models deployed in user plane function (UPF) gateways, enabling real-time application recognition and service treatment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional application recognition methods are used, then the system can identify applications in unencrypted traffic, but it fails to accurately recognize applications in encrypted traffic (80% of network traffic by 2019)
Solution Approach 1:
The patent introduces Decode Equivalent Class (DEC) information as an intermediary element that bridges the gap between encrypted traffic and application recognition. DEC information serves as a mediator that captures essential application characteristics without requiring decryption of the actual traffic content, enabling accurate application identification while maintaining encryption privacy protection.
Solution Approach 2:
The patent replaces traditional mechanical inspection methods (deep packet inspection, protocol analysis) with AI-based inference models. Instead of mechanically examining encrypted packet contents, the system uses machine learning models trained on DEC information to infer application types, transitioning from direct observation to intelligent deduction.
2Adaptability or versatility
If the network recognizes every specific application individually, then service operators can provide differentiated service treatments, but the complexity of managing numerous applications (e.g., Google's multiple apps sharing CDN servers) increases significantly
Solution Approach 1:
The patent segments the application recognition system into hierarchical levels: DEC information extraction layer, AI model training layer, and inference model deployment layer. This segmentation allows the system to handle application diversity through structured classification without overwhelming complexity at any single level, enabling manageable service differentiation.
Solution Approach 2:
The patent creates universal AI models that can handle multiple applications through DEC information. Instead of requiring separate recognition mechanisms for each application, the system develops multi-functional inference models trained on DEC characteristics that can identify various applications (Gmail, Google Drive, YouTube, etc.) using a unified approach.
3Measurement precision
If the network performs deep inspection to recognize applications in less than ten packets for billing purposes, then billing accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary action by training AI models offline using extensive DEC information from network traffic. This pre-training phase prepares inference models in advance, so that during actual billing operations, the system can quickly recognize applications without performing time-consuming deep inspections, achieving both accuracy and speed.
Solution Approach 2:
The patent extracts only the essential DEC information from traffic packets rather than performing complete deep inspection of all packet contents. This partial action approach captures sufficient characteristics for accurate application identification while significantly reducing processing time and computational overhead for billing operations.
Data Source
AI summary
Systems and methods are provided for recognition of an application in communication traffic flow in a network using an artificial intelligence (AI) based hierarchical service awareness engine. A decode equivalent class (DEC) can be used to provide information on the application. A DEC corresponds to a class of traffic that is mapped to an artificial intelligence (AI) model associated with parameters related to the class of traffic. DEC information can be fed to an AI model set and an inference model can be derived from a AI model of the AI model set corresponding to a DEC. The inference model can be provided to a gateway of the network to recognize a specific application of a service in communication flows. In various embodiments, in training the AI models, the gateway can provide DEC information for the AI model set from classifying flows of data traffic received from the network into DECs.