Network Device Learning Model for Packet Group Quality Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In moving image distribution systems, the correlation between changes in communication condition information and quality information on terminal devices is often weak, leading to inaccurate estimation of quality information even when using learning models.
Innovation Solution
A network device that calculates group information indicating packet acquisition statuses for each packet group, and generates a learning model by learning teacher data including this group information, communication condition information, and quality information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a learning model is generated using conventional communication condition information, then the model can be trained with available data, but the estimation accuracy of quality information remains low due to weak correlation between communication conditions and quality outcomes
Solution Approach 1:
The patent segments packets into multiple packet groups based on their relationships (e.g., RTP packet groups, TCP packet groups), and calculates acquisition status separately for each group. This segmentation allows the system to capture nuanced quality information that reflects the actual impact of packet loss on reproduced quality, thereby improving estimation accuracy despite weak overall correlation between communication conditions and quality outcomes.
Solution Approach 2:
The patent introduces a new dimension of analysis by calculating acquisition status for multiple different packet groups (RTP groups, TCP groups, etc.) rather than treating all packets uniformly. This multi-dimensional approach to measuring packet acquisition status enables the learning model to capture quality-relevant patterns that are not apparent in conventional single-dimension communication condition metrics.
2Measurement precision
If packet acquisition status is calculated for each packet individually, then detailed quality information can be obtained, but the computational complexity and data volume increase significantly
Solution Approach 1:
Instead of calculating acquisition status for each individual packet, the patent segments packets into groups and calculates the acquisition status at the group level. This reduces the computational burden from O(n) individual packet calculations to O(n/g) group calculations (where g is the average group size), significantly lowering device complexity while preserving quality information detail through the group-based measurement approach.
Solution Approach 2:
The patent merges multiple related packets into packet groups (such as RTP packet groups containing related media packets, or TCP packet groups containing related data packets) and calculates a single acquisition status for each group. This merging approach maintains detailed quality information by preserving group-level distinctions while reducing overall computational complexity through aggregated calculations.
Data Source
AI summary
A network device includes: a memory; and a processor coupled to the memory and the processor configured to: calculate, based on acquisition statuses of packets in a capture device that acquires the packets transmitted from an application device to a terminal device over a network, first group information indicating the acquisition statuses for each of a plurality of packet groups each including a plurality of packets having a predetermined relationship with each other; and generate a learning model by learning teacher data including the first group information calculated, first communication condition information indicating communication conditions of the packets in the network, and first quality information indicating quality in terms of output of the packets on the terminal device.