Adaptive Bit Rate Streaming Using On-Device ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current adaptive bit rate streaming algorithms fail to provide optimal video quality and smooth playback in varying network conditions, leading to buffering issues and lack of personalization, as they rely heavily on server-side prediction and require extensive data sets for effective performance.
Innovation Solution
A method using a combination of supervised learning and reinforcement learning to generate a customized adaptive bit rate application, which can be deployed on user devices or edge computing servers, capturing user-specific parameters to enable seamless and personalized video streaming by selecting the appropriate predictive analysis model based on real-time network conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If server-side prediction is used for adaptive bit rate streaming, then video quality can be maintained, but latency increases and performance deteriorates under low network conditions
Solution Approach 1:
The client device performs local prediction of network conditions and video quality using machine learning models deployed on-device. This self-service approach eliminates dependency on server-side prediction, reducing latency while maintaining video quality through local intelligent decision-making about bit rate selection
Solution Approach 2:
The system pre-trains machine learning models on-device using synthetic data and historical patterns, enabling the client to perform prediction actions in advance without requiring real-time server intervention. This preliminary preparation allows rapid local responses to network condition changes
2Measurement precision
If machine learning models are trained on diverse global network conditions, then prediction accuracy improves, but data collection requirements and training complexity increase
Solution Approach 1:
The training process is segmented into two stages: (1) training on synthetic network condition data that covers diverse scenarios without requiring real global data collection, and (2) fine-tuning with limited real device data. This segmentation enables high prediction accuracy across various network conditions while dramatically reducing the complexity of data collection and training operations
Solution Approach 2:
The system creates synthetic copies of real network conditions through simulation, generating training data that represents diverse global scenarios without actually collecting data from all regions. This copying approach maintains prediction accuracy while eliminating the complexity of comprehensive real-world data collection
3Ease of manufacture
If rule-based adaptive bit rate algorithms are used, then implementation is simple, but they fail to provide optimal video quality and smooth playback in varying network conditions
Solution Approach 1:
The system transitions from fixed rule-based parameters to dynamic machine learning models that adapt bit rate selection based on predicted network conditions. This parameter change enables the system to learn optimal video quality settings from historical data and synthetic training, achieving superior playback smoothness and quality while maintaining reasonable implementation through existing ML frameworks
Data Source
AI summary
The disclosure relates to a method and system of selectively deploying an application for facilitating quality-of-experience (QoE) in terms of streaming multimedia content in a networking environment comprising a user-equipment (UE) and a networking node provided with a predictive analysis module. The method comprises: capturing parameters pertaining to UE from at least one of a version of the predictive analysis module with respect to the UE, a current processor occupancy within the UE, a power-level within the UE, network conditions pertaining to the access network etc. One or more of the captured parameters and the observed network conditions is analyzed. Based on analysis, inference is drawn for selecting between the predictive analysis model of the UE and of the networking node for thereby enabling a customized streaming of multimedia content at the UE.