Improving Perceived Quality of Live Adaptative Video Streaming.

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Live streaming services like Dynamic Adaptive Streaming over HTTP (DASH) are sensitive to network issues such as delay, jitter, and packet loss.
• Large buffers, common in video-on-demand, are unsuitable for live streaming due to induced delay, potentially degrading quality during network congestion.
• Active Queue Management (AQM) aims to control network congestion by managing router queues and influencing transmission rates.

Purpose of the Study:

• To evaluate the performance of existing AQM strategies for real-time adaptive video streaming.
• To propose and validate a novel AQM algorithm leveraging Long Short-Term Memory (LSTM) neural networks for improved user-perceived video quality in live streaming.

Main Methods:

• Performance evaluation of recent AQM strategies in live streaming scenarios.
• Development of a new AQM algorithm utilizing LSTM neural networks to predict queue delay trends.
• Implementation of early packet discard mechanisms based on LSTM predictions to preempt network congestion.

Main Results:

• The proposed LSTM-based AQM algorithm demonstrates superior performance compared to existing AQM methods.
• Significant improvements in user-perceived video quality were observed, particularly in congested network conditions.
• The LSTM model effectively forecasts queue delay, enabling proactive congestion management.

Conclusions:

• The LSTM-based AQM approach offers a promising solution for enhancing the quality of live adaptive video streaming.
• Proactive congestion control through predictive modeling is crucial for maintaining seamless video playback.
• This method effectively addresses the challenges posed by network congestion in unmanaged networks for streaming applications.