[Paper Review] Adaptive Video Streaming over LTE Unlicensed

Apostolos Galanopoulos, Antonios Argyriou|arXiv (Cornell University)|Jul 31, 2016

Advanced Wireless Network Optimization16 references3 citations

TL;DR

This paper proposes a two-stage framework for adaptive video streaming over LTE-Unlicensed (LAA) networks, where the eNodeB uses utility maximization and ADMM to select optimal video quality per user based on unlicensed spectrum availability, followed by a Lyapunov-based scheduler that prioritizes users with high queue backlogs. The approach reduces buffer under-runs and video freezes by 3× compared to PFS and AVIS, significantly improving QoE.

ABSTRACT

In this paper we consider the problem of adaptive video streaming over an LTE-A network that utilizes Licensed Assisted Access (LAA) which is an instance of LTE Unlicensed. With LTE Unlicensed users of the LTE-A network opportunisti- cally access radio resources from unlicensed and licensed carriers through Carrier Aggregation (CA). Our objective is to select the highest possible video quality for each LTE-A user while also try to deliver video data in time for playback, and thus avoid buffer under-run events that deteriorate viewing experience. However, the unpredictable nature of the wireless channel, as well as the unknown utilization of the unlicensed carrier by other unlicensed users, result in a challenging optimization problem. We first focus on developing an accurate system model of the adaptive streaming system, LTE-A, and the stochastic availability of unlicensed resources. Then, the formulated problem is solved in two stages: First, we calculate a proportionally fair video quality for each user, and second we execute resource allocation on a shorter time scale compatible with LTE-A. We compare our framework with the typical proportional fair scheduler as well as a state-of-the-art LTE-A adaptive video streaming framework in terms of average segment quality and number of buffer under- run events. Results show that the proposed quality selection and scheduling algorithms, not only achieve higher video segment quality in most cases, but also minimize the amount and duration of video freezes as a result of buffer under-run events.

Motivation & Objective

• Address the challenge of maintaining high video quality and minimizing buffer under-runs in LTE-A networks using Licensed Assisted Access (LAA) with unlicensed spectrum.
• Overcome the unpredictability of unlicensed carrier access and variable channel conditions in adaptive video streaming.
• Design a system where the eNodeB centrally makes video quality decisions using knowledge of unlicensed band dynamics, improving fairness and reliability.
• Minimize video freeze duration and frequency by integrating queue backlog awareness into the scheduling policy.
• Outperform standard proportional fair scheduling (PFS) and state-of-the-art frameworks like AVIS in both segment quality and playback smoothness.

Proposed method

• Formulate a utility maximization problem at the eNodeB to determine proportionally fair video segment quality per user, considering unlicensed spectrum availability and resource constraints.
• Solve the convex optimization problem using the Alternating Direction Method of Multipliers (ADMM) to enable distributed and scalable quality selection.
• Implement a Lyapunov-based scheduling policy that prioritizes users with high queue backlogs and high instantaneous data rates to ensure timely delivery.
• Use Carrier Aggregation (CA) to combine licensed and unlicensed carriers, enabling dynamic access to unlicensed spectrum while respecting coexistence rules.
• Integrate real-time unlicensed channel access probability estimates into the quality selection process to avoid overestimating available data rates.
• Decouple quality selection from scheduling: quality is chosen in a longer timescale, while scheduling is executed on a shorter timescale to meet delivery deadlines.

Experimental results

Research questions

• RQ1 How can video quality be optimally selected in LTE-A networks using unlicensed spectrum when channel conditions and unlicensed carrier availability are highly dynamic?
• RQ2 To what extent does centralizing video quality decision-making at the eNodeB—based on unlicensed band activity—improve QoE compared to UE-based decisions?
• RQ3 Can a Lyapunov-optimization-based scheduler reduce video freeze duration and frequency by prioritizing users with high queue backlogs?
• RQ4 How does the proposed framework compare to standard PFS and AVIS in terms of average segment quality and buffer under-run events?
• RQ5 What is the impact of unlicensed spectrum variability (modeled via Poff standard deviation) on video freeze probability and duration?

Key findings

• The proposed framework reduces video freeze probability by approximately 3× compared to both PFS and AVIS across all tested unlicensed channel variability levels.
• Average freeze duration is significantly lower under BCASP than under PFS and AVIS, due to proactive queue management and backlog-aware scheduling.
• The framework achieves higher average segment quality than PFS and AVIS, especially in high-variability unlicensed spectrum conditions.
• The use of ADMM enables efficient and scalable solution of the utility maximization problem for video quality selection under resource constraints.
• Lyapunov-based scheduling effectively balances fairness and delay by prioritizing users with high queue backlogs, minimizing the risk of buffer under-runs.
• The eNodeB’s centralized knowledge of unlicensed band dynamics leads to more accurate quality selection, reducing the likelihood of selecting unachievable video qualities.