Wireless Multicast for Zoomable Video Streaming

Hui Wang, Mun Choon Chan, Wei Tsang Ooi

TL;DR

This paper tackles the challenge of efficiently delivering zoomable video streams to a diverse group of users over wireless networks. They propose an optimization algorithm that intelligently decides which resolution each video tile should be sent at, maximizing overall video quality for all users and achieving significant improvements over existing methods.

What Problem Does This Paper Address?

Interactive video applications, like those offering zoom and pan functionality, are becoming increasingly popular. These "zoomable" videos are typically broken down into a grid of independent "tiles." The challenge arises when trying to deliver these tiles wirelessly to a group of users with varying network speeds (link rates) and different viewing interests (regions of interest or RoIs). Simply sending the highest resolution to everyone is inefficient, while sending the lowest resolution degrades the user experience. The core problem is how to dynamically allocate tile resolutions and transmission rates to maximize the collective video quality for all users, given their individual needs and network constraints.

Methods

The authors introduce a novel optimization algorithm to solve this complex resource allocation problem. Their approach involves:

  • Modeling the Problem: They frame the problem as a utility maximization task, where the goal is to maximize the sum of user utilities (representing video quality) subject to the total available transmission time slots.
  • Adaptive Utility Assignment: To ensure fairness and prevent severe quality degradation, they adaptively tune a lower bound for the guaranteed resolution level each user must receive.
  • Optimal Allocation for a Single Tile: They first design a dynamic programming algorithm to determine the optimal resolution and link rate for transmitting a single tile. This involves considering all users' interests and link rates.
  • Virtual Clustering: To make the single-tile algorithm scalable to a large number of users, they employ a virtual clustering technique. Users with identical link rates are grouped, reducing the effective number of users to the number of distinct link rates.
  • Optimal Allocation for Multiple Tiles: Finally, they extend the single-tile allocation to handle multiple tiles by optimally distributing the total available time slots across all tiles. This is achieved through another dynamic programming approach.

Datasets

The paper does not explicitly mention using pre-existing datasets for training. Instead, they evaluate their algorithm using:

  • Test Videos: Standard HD video files (Crowd-Run, Old-Town-Cross, Rush-Hour) were used.
  • Real User Interaction Logs: RoI requests and variations were based on logs from 10 users who had previously used a zoomable video system.
  • Simulated Wireless Channels: Mobile devices were placed at varying distances from an Access Point (AP) to simulate different channel conditions and link rates.

Key Insights and Ideas

  • Feasibility of Mixed Resolutions: A key insight is that mixing different resolution levels for tiles within a single video frame is perceptually acceptable to users in many scenarios, especially for lower motion content. This opens the door for significant bandwidth savings.
  • Balancing Quality and Efficiency: The core challenge is finding the sweet spot between delivering high-quality video (higher resolution tiles) and efficient bandwidth usage (lower resolution tiles or fewer transmissions). Their algorithm aims to strike this balance optimally.
  • Scalability through Clustering: The problem's complexity grows significantly with the number of users. Virtual clustering based on link rates is a crucial technique to make the optimal solution practical for real-world scenarios.
  • Fairness is Key: Simply maximizing total utility can lead to unfairness. The adaptive utility assignment mechanism is vital for ensuring that all users receive a minimum acceptable quality, preventing some users from being starved of bandwidth.

Key Results

  • Significant Quality Improvement: The proposed optimal multicast algorithm achieved up to 12dB improvement in PSNR compared to other heuristic methods (adaptive unicast, adaptive multicast, and approximation multicast).
  • Goodput Gains: The algorithm also demonstrated substantial goodput improvements, with up to 19% and 34% gains over approximation and adaptive multicast, respectively, when supporting 10 users.
  • Efficiency: The optimal algorithm's average running time was reported as 49.18ms, incurring only a 2.5% overhead, making it practical for real-time deployment.
  • Robustness: The algorithm showed good performance across different video content types (motion levels) and was resilient to client mobility and varying wireless channel conditions.

Limitations and Future Work

  • No Packet Loss/Retransmission Handling: The current algorithm does not explicitly incorporate packet loss or frame retransmissions. The authors mention using conservative threshold parameters to mitigate this.
  • Single Access Point: The current system setup assumes a single access point.
  • Future Work: The authors suggest extending their work to scenarios with multiple access points (APs), where AP association mechanisms could be further exploited to enhance multicast performance. They also note that their algorithm can be directly applied to general multi-session video multicast problems.

Summary updated

Abstract

Zoomable video streaming refers to a new class of interactive video applications, where users can zoom into a video stream to view a selected region of interest in higher resolutions and pan around to move the region of interest. The zoom and pan effects are typically achieved by breaking the source video into a grid of independently decodable tiles. Streaming the tiles to a set of heterogeneous users using broadcast is challenging, as users have different link rates and different regions of interest at different resolution levels. In this article, we consider the following problem: Given the subset of tiles that each user requested, the link rate of each user, and the available time slots, at which resolution should each tile be sent, to maximize the overall video quality received by all users. We design an efficient algorithm to solve this problem and evaluate the solution on a testbed using 10 mobile devices. Our method is able to achieve up to 12dB improvements over other heuristic methods.
BibTeX
@article{wang2015wireless,
  author = {Hui Wang and Mun Choon Chan and Wei Tsang Ooi},
  title = {{Wireless Multicast for Zoomable Video Streaming}},
  journal = {ACM Trans. Multim. Comput. Commun. Appl.},
  year = {2015},
  url = {https://openreview.net/forum?id=XKUSt5iU2s0},
}
Figure