[Paper Review] Delay-Sensitive and Power-Efficient Quality Control of Dynamic Video Streaming using Adaptive Super-Resolution

Minseok Choi, Won Joon Yun+1 more|arXiv (Cornell University)|Oct 12, 2021

Image and Video Quality Assessment53 references4 citations

TL;DR

This paper proposes a delay-sensitive, power-efficient video streaming system that jointly optimizes video quality, buffering, and resource usage by adaptively compressing video chunks at the transmitter and using an adaptive super-resolution (SR) network at the receiver. By dynamically adjusting compression rates and SR network depth based on network and device states, the scheme achieves superior quality-of-service with reduced delay, power consumption, and CPU usage compared to non-adaptive or non-cooperative methods.

ABSTRACT

In a decade, the adaptive quality control of video streaming and the super-resolution (SR) technique have been deeply explored. As edge devices improved to have exceptional processing capability than ever before, streaming users can enhance the received image quality to allow the transmitter to compress the images to save its power or pursue network efficiency. In this sense, this paper proposes a novel dynamic video streaming algorithm that adaptively compresses video chunks at the transmitter and separately enhances the quality at the receiver using SR. In order to allow transmission of video chunks with different compression levels and control of the computation burden, we present the adaptive SR network which is optimized by minimizing the weighted sum of losses extracted from different layer outputs. for dynamic video streaming. In addition, we jointly orchestrate video delivery and resource usage, and the proposed video delivery scheme balances the tradeoff well among the average video quality, the queuing delay, buffering time, transmit power, and computation power. Simulation results show that the proposed scheme pursues the quality-of-services (QoS) of the video streaming better than the adaptive quality control without the cooperation of the transmitter and the receiver and the non-adaptive SR network.

Motivation & Objective

• To address the growing demand for high-quality video streaming in resource-constrained environments such as IoT and V2X networks.
• To reduce playback stalls, queuing delay, and power consumption in dynamic video streaming under time-varying channel and device conditions.
• To jointly optimize video delivery and computation tasks at the transmitter and receiver, balancing conflicting performance metrics like quality, delay, and energy use.
• To enable efficient use of limited CPU and transmit power by adapting the depth of the super-resolution network based on available resources.
• To improve quality-of-service (QoS) and quality-of-experience (QoE) in dynamic video streaming through adaptive control of compression and SR processing.

Proposed method

• The system uses an adaptive super-resolution generative adversarial network (ASRGAN) whose depth is dynamically selected based on available CPU resources to balance quality and computational cost.
• The transmitter adaptively compresses video chunks using different bitrates based on queue state, channel conditions, and buffer levels at the receiver.
• A drift-plus-penalty optimization framework is employed to jointly control transmission rate, compression level, and SR network depth, minimizing a weighted sum of delay, power, and quality degradation.
• The adaptive SR network is trained by minimizing a weighted sum of losses from multiple intermediate layer outputs, enabling flexible trade-offs between quality and computation.
• The system monitors real-time metrics such as transmitter queue length, receiver buffer state, transmit power, and CPU usage to make dynamic decisions.
• The optimization framework uses a parameter $ V $ to weight the importance of video quality degradation, allowing trade-offs between quality and delay/power.

Experimental results

Research questions

• RQ1 How can video streaming systems balance video quality, queuing delay, and power consumption in dynamic, resource-constrained environments?
• RQ2 What is the optimal way to adaptively control video compression at the transmitter and super-resolution processing at the receiver to improve QoS and QoE?
• RQ3 How does joint optimization of transmission and computation decisions affect the performance of video streaming under limited CPU and transmit power?
• RQ4 To what extent can adaptive super-resolution reduce computational load while maintaining acceptable video quality in real-time streaming?
• RQ5 How does the proposed scheme compare to non-adaptive or non-cooperative approaches in terms of delay, quality, and resource efficiency?

Key findings

• The proposed scheme reduces playback stall events and average buffer length significantly compared to the non-adaptive baseline ‘Comp1’, despite using more CPU cores.
• With $ heta = 2.5 $, the proposed scheme utilizes CPU 2.5 times more efficiently than ‘Comp1’, indicating better resource utilization.
• The average PSNR of received video chunks is higher in the proposed scheme than in ‘Comp1’, demonstrating improved video quality.
• As the weight parameter $ V $ increases, the system prioritizes higher video quality, leading to deeper ASRGAN networks and reduced delay incidence, even with higher CPU usage.
• The proposed scheme achieves a better trade-off between video quality, delay, and power consumption than both non-adaptive SR and non-cooperative adaptive streaming methods.
• The joint control of transmitter and receiver decisions enables a more balanced and stable performance across multiple conflicting metrics, including buffering, delay, and energy use.