top of page

Enhancing End-to-End Video Delivery by Looking-ahead at User Mobility

The phenomenal growth of mobile video has led researchers and industry to seek novel disruptive paradigms for media delivery. Since much of this content is non real-time, there is ample room to optimize it's end-to-end delivery by exploiting user location-awareness and mobility predictions, as illustrated in Figure 1. This includes 'smarter' 1) caching, 2) content pre-fetching, 3) radio resource allocation and 4) video quality planning. For example, if the radio map indicates that a user's future signal strength will be low, then extra radio resources can be granted to pre-buffer additional content. Furthermore, if the next cell is congested, then (a large number of) lower quality segments will be quickly delivery to prevent future video stalls. 
Current work focused on developing a cross-layer framework that jointly optimizes multi-user radio access and long-term quality planning of adaptive video streaming. We first modeled and formulated the problem as a mixed-integer linear program (MILP) and then developed close to optimal polynomial-time algorithms. Effects of prediction errors and stochastic formulations incorporating uncertainty are currently being addressed. 


Predictive Green Wireless Access: Exploiting Mobility and Application Information

The increasing mobile data traffic and dense deployment of wireless networks have made energy efficient radio access imperative. 
As networks are designed to satisfy peak user demands, radio access energy can be reduced significantly at times of lower demand. This includes putting base stations (BSs) to intermittent short sleep modes,
as well as powering down select BSs completely where demand is low for prolonged time periods. In order to fully exploit such energy conserving mechanisms, networks should be aware of the user temporal and spatial traffic demands which is what we address in this project. To this end, we: 

  • propose an architecture for a predictive green wireless access (PreGWA) framework and identify its key functional entities and their interaction/signalling requirements, 

  • apply the predictive access schemes to improve the transmission efficiency of stored videos through algorithms that incorporate rate predictions. We also minimize downlink BS transmission power for HTTP-based adaptive video streaming. BS on-off switching patterns are developed based on traffic and user location-awareness,

  • develop simple distributed multi-cell resource allocation solutions, and highlight the need for and user equipment assistance (e.g. by reporting video buffer status during handover). 


 Operations and information exchange in predictive wireless acces

 Operations and information exchange in predictive wireless acces

 Illustration of air-time minimization using rate predictions

  • Machine Intelligence and Big Data Analytics for 5G  Networks and Beyond (NSERC Strategic-Queen's University): Topics of research include network traffic prediction, hybrid deep learning architectures for spatio-temporal mobility modeling, anticipatory edge computing and NFV, predictive mobility-aware scheduling.

  • Autonomous Multi-RAT Radio Resource Management for 5G (OCE ENCQOR): The goal of this project is to investigate how unsupervised machine learning and deep reinforcement learning techniques can be used in dual connectivity management of 5G networks.

  • Sensor-less Sensing of 5G Application Traffic (OCE ENCQOR)The goal of this project is to develop artificial intelligence techniques that can model, classify and predict real-time traffic of 5G applications like virtual reality, autonomous vehicle communication, and massive MTC.

bottom of page