Optimal service decomposition for mobile augmented reality with edge cloud support

Abstract

Mobile augmented reality (MAR) applications are starting to attract significant attention due to the enhanced capabilities stemming from both the network and the end devices that propel their realization. However, despite the progress on the end user devices, MAR applications are inherently hugely demanding in terms of computational and memory requirements since they combine, inter alia, video streams, computer generated images, intense computer vision algorithms and geolocation. To this end, edge cloud computing is envisioned as a key technology for supporting such applications where part of the computationally demanding algorithms could be offloaded to suitably selected edge clouds. Within that context the inherent user mobility should be considered to allow an efficient service continuum between edge and the end-terminal. To this end, in this paper, an optimal edge cloud resource MAR service decomposition is presented that takes explicitly into account the AR service composition as well as the inherent user mobility to proactively allocate resources to satisfy the required strict latency and frame accuracy requirements of MAR applications. In addition to the optimal decision making using mathematical programming, and as a mean to provide real-time decision making two advanced heuristic techniques are proposed. A Simulated Annealing based mobility aware AR algorithm (SAMAR) is developed to enhance computing efficiency and a Long Short-Term Memory (LSTM) neural network which is trained offline with optimal solutions. Numerical investigations reveal that significant gains can be achieved by the proposed schemes compare to a number of baseline previously proposed techniques.