Towards Resource-Efficient Wireless Edge Analytics for Mobile Augmented Reality Applications

From entertainment to education, augmented reality (AR) is about to impact positively our everyday lives. Enhanced capabilities of mobile devices, such as smartphones or wear-ables, as well as ubiquitous network connectivity give AR the opportunity to prosper. Despite these improvements, AR requires computationally heavy tasks, such as context recognition and classification through image or video processing, which are hard to fulfill on mobile devices. To this end, solutions for computation offloading to cloud servers have been proposed. We consider a scenario where context identification is performed through elicitation of user-generated information, such as images or small video files. It is the quantity of this information that ultimately determines the context classification precision, which we model as a Binomial random variable. We introduce the problem of maximizing a lower bound of the precision of context classification through prudent resource allocation, namely computation offloading, and bandwidth and computational capacity allocation at the wireless network edge. We define the context classification precision as a function of the quantity of information that users provide, and we demonstrate through numerical experiments that appropriate management of the limited resources at the wireless edge can maximize the classification precision of data analytics mechanisms needed for augmented reality applications.