Data-Centric Mobile Crowdsensing

Abstract

Mobile crowdsensing (MCS) is a novel and appealing sensing paradigm that leverages the diverse embedded sensors of massive mobile devices to collect different kinds of data. One of the key challenges in MCS is to efficiently schedule mobile device users to perform different sensing tasks. Prior effort to this problem mainly focused on the interaction between the task-layer and the user-layer, without considering the similar data requirements of tasks and the heterogeneous sensing capabilities of users. In this work, we introduce a new data-layer between tasks and users, and propose a three-layer data-centric MCS framework, which enables different tasks to reveal their common data requirements and hence reuse the common data items. We focus on studying the joint task selection and user scheduling problem under this new framework, aiming at maximizing the social welfare. Specifically, we first analyze theoretical performance gain due to data reuse in the ideal scenario with complete information. We then consider the practical scenario with private information of both tasks and users, and propose a two-sided randomized auction mechanism, which is computationally efficient, individually rational, incentive compatible (truthful) in expectation, and close-to-optimal. We further show that the proposed randomized auction may not be budget balanced, and hence introduce a reserve price into the auction to achieve the desired budget balance at the cost of certain welfare loss. Simulation results show that with data reuse, the social welfare achieved in the proposed randomized auction can be increased from 270 up to 4,500 percent, comparing with those without data reuse.