Multi-class labeling with BCH codes for mobile crowdsensing

Mobile Crowdsensing is an effective paradigm to perform tasks in many scenarios by utilizing crowd intelligence and sensing resources. However, the responses from users are often unreliable due to the low-paid rewards and their biased expertise of the crowd. In this paper, we propose a multi-class labeling scheme based on BCH codes (BCH-MCL) for the mobile crowdsensing system where the quality of crowd is unknown. For multi-class labeling tasks, BCH-MCL provides each label a BCH codeword with maximum error correction capability, and maps the responses of the crowd into an estimated codeword to determine the final estimated label. The theoretical analysis characterizes the fault-tolerance capability of BCH-MCL, and derives an upper bound of the mis-label probability for the proposed BCH-MCL scheme, with a necessary condition and a sufficient condition presented. Furthermore, we prove that BCH-MCL can achieve a more accurate estimation and much lower computational complexity, compared with prior work DCFECC. Simulation results validate our theoretical analysis and indicate the effectiveness and efficiency of the proposed BCH-MCL.