Learning to detect frame synchronization

Abstract

Conventional frame synchronization detection algorithms in the communication system, such as correlation and maximum likelihood, either fails to deal with frequency deviation problem or requires great efforts to be implemented on the hardware platform. Therefore, we parallel the frame synchronization detection problem with the Multi-instance Learning (MIL) in the field of machine learning, and propose a novel Learning to Detect Frame Synchronization algorithm (LDFS). This algorithm is mainly conducted offline to learn preamble signal detectors, which can then be efficiently implemented on the hardware platform to accomplish the synchronous detection task. In this paper, we first solve frame synchronization detection problem from the point of machine learning, and thus our algorithm displays some advantages over the existing algorithms. First, the resulted detector is simple and efficiently realized on the hardware platform. Second, the learned detector is adaptive to work under different communication frequencies straight-forwardly without extra modifications. Experimental results demonstrate the effectiveness and promise of the proposed LDFS algorithm. © Springer-Verlag 2013.