SSA: Microsecond-Level Clock Synchronization Based on Machine Learning for IoT Devices.

Clock synchronization is an essential but challenging task for internet of things (IoT) devices. The state-of-the-art data-driven Huygens solution cannot achieve accuracy for IoT networks, because the devices are usually weak in power to make massive timestamp probing for data-driven solutions. We propose the SSA clock synchronization scheme to improve the Huygens algorithm. First, SSA has a sliding window mechanism to accumulate data points for the data-driven support vector machine (SVM) algorithm in Huygens, which complements the issue of insufficient data points. Second, SSA applies a smoothing method to the periodical estimated clock offset and drift, which eliminates the noise introduced by the larger sliding window. Third, SSA makes an adaptive clock correction instead of the periodical correction in Huygens so as to avoid correcting the clock before the algorithm could stably estimate and smooth the clock offset and drift. We conduct extensive experiments on a real device (Huawei Sound X), and the results shows that our SSA can achieve synchronization accuracy of around 20 mu s in the actual working environment.