Modeling and tracking time-varying clock drifts in wireless networks
Kim, Ha Yang
MetadataShow full item record
Clock synchronization is one of fundamental requirements in distributed networks. However, the imperfection of crystal oscillators is a potential hurdle for network-wide collaboration and degrades the performance of cooperative applications. Since clock discrepancy among nodes is inevitable, many software and hardware attempts have been introduced to meet synchronization requirements. Most of the attempts are built on communication protocols that demand timestamp exchanges to improve synchronization accuracy or resource efficiency. However, link delay and environmental changes sometimes impede these synchronization efforts that achieve in desired accuracy. First, the clock synchronization problem was examined in networks where nodes lack the high accuracy oscillators or programmable network interfaces some previous protocols depend on. Next, a stochastic and practical clock model was developed by using information criteria which followed the principle of Occam's razor. The model was optimized in terms of the number of parameters. Simulation by using real measurements on low-powered micro-controllers validated the derived clock model. Last, based on the model, a clock tracking algorithm was proposed to achieve high synchronization accuracy between unstable clocks. This algorithm employed the Kalman filter to track clock offset and skew. Extensive simulations demonstrated that the proposed synchronization algorithm not only could follow the clock uncertainties shown in real measurements but also was tolerant to corrupted timestamp deliveries. Clock oscillators are vulnerable to noises and environmental changes. As a second approach, clock estimation technique that took circumstances into consideration was proposed. Through experiments on mobile devices, the obstacles were clarified in synchronization over wireless networks. While the causes of clock inaccuracy were focused on, the effect of environmental changes on clock drifting was investigated. The analysis of the observations inspired an M-estimator of clock error that was accurate but under dominant disturbances such as oscillator instability and random network delay. A Kalman filter was designed to compensate with temperature changes and estimate clock offset and skew. The proposed temperature-compensated Kalman filter achieved the better estimates of clock offset and skew by adjusting frequency shifts caused by temperature changes. The proposed Kalman filter-based clock synchronization was implemented in C. A real-time operation was proved by clock tracking between two mobile platforms that the synchronization technique was implemented on. Moreover, the technique was converted to fixed-point algorithm, which might degrade performance, to evaluate the synchronizing operation on fixed-point processors. The fixed-point simulation reported performance degradation caused by limited hardware resources; however, it also corroborated the applicability of the synchronization technique.