Hough Transform-Based Clock Skew Measurement by Dynamically Locating the Region of Offset Majority

Abstract

SUMMARYA network-based remote host clock skew measurement involves collecting the offsets, the differences between sending and receiving times, of packets from the host within a period of time. Although the variant and immeasurable delay in each packet prevents the measurer from getting the real clock offset, the local minimum delays and the majority of delays delineate the clock offset shifts, and are used by existing approaches to estimate the skew. However, events during skew measurement like time synchronization and rerouting caused by switching network interface or base transceiver station may break the trend into multi-segment patterns. Although the skew in each segment is theoretically of the same value, the skew derived from the whole offset-set usually differs with an error of unpredictable scale. In this work, a method called dynamic region of offset majority locating (DROML) is developed to detect multi-segment cases, and to precisely estimate the skew. DROML is designed to work in realtime, and it uses a modified version of the HT-based method [8] both to measure the skew of one segment and to detect the break between adjacent segments. In the evaluation section, the modified HT-based method is compared with the original method and with a linear programming algorithm (LPA) on accumulated-time and short-term measurement. The fluctuation of the modified method in the short-term experiment is 0.6 ppm (parts per million), which is obviously less than the 1.23 ppm and 1.45 ppm from the other two methods. DROML, when estimating a four-segment case, is able to output a skew of only 0.22 ppm error, compared with the result of the normal case.

Coexisting Parallelogram Method to Handle Jump Point on Hough Transform-based Clock Skew Measurement

“…As a result, the Hough transform-based method will produce a large parallelogram-like region that covers all the offsets, including the jump point, and finally produce an inaccurate skew result. As far as we are concerned there is a solution to handle this jump point problem, a method called DROML presented in [31]. DROML created a dynamic region to cover the densest part of the offsets collection, which makes the calculated skew is spared from the effect of the jump point.…”

“…Hence, we omit the effect of the jump point by obtaining the skew from skews of some subsets of the offsetset which are no longer contain the jump point. Our evaluation results show that compared with LPA [20], Aoki's method [2], and the Hough transform based method [31], only the proposed method can handle the problem accurately. The following section describes in detail the voting scheme we used in this paper.…”