In addition to operating Internet and telephone-based time services, we also conduct research on improving the accuracy and efficiency of these services, and on developing ways to authenticate and certify the time signal.

Service Accuracy

The accuracy of the digital time messages is limited primarily by the jitter in the delay through the transmission network, although the noise in the synchronization of the servers also makes a smaller contribution. We are studying ways of modeling the network jitter and of designing algorithms that can automatically respond to changes in the network delay. In addition, we are upgrading the clocks in the servers to improve their short-term stability and to provide increased reliability in the event of a temporary failure in the link between a remote server and the primary clock ensemble in Boulder.

Service Efficiency

We are developing algorithms that can automatically configure themselves to provide a specified level of accuracy at minimum cost. The algorithms combine the accuracy requirement, the quality of the network connection and the stability of the local clock oscillator to adjust this configuration in real-time as the parameters change in response to changes in the network delay, fluctuations in the ambient temperature of the client system and similar factors.

One of the most important conclusions of this work is that the ratio of cost to performance is not linear. Small decreases in the specified accuracy can result in significant savings both in terms of the network bandwidth that is required and in terms of the load on the NIST time servers. Since the load on the NIST time servers is increasing at a compounded rate of nearly 7% per month, improvements in the efficiency of the client software is very important to the continued operation of the service on a finite budget.

A related area of study involves studies of the algorithms used to discipline the local clock oscillator. This work is particularly relevant to larger workstations and servers, since these systems usually have background processes which query the NIST primary servers quite frequently. Two common solutions are based on pure frequency lock loops or phase-lock loops; recent work has shown that a hybrid combination of the two may be most suitable in some circumstances.

General References

Judah Levine, An Algorithm to Synchronize the Time of a Computer to Universal Time, IEEE Trans. On Networking, vol. 3, pp. 42-50, February 1995.

Judah Levine, Time Synchronization over the Internet Using an Adaptive Frequency-Locked Loop, IEEE Trans. On Ultrasonics, Ferroelectrics and Frequency Control, vol. 46, pp. 888-896, July 1999.

Judah Levine, Introduction to Time and Frequency Metrology, Rev. Scientific Instruments, vol. 70, pp. 2567-2596, June 1999.

For questions or more information contact Judah Levine at jlevine@boulder.nist.gov.