Resilient, Trustworthy, Ubiquitous Time Transfer using DTM and NTP

Abstract

Packet delay asymmetry and jitter are known to be major factors limiting the time accuracy in network-based synchronization protocols. One aim of the UTIME project was to better understand the network asymmetry and jitter problem and to try improving the current limitations. To this purpose two network links have been established in Germany. Link A is between GMV offices in Darmstadt and Munich, separated by 300 km. Link B is between Darmstadt and PTB, the German national metrology institute in Braunschweig, also 300 km away. Link B is based on an inexpensive Fibre To The Home (FTTH) network access on each site. Link A is based on a more professional (but slightly more expensive) Multiprotocol Label Switching (MPLS) network provision. Accurate time generated by a time-calibrated multi-frequency GNSS receiver is sent from the server in Darmstadt to the client using two protocols: DTM for Link A (MPLS), and NTP for Link B (FTTH). DTM is an ETSI standard (ETSI ES 201 803-x) with server and client running on Nimbra equipment developed by Net Insight in Sweden. The selected NTP implementation is Red Hat’s Chrony software with server and client running on a Raspberry Pi. Chrony implements the Network Time Security (NTS) protocol, which adds guaranteed authenticity to NTP traffic. On the client side, the resulting DTM and NTP synchronization accuracy is monitored by means of a calibrated GNSS receiver in Munich, and by comparison with the time scale UTC(PTB), respectively. The main conclusion of the UTIME project is that it is possible to achieve sub-microsecond accurate time distribution (both in offset and jitter) over several hundred km (Munich-Darmstadt) using packet-exchange network technology (DTM) over relatively inexpensive and commercially available end-to-end network links (MPLS), without a detailed knowledge of the underlying network topology and structure (no “network engineering” required). In the case of NTP over FTTH, the time distribution presents a large time offset, of the order of few milliseconds, due to the unavoidable packet delay asymmetry in FTTH-like networks. This large error is not acceptable for most industrial applications today in telecom, energy, finance, etc. Furthermore, NTP shows in general a much higher time error in jitter, of the order of 100 microseconds, which is two orders of magnitude larger than the DTM jitter. This large NTP jitter is completely normal, due to limitations in the FTTH network and in the NTP hardware and software. By carefully tuning the NTP configuration and choosing packets with minimal round-trip time, we were able to reduce the NTP jitter down to 10 microseconds, but not further. NTP could potentially reach a microsecond-level jitter using a combination of techniques inspired in DTM, in particular, using a very high packet exchange rate, using a more stable computer clock, and using Artificial Intelligence (AI) to select “lucky packets” with positive properties, such as minimal round-trip time (and hence minimal time jitter). Since DTM and NTP are both basically using the same client-server packet exchange mechanism, it should be possible to upgrade NTP to achieve a level of performance similar to DTM. Some modifications in NTP at hardware and software level would be required. After such modifications, NTP should be able to achieve sub-microsecond accuracy (in offset and jitter) over MPLS links. Over FTTH links, NTP should be able to achieve a sub-microsecond jitter similar to the MPLS case, but with a time offset at the millisecond level due packet delay asymmetry in the network. The NTP offset is unavoidable over FTTH, however this delay is typically constant over several days and therefore it could be calibrated using space-based PNT systems, in particular GNSS, or other, for example LEO satellite systems. This work was funded under the European Space Agency Navigation Innovation and Support Programme Element 1, which is dedicated to the technology innovation in the PNT sector.