On Tuesday, 26 january 2016 a little after noon at Aalto University Metsähovi Radio Observatory, the Hydrogen Maser atomic clock automatic monitoring system alerted the GPS satellite positioning that the time jumped suddenly. The automatic monitoring system of a hydrogen maser atomic clock triggered an alarm which reported a deviation of 13.7 microseconds. While this is tiny, it is a sign of a problem somewhere, and does not exclude the possibility of larger timekeeping problems happening. The specific source of the problem is not known, but candidates are a faulty GPS satellite or an atomic clock placed in one. Particle flare-up from sun is unlikely, as the observatory has currently not detected unusually high activity from sun.
14 microseconds in the realm of high-frequency trading is around 50 times the usual granularity at which decisions are made by electronic systems these days. Therefore, this is HUGE. The consequences are many. Automated strategies rely on precise timestamping to send or receive orders. These timestamps come from very precise, disciplined clocks that rely on GPS, like the popular Symmetricom S350. The S350s then distribute the clock to the compute farm using the IEEE 1588 Precise Time Protocol (PTP). But without a precise souce, all this expensive infrastructure is as reliable to tell time as Brutus, the Rotweiller.
Further, data timestamps are used to correlate events that happen in distinct exchanges, looking for leaders and laggers, which is the bread and butter of most HFT strategies. With a wrong timestamp, many algorithm optimizers will either (1) flag a trade that does not exist or (2) ignore a trade opportunity, both of which are pretty bad. Given that this is an isolated episode, it seems that it can be ignored.
The reliance of the US and Europe financial trading industry on GPS timestamps is so pervasive and still, the GPS system seems so frail and open to local cyber/terrorist attacks even from a short distance from the exchange buildings that it begs the question: why isn’t anyone scared?
Let’s watch out for future mishaps. This will be interesting.
UPDATE: NAVCEN confirmed GPS SATELLITE SVN23 (PRN32) suffered a malfunction and was removed from the constellation. Reasons are still unknown. The time warp was confirmed in UK, US, Finland, China, Germany and Australia at least.
UPDATE: there were some recent news regarding this satellite http://gpsworld.com/last-block-iif-to-replace-oldest-gps-satellite/
“SVN-23, launched on Nov. 26, 1990, has been an ‘Iron Bird’ workhorse in the E-plane and has successfully served the world’s GPS users for over 25 years,” said Rick Hamilton, CGSIC Executive Secretariat, in an email. “This is over 18 years past its designed service life, having operationally outlasted (and, in many cases, outperformed) its peers on-orbit due to the diligent efforts of the men and women of the U.S. Air Force.”
UPDATE: There is also a discussion about this satellite on Google Groups from Feb 7, 2008
The U.S. military announced earlier this week that GPS satellite SVN23, transmitting L-band code as PRN32, will be set to usable status on Feb. 19. This is notable for a couple of reasons, but namely because it is the first time that the PRN32 designation will be used by an operational, healthy GPS satellite.
GPS receivers initially were built to accommodate up to 31 satellite signals, and a PRN designated with the number 32 can’t be tracked by some manufacturers’ devices that look for PRNs numbered 0 through 31. The U.S. Air Force began testing the PRN32 designation late in 2006 — while SVN23/PRN32 was set to unhealthy and not included in the operational GPS constellation almanac, some all-in-view GNSS tracking stations received the L-band signal. As of January 2007, SVN23 has been broadcasting but set to unhealthy and not included in the almanac. Nevertheless, a number of civilian users have reported being able to track PRN32 since then.
The setting of satellite SVN23/PRN32 to healthy is also notable because in GPS satellite terms, it’s coming back from the dead, or at least from decommissioning. It is the first Block IIA satellite, launched on Nov. 26, 1990, and initially decommissioned on Feb.13, 2004, after more than 13 years of service. Apparently the U.S. Air Force reactivated it at some point and set it to broadcast a non-standard code that could not be tracked by standard GPS receivers. However, on Dec. 2, 2006, it started to transmit the standard PRN32 code as part of the Air Force’s initial test.