How GPS Keeps Time: Atomic Clocks, Relativity, 18 s

GPS keeps time with atomic clocks on every satellite, steered to a single system time called GPS Time that has counted continuously since January 6, 1980 and never adds leap seconds. Because UTC has added 18 leap seconds since then, GPS Time is now 18 seconds ahead of UTC, and every receiver uses a correction broadcast by the satellites to show civil time. The satellite clocks are also deliberately tuned slow before launch, by about 38 microseconds a day, to cancel the effects of relativity. This guide explains each of these pieces with the official numbers and our own check of the arithmetic. Facts are as of November 2026.

Key facts

  • “Each GPS satellite contains multiple atomic clocks”, and users can determine the time “to within 100 billionths of a second” (GPS.gov).
  • “GPS is ahead of UTC by 18 seconds” and “TAI is ahead of GPS by 19 seconds” (US Naval Observatory).
  • Satellite clock rates are offset by Δf/f = −4.4647 × 10−10, so the 10.23 MHz reference becomes 10.2299999954326 MHz (IS-GPS-200). That is 38.6 microseconds per day.
  • The GPS week counter reset to zero “at 23:59:42 UTC on April 6, 2019” (GPS.gov). By our count the next reset of the old 10-bit counter is on November 20, 2038.

Atomic clocks in orbit

GPS works by timing. Each satellite broadcasts where it is and what time its clock says, and a receiver compares those times with its own to work out how far away each satellite is: a radio signal travels about 30 metres in 100 billionths of a second. GPS.gov says “Each GPS satellite contains multiple atomic clocks that contribute very precise time data to the GPS signals”, which “enables users to determine the time to within 100 billionths of a second” (GPS.gov: Timing). A cheap receiver does not need an atomic clock of its own: in the standard navigation solution it treats its own clock error as a fourth unknown, next to the three position coordinates, which is why it needs signals from at least four satellites for a position fix. That is our summary of the method, not a quote from an official source.

How atomic clocks themselves work is covered in our guide to how atomic clocks work.

GPS Time, UTC and TAI

GPS Time started equal to UTC at midnight at the start of January 6, 1980 (the GPS.gov rollover page counts weeks from “January 5, 1980”, meaning the end of that day). At that moment, International Atomic Time (TAI) was 19 seconds ahead of UTC, according to the leap second list in the tz database. GPS Time has never added a leap second, so it stays exactly 19 seconds behind TAI, while UTC has fallen further behind with each leap second:

Time scale Offset from UTC in November 2026 Adds leap seconds?
UTC 0 s Yes, until May 2027 (see below)
GPS Time +18 s No
TAI +37 s No

The arithmetic checks both ways: TAI − UTC is 37 s today and was 19 s in 1980, so 37 − 19 = 18 s, matching the USNO’s “GPS is ahead of UTC by 18 seconds” (USNO), and there have been 18 leap seconds since January 6, 1980 in the list. Receivers convert GPS Time to UTC using the offset the satellites broadcast, so your phone shows civil time, not GPS Time.

The gap should now stay at 18 s. The 28th General Conference on Weights and Measures decided in October 2026 “that continuous UTC will become effective on 20 May 2027”, meaning no more leap seconds are planned (CGPM 2026, Resolution 5). Our leap seconds guide has the full list, and UTC vs GMT explains the time scales.

Relativity: why the clocks are set slow

The satellites’ clocks run at a different rate from clocks on the ground for two reasons. Their weaker gravity, high above the Earth, makes them run faster (general relativity), and their orbital speed makes them run slower (special relativity). Neil Ashby’s review in Living Reviews in Relativity describes these as “gravitational and motional frequency shifts which are so large” that they must be accounted for (Ashby, Relativity in the Global Positioning System).

The official GPS interface specification handles it before launch: “The clock rates are offset by Δf/f = -4.4647 × 10-10”, which “is equal to 10.2299999954326 MHz” instead of the nominal 10.23 MHz (IS-GPS-200, section 3.3.1.1). Multiplied by the 86,400 seconds in a day, that offset is 38.58 microseconds per day.

Our independent check. Using standard constants (GM = 3.986004418 × 1014 m³/s², the geoid potential W0 = 62,636,856 m²/s², c = 299,792,458 m/s) and a nominal orbit radius of about 26,560 km for a circular orbit. That radius is the reference semi-major axis in IS-GPS-200 (AREF = 26,559,710 m), and it matches GPS.gov’s “altitude of approximately 20,200 km” plus the Earth’s radius (GPS.gov: Space segment):

Effect Rate Per day
Weaker gravity (clock runs faster) +5.299e-10 +45.8 µs
Orbital speed (clock runs slower) −8.349e-11 −7.2 µs
Net +4.4646e-10 +38.6 µs

Our net figure of 4.4646e-10 matches the specification’s 4.4647 × 10−10 to within 0.003%, which is as close as a circular-orbit sketch should get. In light-travel terms, 38.6 microseconds is about 11.6 km: a sense of how large the effect is compared with the timing GPS needs. The orbits are slightly elliptical, so receivers also apply a small periodic correction from the broadcast orbit data.

The GPS week number rollover

GPS signals carry the date as a week number plus seconds into the week. In the original navigation message, GPS.gov says the system “broadcasts the GPS week number using a 10-bit code with a maximum value of 1,023 weeks”, so the counter wraps to zero every 1,024 weeks (about 19.6 years). It “reset to zero at 23:59:42 UTC on April 6, 2019”, and this “occurred for the first time in August 1999” (GPS.gov: Week number rollover).

We checked the dates by adding 1,024-week steps to January 6, 1980 in GPS Time and converting to UTC with the leap second list:

Rollover GPS weeks since 1980 UTC
First 1,024 August 21, 1999, 23:59:47
Second 2,048 April 6, 2019, 23:59:42
Third (projected) 3,072 November 20, 2038, 23:59:42

The 2019 time matches GPS.gov to the second, which confirms the method; the 1999 rollover came at 23:59:47 UTC because only 13 leap seconds had been added by then. The 2038 date assumes the GPS-UTC offset stays at 18 s. Old receivers that only understand the 10-bit counter can jump back 19.6 years at a rollover; this is a cousin of the Unix year 2038 problem, which falls on January 19, 2038.

Who relies on GPS time

Positioning is only half the story. GPS.gov lists timing users directly: “wireless telephone and data networks use GPS time to keep all of their base stations in perfect synchronization”; “Power companies and utilities have fundamental requirements for time and frequency”; and “Major financial institutions use GPS to obtain precise time for setting internal clocks” (GPS.gov: Timing). GPS.gov adds that “National laboratories around the world use this ‘common view’ technique to compare their time scales”.

What this means for you

  • Your phone’s clock may come from the mobile network or from internet time servers rather than from GPS directly; GPS.gov notes that the networks themselves use GPS time to keep base stations in sync.
  • Raw GPS logs (drones, survey kit, dashcams) may store GPS Time. Subtract 18 s to get UTC for any timestamp after January 1, 2017.
  • Time zones are applied last. GPS gives UTC; your device then applies its time zone rules, which is why a device with an old time zone database can show the right UTC but the wrong local time. Check yours against our time zones pages or the current UTC time.

How we checked

We read the GPS.gov, USNO and IS-GPS-200 texts and the CGPM 2026 resolution. The leap second offsets come from leap-seconds.list in tzdata 2026e. The rollover dates were computed in Python by adding 1,024-week steps to the 1980 epoch and converting GPS Time to UTC; the 2019 result reproduces GPS.gov’s 23:59:42 UTC exactly. The relativity table uses standard physical constants and a circular-orbit approximation; the full treatment is in Ashby’s review.

Sources

Frequently asked questions

Is GPS time the same as UTC?

No. GPS Time is 18 seconds ahead of UTC because it does not add leap seconds. Receivers correct for it.

How accurate is GPS time?

GPS.gov says users can determine the time to within 100 billionths of a second (100 nanoseconds).

Why do GPS clocks need relativity corrections?

In orbit, weaker gravity makes clocks run faster and speed makes them run slower. The net effect is about 38.6 microseconds a day, which is removed by setting the satellite clocks slightly slow.

When did GPS time start?

At 00:00 UTC on January 6, 1980.

When is the next GPS week rollover?

For the original 10-bit week counter, on November 20, 2038, by our calculation.

Similar Posts