Two or three sentences each. Where a term deserves more than that, the entry links to the page that gives it more — here if we have written one, and out to the canonical definition where somebody has already done it properly.
How close a measured value is to the true value. In the metrologist's vocabulary this is not one axis but two combined — trueness (freedom from systematic error) and precision (freedom from random error) — which is why "accurate to 15 ns" is ambiguous until you say which of the two you are claiming.
TruenessPrecisionSeen side by sideVIM 2.13, measurement accuracy
The standard measure of frequency stability, written ADEV(τ) because it is a function of the averaging interval τ rather than a single number — a source can be excellent at 1 s and poor at 10,000 s. MDEV(τ), the modified form, is used where you need to distinguish white phase noise from flicker phase noise, which plain ADEV cannot separate.
The distance between two GNSS stations whose observations are being compared. It matters because every differential technique works by cancelling the errors the two stations share, and sharing decays with distance: at zero baseline — two receivers on one antenna through a splitter — everything cancels except the receivers themselves, which is what makes it the sharpest way to characterize a receiver. Over a few tens of kilometers the atmosphere still largely cancels; over thousands, only the satellite's own errors do, and the ionosphere becomes the limit.
Satellite jargon for a relay function: content goes up, comes back down, and the satellite is neither its source nor its consumer — it does not read what it is carrying, act on it, or care what it says. Worth one caveat, since we use the phrase on this site: GNSS satellites do not relay the way most satellites do. There is no transponder amplifying and frequency-shifting an uplink. The navigation message — ephemeris and clock corrections — is uploaded from the ground segment, stored, and then re-encoded onto a signal the satellite generates itself, so "regeneration" is the more accurate word for the mechanism. What survives, and what the phrase is doing when we use it, is the distinction between where the content comes from and where the signal comes from: the corrections are computed on the ground, while the ranging and carrier-phase signals carrying them originate on board, timed by the satellite's own free-running clock.
What are the limits of GNSS time accuracy?How does GNSS time relate to UTC?
The file format national laboratories use to exchange time-transfer measurements, one line per satellite track, carrying the offset between the satellite's signal and the local clock. Its header states every delay in the chain — antenna cable, receiver internal, reference — which is what makes a CGGTTS file self-describing enough for somebody else to use. Version 2E covers every current constellation.
An error that affects every measurement in a set equally, and therefore cancels exactly when you take a difference between any two of them. A set of clocks that is uniformly 100 ns late has a 100 ns common-mode error and still measures intervals between its own clocks perfectly. The cancelling stops the moment you compare against a clock outside the set.
The arbitrary origin a set of measurements is expressed against, fixed by convention because the observations themselves only ever determine differences. It appears twice in this subject and it is the same idea both times: a position datum such as ITRF2020 says where coordinates are measured from, and a clock datum is the constant every satellite clock in a correction stream is offset by. Neither is discoverable from the data — somebody has to declare it.
GNSS Analysis CenterWhy differencing two solutions cancels itPPP-AR
A named instant — and the confusion is real, because the word carries two senses that pull in opposite directions. The origin sense is a fixed starting point a timescale counts from: the epoch of GPS Time is 1980-01-06, of Unix time 1970-01-01. The measurement sense is one sample instant in a stream of them: a GNSS receiver reports an epoch of observations — every satellite measured at that moment — and the next one arrives however far apart the sampling rate says. It reads as a synonym for "second" only because 1 Hz is the common sampling rate; a 30-second file has 2880 epochs in a day, and an epoch need not land on a second boundary at all. An epoch is a point, never a duration.
Timescales — where the origin sense livesRINEX, where you meet epochs one block at a timePPS, which marks a moment without naming itIGS, RINEX 3.05 — epoch records in observation files
An institution that takes raw observations from a global network of tracking stations and works out where each satellite really was and what its clock really read, then publishes the answer as a correction stream. The IGS lists thirteen of them, including CNES, JPL, Wuhan University and the U.S. Naval Observatory. Say GNSS Analysis Center on first use — "analysis center" alone means something different in almost every other field.
Where corrections come fromPPP-AR, which needs one center's stream and not a mixtureDatumIGS, list of Analysis Centers
A rack-mounted appliance that turns an antenna into NTP and PTP on your network. There is a GNSS receiver module inside it, but as a hidden component: the vendor has already wrapped it in an oscillator, a discipline loop and a protocol stack, so you never see the module's raw PPS jitter or its observation stream. You are buying the clock rather than the parts — which is why the specifications that matter for a module are largely meaningless here, and vice versa.
Which traits matter for whichThe protocols it exists to speakHoldover, which is what its oscillator is really for
What a clock does once its reference disappears: it runs free on its own oscillator and accumulates error. Not the same as merely undisciplined — a clock in holdover was disciplined until recently, so it knows its own frequency offset and can correct for it, which one that never had a reference cannot. A holdover figure is meaningless without both the error threshold and the oscillator it assumes.
A national metrology institute — the body that maintains a country's measurement standards, such as NIST in the United States or NPL in the United Kingdom. Worth knowing when reading about timekeeping, but note that not every contributor of a UTC(k) is an NMI: the BIPM says "institutes", and the list includes observatories and designated laboratories as well.
My own research engine for real-time GNSS timing: it takes raw observations and a single GNSS Analysis Center's correction stream and solves for the receiver clock, pinning a surveyed antenna position rather than estimating it. The name is for Persistent PPP-AR. It is where the ambiguity-resolving path on this site gets tested, and there is no commercial product doing the same thing yet.
A single-receiver solution that adds the carrier phase to the pseudorange and replaces the broadcast corrections with precise orbit and clock corrections from an analysis center. The carrier is a far quieter observable, but it carries an unknown whole number of cycles, and plain PPP estimates that unknown as a real number — hence float PPP — which is why it needs tens of minutes to converge.
SPPPPP-ARWhere the corrections come fromNavipedia, Precise Point Positioning
PPP in which the carrier-phase ambiguity is fixed to the integer it physically is, rather than estimated as a float. Doing so needs satellite phase biases as well as orbits and clocks, and those are only self-consistent when they come from a single analysis center — mixing correction sources destroys the property that makes fixing possible. The payoff is faster convergence and a better solution from the same antenna.
PPPMatching acquisition and distributionNavipedia, Carrier Phase Ambiguity Fixing
A once-per-second electrical edge marking the start of a second. It is the lingua franca between timing equipment — almost every receiver emits one and almost every instrument accepts one — and its limitation is exactly its simplicity: it marks a moment without naming it, and it compensates for nothing. The pulse says now; it does not say which second, and it carries no correction for the cable it travelled down.
What the protocols do that a PPS cannotResolution, and why a quantized edge is not a precise oneWhat makes an accurate timing receiver
How closely repeated measurements agree with each other, whether or not they agree with the truth. Averaging improves precision, and that is the whole of what averaging does — a tight cluster in the wrong place is precise and untrue.
TruenessThe limits of averagingVIM 2.15, measurement precision
Precision measured under conditions held as constant as you can manage: the same procedure, operator, instrument and location, over a short enough period that nothing drifts. Its counterpart, reproducibility, deliberately varies those things — so a result that repeats well but reproduces badly is telling you something about your setup rather than about the thing you are measuring.
The smallest change in the measured quantity that produces a perceptible change in what the instrument reports. It bounds what you can see, not what is true — a reading with twelve digits can be twelve digits of confidently reported nonsense. And resolution coarser than your actual scatter conceals that scatter: a counter that cannot measure intervals smaller than 10 ns, watching a clock that jitters by 1 ns, may see no jitter at all.
PrecisionTruenessWhat coarse resolution hidesVIM 4.14, resolution
The universal text format for raw GNSS observations — one epoch per block, every satellite's code and carrier-phase measurement written out as the receiver saw them, before any solution. Its whole value is in the name: a RINEX file from any manufacturer's receiver can be processed by anybody's software, which is what makes it possible to reprocess somebody else's observations years later, or to compare two clocks whose owners never agreed on anything but the format.
CGGTTS, the time-transfer counterpartComparing distant clocksIGS, RINEX 3.05 format specification
The baseline GNSS solution: pseudoranges from four or more satellites, corrected only by what the satellites themselves broadcast. It needs no internet connection, no subscription and no convergence time, and it is what every receiver falls back to. The price is that its accuracy is capped by the freshness of the broadcast orbit and clock corrections — refreshed every ten minutes to two hours — which puts it in the tens of nanoseconds.
Why GNSS time is a predictionPPPNavipedia, Code Based Positioning (SPS)
In time and frequency, how well a source holds its output over a given averaging interval — always a function of that interval, never a single number, because the answer depends on how long you watch. Two families, depending on what you are holding steady: frequency stability by [ADEV or MDEV](#adev), time stability by [TDEV](#tdev). A clock can be superbly stable and still be wrong, which is why stability and accuracy are quoted separately.
VIM 4.19, stability of a measuring instrument (the general sense)
The rule of thumb that your measuring equipment should be about ten times better than the error you are chasing, so its own contribution can be ignored rather than budgeted. Beware the folklore: the ratio the standards actually require is 4:1, not 10:1 — ANSI/NCSL Z540-1 caps the measurement standard's uncertainty at 25% of the tolerance — and modern practice has replaced the accuracy ratio (TAR) with the uncertainty ratio (TUR), which accounts for the whole measurement process rather than a headline spec.
A way to recover the stability of each individual clock from measurements that can only ever show differences. Compare three clocks pairwise — A−B, B−C, A−C — and if their errors are uncorrelated, the three measured variances are three equations in the three unknowns, so each clock's own variance falls out: σ²A = ½[σ²(A−B) + σ²(A−C) − σ²(B−C)]. The catch is the word uncorrelated. A shared antenna, a shared reference, a shared room or a shared afternoon all violate it, and the method has a memorable way of saying so: it returns a negative variance, which is impossible, and which means the assumption rather than the arithmetic was wrong.
Questions to ask when benchmarking clocksADEVCommon-modeGray & Allan (1974), A method for estimating the frequency stability of an individual oscillator
The counterpart of ADEV for time rather than frequency stability, also a function of averaging interval, written TDEV(τ). It is derived from the modified Allan deviation — TDEV(τ) = τ·MDEV(τ)/√3 — so the two describe the same underlying noise in the units you happen to care about.
A way of naming instants, built from three things: the length of a tick, a zero point, and a count of ticks since. UTC and TAI are timescales; so, absurdly but perfectly validly, is years since Victoria's reign.
NIST's remote time calibration service. NIST installs a GNSS receiver at your site — the equipment stays theirs — takes the 1 PPS output of your clock, and compares it against UTC(NIST), reporting every ten minutes to a web page. The base tier measures rather than steers, and what it really sells is a calibrated end of the link and a report with NIST's uncertainty on it.
Could you build the same link yourself?TraceabilityNIST, Time Measurement and Analysis Service
The property of a result that it can be related to a stated reference through a documented, unbroken chain of calibrations, each one contributing to the total uncertainty. Traceable is not a synonym for accurate: it means the path back to the definition is known and its uncertainty accounted for, not that the number is good.
What "traceable to UTC" actually requiresVIM 2.41, metrological traceability
How close the average of infinitely many repeated measurements sits to the reference value — in datacenter timing, to UTC. It is inversely related to systematic error and unrelated to random error, which is the formal way of saying that no amount of averaging will ever fix a bias. Mind the grammar: your offset from UTC is the systematic error, a quantity you can state and defend; trueness is the property of that offset being small, and is not itself a number.
The world's civil timescale, computed by the [BIPM](https://www.bipm.org/) (the international bureau that keeps the world's measurement standards) from the clocks of about eighty contributing institutes, and published monthly in Circular T. It is a virtual timescale — nothing ticks it, and the literature's term of art, paper timescale, means the same thing. And it is a deferred-time scale: nobody can tell you what UTC reads right now, because this month's answer is settled next month. So everything that claims to distribute UTC in real time is distributing a realization of it, and every realization is necessarily a prediction.
A national institute's own realization of UTC — a physical clock that ticks, in contrast with [UTC](#utc) itself, which is virtual and ticks nowhere. It is the thing you can actually steer to. Because the virtual scale is settled in arrears, a realization cannot copy UTC and must instead predict it: Circular T publishes the difference [UTC − UTC(k)] every five days, so how good the prediction was is knowable, but only afterwards. BIPM's weekly UTCr narrows the wait without closing it.
The speed at which a signal travels down a cable, as a fraction of the speed of light in vacuum — typically about 0.66 for solid-polyethylene coax and 0.85 for foam. It is the number that converts a cable's length into its delay, which is the thing that actually matters, and it differs enough between coax types that assuming the wrong one is worth tens of nanoseconds on an ordinary run.