A satellite can help compare two clocks without being the clock you are trying to measure.
Two laboratories want to know whether their clocks differ by a few nanoseconds. Each can observe the same navigation satellite. Subtract the two local comparisons and the common satellite time contribution can largely cancel. What remains can still include unequal receiving equipment and propagation errors.
Common-view GNSS time transfer uses matched observations to compare clocks at different places. This guide follows one pair of readings all the way to a calibrated difference, then changes the satellite, signal and equipment to see which parts of the original cancellation survive.
The result is a time comparison, not a position fix or an automatic statement about UTC. You can work the examples with subtraction and units. The useful habit is to separate the reference that is shared from the paths that are not.
Geodesy subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1 · Chapters 1–2: Follow the shared reference
Build one-way comparisons and subtract the matched pair.
Route 2 · Chapters 3–5: Find what stays in the difference
Do not cancel the two propagation paths by name
Inspect path effects, equipment delays and their calibration.
Route 3 · Chapters 6–8: Handle changes without hiding steps
Change the signal and predict the apparent step
Compare signals, constellations and observation windows.
Route 4 · Chapters 9–10: Check the interpretation
Separate a time difference from a rate difference
Separate time offset from rate and carry the right uncertainty.
Route 5 · Chapters 11–12: Transfer and report
Transfer the logic to a link with a new interruption
Test a changed link and write a comparison another laboratory can trace.
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Follow the shared reference
Use one common messenger
Write a signed comparison you can reconstruct
Find what stays in the difference
Do not cancel the two propagation paths by name
Locate the point where the local clock enters
Use a common-clock test to reveal differential hardware
Handle changes without hiding steps
Change the signal and predict the apparent step
Treat another constellation as another reference and signal system
Match what the time tags actually represent
CHAPTER 1 OF 12 · Follow the shared reference
1. Use one common messenger
Imagine that two people in separate rooms hear a single bell and compare the event with their own clocks. If the sound reaches the rooms at different times, they must account for the different travel times. Once those paths are addressed, subtracting the two local comparisons can tell them something about how far apart their clocks are. They do not need the bell itself to be a perfect clock.
In GNSS common view, two receiving stations observe a common satellite over a matched observation interval. Each station relates its local clock to the satellite signal using an observation and propagation model. The two comparisons are then combined with the right signs. In the ideal shared-reference calculation the same satellite clock term appears in both and cancels.
A satellite observation is not a direct view of someone else’s clock face. The observation depends on when the signal was transmitted, the path to the receiver, and the receiving chain that delivers a comparison to the local clock reference. Time and event conventions matter, especially when the desired accuracy is small.
Distinguish this from taking two independent one-way timing outputs at unrelated times and subtracting their displayed values. The common-view cancellation is supported by using the same satellite contribution in compatible observations. If the comparison does not actually share that contribution, the algebra does not make it shared.
CHAPTER 2 OF 12 · Follow the shared reference
2. Write a signed comparison you can reconstruct
Use a teaching convention: after the named geometric and propagation models are applied, the reading at site i is local clock minus satellite clock plus a receiving delay contribution di and a residual ei. The sign of a real system’s calibration field depends on how its measurement and corrections are defined. These examples use the declared sign, not a universal file convention.
Suppose the modelled comparison at A is +32 ns and that at B is +19 ns for the same satellite and matched interval. A minus B gives +13 ns before differential equipment calibration. The satellite clock contribution disappears in this ideal subtraction because it occurs with the same sign in both local comparisons.
Let the effective receiving contributions under this convention be +5 ns at A and +2 ns at B. Their difference is +3 ns. Subtract it from +13 ns to estimate A minus B = +10 ns, assuming the remaining errors are negligible in this toy case. The sign says A is ahead of B under the convention, not that its oscillator necessarily runs faster.
Reverse the comparison as a check. B minus A before calibration is −13 ns; the delay difference is −3 ns; the corrected result is −10 ns. If the direction changes but the result does not reverse, review the ordering or sign of the correction before interpreting a clock discrepancy.
CHAPTER 3 OF 12 · Find what stays in the difference
3. Do not cancel the two propagation paths by name
The stations see different paths even when observing the same satellite. Their geometric distances differ. Residual atmosphere and multipath can differ as well. Common-view processing models travel times and can reduce effects that are sufficiently common, but the remaining differential terms are not automatically zero.
One metre of unmodelled path length corresponds to about 3.34 ns in one-way light time. A 0.30 m path difference corresponds to roughly 1.0 ns. This conversion helps put an apparent clock change in perspective. It does not imply that any particular mismatch is caused by geometry or that an instrument error of the same numerical size is interchangeable.
Imagine two station position errors. At the first site an error projects by +0.20 m onto the relevant range direction; at the second it projects by +0.05 m in the same sense. The differential scale is 0.15 m, or about 0.50 ns of light time before applying the sign of the observation model. If the satellites or geometry change, the projections can change too.
Write which propagation terms were already applied and which residual uncertainty remains. Shorter baselines may make some effects more correlated in a given setup, but do not rely on that as a substitute for checking the actual geometry and observations. Matching the satellite is the beginning of the comparison, not its entire error budget.
CHAPTER 4 OF 12 · Find what stays in the difference
4. Locate the point where the local clock enters
A receiving chain can include an antenna, cable, receiver, timing input, distribution electronics and a reference point at which the local clock is compared. The relevant calibration applies to a defined chain and signal or observable. A label saying receiver delay may not include every cable or reference distribution term you need.
Imagine the antenna cable is replaced and the effective signed chain contribution at A rises by +4 ns. With the clocks unchanged, the uncalibrated A-minus-B comparison rises by +4 ns in the example convention. If you continue using the old delay value, the change masquerades as a clock step.
The reverse can happen too. If a step is real but you apply an invented equipment correction chosen solely to flatten the time series, the clock event disappears from the result. Use maintenance logs, parallel observations and calibration evidence to distinguish the possibilities. A smoother output is not an independent reason to change a delay.
Retain equipment identities, cable and configuration changes, signal definitions, reference plane and the period over which each calibration applies. When the receiving chain changes, write down what was changed and what evidence carries the calibration across the change. A calibrated receiver on the wrong chain is not a complete link calibration.
CHAPTER 5 OF 12 · Find what stays in the difference
5. Use a common-clock test to reveal differential hardware
A simplified calibration comparison can connect two receiving setups to the same known local clock reference, under a defined geometry and cable arrangement. Because their physical clock difference is zero at the declared comparison plane, a persistent measured difference can inform the differential delay under those conditions. Other path and configuration differences still need attention.
Suppose two receivers on the same reference show modelled comparisons +27 and +23 ns during a controlled test. Their +4 ns difference is consistent with a differential contribution of +4 ns in this example. It does not reveal the absolute delay of either receiver. Adding +10 ns to both unknown receiver contributions would leave their difference unchanged.
If a calibrated transfer setup is moved between laboratories, its own stability, setup and reference connection matter. A change in the travelling chain can become entangled with the site differences. Before treating the transfer as a clean independent check, document where the calibration comes from and how repeatability was assessed.
Do not assume all outputs of one receiver share the same delay. A code observation, an ionosphere-free code combination, or a different signal can be affected differently by hardware and processing. Check whether the calibration explicitly supports the quantity you are using, not merely the same model number on the front panel.
CHAPTER 6 OF 12 · Handle changes without hiding steps
6. Change the signal and predict the apparent step
Return to the first example. The real clock difference is +10 ns. Signal X has effective receiving contributions of +5 and +2 ns, so the uncalibrated difference is +13 ns. On signal Y let those contributions be +8 and +1 ns. The uncalibrated difference becomes +17 ns even though the clocks did not move.
If you subtract the signal-X calibration difference of +3 ns from the Y comparison, you report +14 ns and invent a +4 ns clock step. Applying the supported Y difference of +7 ns instead returns +10 ns. This does not make the toy calibration universal; it shows why signal identity is part of the result.
A useful overlap test collects both comparisons during a period when the same physical clock difference should apply. After the named calibrations and other models, compare the two. A persistent difference suggests a signal-dependent issue or a model mismatch. If the two estimates share much of their processing, agreement may be less informative than it first looks.
If an overlap period is missing, report the boundary as a change in the measurement system. Do not silently stitch the records together by choosing whatever offset makes them match. You can present separate segments and state what additional calibration or comparison would be needed to support a continuous result.
CHAPTER 7 OF 12 · Handle changes without hiding steps
7. Treat another constellation as another reference and signal system
Switching from one constellation to another can change system time conventions, observables and hardware contributions. For a clean matched comparison within one satellite and signal, an explicitly common clock term can cancel under the model. Mixing unmatched outputs across constellations is not the same operation.
Suppose A compares its clock with satellite X and B with satellite Y. Subtracting their records leaves a difference involving the two satellite clock contributions unless a consistent model explicitly accounts for them. Even when they are in the same constellation, they are different satellites. Sharing the GNSS label is not enough.
For multi-constellation processing, keep the time-system and receiver-delay conventions explicit. An inter-system term used for positioning may bundle effects under that estimator’s reference convention. It is not automatically a portable calibration of the receiving chain for time transfer. Use the quantity and evidence intended for the time comparison.
A practical test is to produce separate calibrated clock-difference estimates from each supported signal or system during an overlap. Compare them at matched epochs before combining them. If one shows a step when an observable changes, preserve the component results so the combined series does not conceal the discrepancy.
CHAPTER 8 OF 12 · Handle changes without hiding steps
8. Match what the time tags actually represent
An observation record can name a tracking interval, a reference epoch or a reported average. When comparing stations, match the convention and time support, not just a printed date. A result evaluated at one epoch and another average over a different interval can disagree because the clocks or geometry changed between them.
For a simple example, suppose A minus B changes at a steady +0.20 ns per minute during a short experiment. Comparing values separated by 15 minutes without alignment introduces 3.0 ns of difference from the elapsed time alone. This is not a correction to copy into real data; it is the scale of a timing mismatch in the stated scenario.
Signal events are not identical simply because both receivers recorded at the same nominal local time. The physical transmission and reception events are related by travel time and the model conventions. Use the matching and processing rules for the actual data format. The teaching subtraction assumes that the shared satellite contribution was made comparable.
Keep missing intervals visible. If the stations did not observe a suitable common satellite at a given time, there may be no direct common-view comparison for that interval. An interpolated clock series can be useful for another purpose, but label it. A continuous plot is not evidence that every value came from a measured pair.
CHAPTER 9 OF 12 · Check the interpretation
9. Separate a time difference from a rate difference
A time offset describes how far apart two readings are at a stated epoch. A rate difference describes how that offset changes. If A is ahead of B by +10 ns in the morning and remains ahead by +10 ns later, there is no evidence of a rate difference from those two numbers alone.
Suppose the offset rises from +10 ns to +22 ns in 600 seconds under unchanged measurement conditions. The change is +12 ns, giving an average fractional rate difference of 12 × 10⁻⁹/600 = 2 × 10⁻¹¹ in the toy example. This is a slope calculation, not a guarantee about the oscillator’s long-term behaviour.
Now suppose a receiver delay changed by +12 ns halfway through while the clocks kept a constant difference. The same two endpoint readings would produce the same apparent slope. More frequent observations, event logs and an independent comparison could reveal a step instead of a gradual rate change.
When reporting rate, state the interval, model and whether equipment steps were tested. A timing link can reproduce a short-term slope and still have an uncertain constant offset. Conversely a known calibration offset does not establish stability over time. Name the performance quantity the evidence supports.
CHAPTER 10 OF 12 · Check the interpretation
10. Carry the uncertainty of the shared and unshared terms
If two comparisons use a genuinely common additive error, subtracting them removes it under the simple model. Errors that are only partly shared can be reduced, unchanged or behave differently depending on their covariance. The uncertainty of A minus B includes the variance of each and subtracts twice their covariance.
Use a fictional pair with standard uncertainties of 2 ns at A and 3 ns at B, and covariance +3 ns² in the two comparison errors. Their difference has variance 4 + 9 − 6 = 7 ns², or about 2.65 ns standard uncertainty. Assuming independence would give √13, about 3.61 ns.
Now add an independent differential calibration uncertainty of 1 ns in the example. The total standard uncertainty becomes √8, about 2.83 ns. Do not add it twice as if it were a separate one-nanosecond quantity at each end; it was already specified for the difference. If calibration and observation terms are dependent, a different propagation is needed.
Repeated comparisons may improve estimates of random variability without reducing a persistent calibration error. The number of samples does not divide away a shared receiver delay. Report how uncertainty was estimated and which terms persist across samples. A nanosecond display resolution is not a nanosecond uncertainty statement.
CHAPTER 11 OF 12 · Transfer and report
11. Transfer the logic to a link with a new interruption
Imagine two clocks compared by pulses distributed through two cables from one signal source. Subtracting the comparisons can cancel a common source time offset. It cannot cancel an extra length added to just one cable. The analogy is simple, but it captures why a changed path can imitate a changed clock even when the source is shared.
Return to the satellite comparison. During an outage the two stations change receiver software and no longer select the same signal. Their first post-outage A-minus-B value differs from the last pre-outage value. Check continuity of the physical clocks, the receiving chains, the time tags and observable definitions before interpreting the jump.
Now give them a third clock and a separate link. If two links disagree only when they involve the changed receiver, that pattern can guide an investigation. It does not automatically identify the component, because those links may share processing or reference inputs. Construct a test that changes the suspected cause without changing everything else.
Keep the practical question specific. Are you trying to restore a continuous record, compare a new interval or decide whether the local oscillator actually changed? Those goals require different evidence. It is better to report a bounded comparison after a switch than to imply continuity that has not been demonstrated.
CHAPTER 12 OF 12 · Transfer and report
12. Write the result with a clear reference and direction
A useful handover identifies the two clocks or reference planes, the sign of the difference, the epoch or interval and the uncertainty. It names the observed satellites and signals, applied models, calibration versions and any equipment changes. If a value is tied to a time scale, state the traceability of that tie rather than inferring it from the satellite name.
A bounded example would read: “At the matched epoch, clock A minus clock B was estimated at +10 ns using the stated signal and link calibration. A switch to a second signal was tested during overlap. The comparison does not establish either clock’s absolute difference from UTC without the separate reference information.” Use only the parts your own evidence supports.
Retain the raw or processed component comparisons alongside the final difference. A later reviewer can then reproduce the sign, identify the observations that were common and inspect why a particular segment was rejected. A final number with no provenance makes a repair much harder when the equipment changes.
The point of common view is not to make every error vanish. It is to use a shared observation to remove a specified common contribution, then control and document what remains. Once those two jobs are separated, a clock difference becomes something another observer can test.
Independent worked case and evidence checks
Two fictional stations make matched observations of one signal. A records +51 ns relative to the modelled satellite clock and B records +34 ns. The uncalibrated difference is +17 ns. The approved effective receiving contributions are +6 ns and +3 ns under the stated sign convention, giving a differential correction of +3 ns and a calibrated A-minus-B estimate of +14 ns.
During overlap, a second signal gives uncalibrated +21 ns and has a supported differential correction of +7 ns. Its corrected result is also +14 ns. Agreement supports compatibility in this test. It does not prove the absolute calibration if both signals share an unrecognised contribution from the same receiving reference.
Now change the setting. A replaces a cable and the second signal’s uncalibrated difference rises to +25 ns. A separate cable and chain check supports a +4 ns increase in the differential contribution. The revised difference is +11 ns and the calibrated result remains +14 ns. Without that separate evidence, the +4 ns step would remain ambiguous.
For a changed-context check, suppose the second record came from a different time window and A minus B was independently found to drift +0.10 ns per minute. A 30-minute mismatch could contribute 3 ns. Do not label it a hardware delay without aligning the comparison first. The same numerical disagreement can have a different explanation when the time support changes.
If you compare the same pair with both signals, resist treating the two uncertainties as independent by default. Shared local distribution, site coordinates and reference-clock behaviour can affect both. Agreement across signals is valuable evidence for some signal-dependent errors; a separately calibrated comparison with a different path tests a wider set of shared dependencies.
Worked comparison
| Question | What to compare | What is not established |
|---|---|---|
| Did a satellite contribution cancel? | Matched satellite, observable and time conventions | All path and equipment effects cancel |
| Did the clock step? | Overlapping signals and independent calibration | A step in one output is physical |
| Do the clocks run at different rates? | Change of aligned offset over time | A single nonzero offset is drift |
| Is the result tied to UTC? | Traceable clock reference and link documentation | Common view alone supplies absolute UTC |
Questions and practical checks
Do the laboratories need the satellite clock to be perfect?
No. In a compatible common-view subtraction, the shared satellite clock contribution can cancel. Other terms still need modelling and calibration.
Can I subtract results from different satellites?
Not as the same simple common-view cancellation. Different satellite clock contributions must be handled explicitly if the observations are unmatched.
Why can a signal change look like a clock jump?
The effective receiving contribution can depend on the signal or observable. An old calibration can leave a new differential delay in the clock estimate.
Is a common-clock test an absolute calibration?
It can inform a differential delay under its conditions. It does not determine two separate absolute receiver delays by subtraction alone.
Does the clock comparison establish absolute UTC at both sites?
No. It gives a relative result under the stated link model. An absolute or traceable time claim needs the appropriate reference information.
Continue with the connected subjects
Understand clock terms and signal events
Multi-constellation clock biases: distinguish a positioning reference from an equipment calibration
Signal emission times: place the transmitter and receiver at the right events
Broadcast clock polynomials: keep clock reference epochs and conventions visible
Gravitational signal delay: separate a path correction from a clock term
Compare differential measurements and time scales
GNSS double differences: see how shared terms cancel and errors become correlated
Optical clock comparisons: distinguish timing measurements from a height inference
VLBI Intensives: distinguish UT1–UTC from an ordinary clock comparison
PPP outages: separate signal loss from product interruptions
Sources and scope
NIST: common-view GNSS time transfer and matched comparisons
NIST primary study: receiver delay for an ionosphere-free GPS code observable
NIST: GNSS and satellite time-transfer methods
The numerical cases in this guide are original fictional teaching examples, not measurements from a real site. This is an educational explanation, not a field safety instruction, engineering approval or an official certification. For a real project, use qualified practitioners, current product documentation and site-specific evidence. Sources were checked on 4 November 2026.