A changing received frequency does not directly report your speed
A receiver reports a Doppler shift for a navigation signal. The number reflects relative signal dynamics and clock effects, not simply how fast the receiver moves across the ground. Even a stationary receiver observes changing frequencies as satellites move. To infer receiver velocity, the model must separate those contributions using geometry.
This guide begins with a declared Doppler sign convention, converts frequency to a range-rate quantity, and solves a small original velocity-and-clock example. It then tests poor geometry, unequal noise and time-window interpretation. The aim is to make the observable-to-velocity relationship inspectable rather than turn one frequency reading into a speed claim.
All wavelengths, residual frequencies, velocities and uncertainties below are fictional teaching inputs. They are not receiver specifications or navigation instructions. Actual processing needs the relevant signal definitions, satellite states, timing and propagation models.
Geodesy subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1 · Chapters 1–2: Define frequency and range rate
Route 2 · Chapters 3–5: Separate motion and clock drift
Subtract the modelled satellite contribution before solving receiver motion
Route 3 · Chapters 6–8: Test geometry and uncertainty
Route 4 · Chapters 9–10: Inspect timing and continuity
Distinguish an instantaneous rate from a time-window average
Route 5 · Chapters 11–12: Transfer and report
Keep short-term velocity separate from long-term station motion
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Define frequency and range rate
Name the measured frequency and its convention
Convert hertz into a length-per-time quantity
Separate motion and clock drift
Subtract the modelled satellite contribution before solving receiver motion
Solve a one-dimensional velocity and clock example
Distinguish clock offset from clock drift
Test geometry and uncertainty
Use geometry to explain a non-unique solution
Propagate equal-noise observations through the solution
Keep unequal noise and shared errors visible
Inspect timing and continuity
Distinguish an instantaneous rate from a time-window average
Separate carrier continuity from derivative information
CHAPTER 1 OF 12 · Define frequency and range rate
1. Name the measured frequency and its convention
Doppler is a frequency-related observable extracted while tracking a signal. ESA Navipedia describes its connection with relative antenna motion and a common contribution proportional to receiver clock-frequency error. It is not a direct three-dimensional velocity vector printed independently of an observation model.
For this guide, define D as received frequency minus the compatible reference frequency, with positive D associated with an increasing received frequency under the simple geometric convention. Define positive geometric range rate as increasing separation. In the elementary low-speed relation used here, geometric range rate is −λD, where λ is the signal wavelength.
Actual data and receiver interfaces can use different phase and Doppler sign conventions. Do not overwrite a file’s interpretation with this classroom definition. Translate the documented observable first, then verify a synthetic approach-and-recession case. A sign copied from another product without its definition can produce a plausible magnitude in the wrong direction.
The model below is deliberately simplified. It assumes the relevant satellite-state, light-time, clock, frame and propagation contributions have been handled consistently or are explicitly omitted in a fictional exercise. It is not a complete relativistic signal-frequency model or a replacement for a validated GNSS velocity solution.
CHAPTER 2 OF 12 · Define frequency and range rate
2. Convert hertz into a length-per-time quantity
Hertz means cycles per second. Multiplying D by a wavelength in metres per cycle produces metres per second. Under our declared convention, q = −λD is a range-rate-like observable contribution. The wavelength must belong to the actual signal represented by the frequency measurement.
Use a fictional wavelength λ = 0.20 metres. A residual frequency D = +12.5 hertz becomes q = −2.5 metres per second. A residual D = −17.5 hertz becomes q = +3.5. These are converted residual observation quantities, not yet receiver velocity components or ground-speed estimates.
Now change only the fictional wavelength to 0.25 metres while retaining D = 12.5 hertz. The converted quantity is −3.125 metres per second. Equal frequency offsets at different wavelengths need not represent equal length rates. Conversely, the same physical range rate corresponds to different frequency shifts when wavelength changes.
Reverse the calculation: D = −q/λ. For q = −2.5 and λ = 0.20, the result is +12.5. This catches a missed negative sign or an incompatible wavelength. It does not check whether the residual frequency had already had satellite and clock contributions removed correctly; that is a separate model-interface question.
CHAPTER 3 OF 12 · Separate motion and clock drift
3. Subtract the modelled satellite contribution before solving receiver motion
In a simplified geometric account, the range derivative is u·(vs−vr), where u points from receiver to satellite and vs and vr are compatible satellite and receiver velocities. A stationary receiver can therefore have a nonzero range rate because vs is not zero. Satellite motion must not be interpreted as receiver movement.
After subtracting a compatible modelled satellite contribution and handling the other specified terms, write a reduced observation as y = −u·vr+b+ε. Here b is a common receiver-clock range-rate term in metres per second and ε represents remaining errors. The negative sign follows the receiver-to-satellite direction used here.
The geometry u is itself based on a position and state account. Doppler observations do not remove the need to know which directions the signals arrived from. A position error or incompatible satellite epoch can affect the modelled direction and satellite contribution, feeding into the inferred receiver velocity.
The existing orbit and emission-time guides explain those upstream state questions. This article concentrates on separating receiver motion from a common clock-drift contribution after those interfaces have been established. A correctly converted hertz value can still be unsuitable if it is paired with the wrong satellite state or time reference.
CHAPTER 4 OF 12 · Separate motion and clock drift
4. Solve a one-dimensional velocity and clock example
Assume only a fictional x-direction receiver velocity v is unknown; other velocity components are stipulated known. Choose two line-of-sight x coefficients +0.6 and −0.6, consistent with two otherwise specified unit directions. The reduced equations are y+ = −0.6v+b and y− = +0.6v+b.
Let y+ = −2.5 and y− = +3.5 metres per second. Subtracting gives y−−y+ = 1.2v = 6, so v = 5 metres per second. Averaging gives b = 0.5 metres per second. Reconstruct both observations: −3+0.5 = −2.5 and +3+0.5 = 3.5.
At the fictional wavelength 0.20 metres, these reduced range-rate quantities correspond to residual frequencies +12.5 and−17.5 hertz under the defined conversion. Their unequal magnitudes do not imply two inconsistent receiver speeds. The common clock term shifts both range-rate equations in the same direction.
The symmetric average isolates b only because the two position-direction coefficients are equal and opposite in this one-dimensional example. Do not average arbitrary real satellite Doppler measurements and call the result receiver clock drift. Their satellite contributions, geometry, signal definitions and other model terms must first be treated appropriately.
CHAPTER 5 OF 12 · Separate motion and clock drift
5. Distinguish clock offset from clock drift
A clock offset is a difference between time labels at an event. Clock drift is a rate of change of that offset under a stated time convention. In the reduced equation, b = c times a receiver clock-offset derivative in the simplified account. Its units are metres per second, not metres.
The fictional b = 0.5 metres per second corresponds to 0.5/299,792,458 seconds per second, approximately 1.66782 nanoseconds per second. This is a teaching conversion, not a typical receiver value. A constant clock offset and a constant clock drift have different effects on range and range-rate observations.
For an illustrative offset history δt(t) = a+kt, changing a leaves the derivative k unchanged. Changing k changes the range-rate clock contribution. A large constant offset can therefore coexist with a small drift, and a small offset at one instant can coexist with a nonzero drift. One snapshot does not identify both quantities.
The broadcast-clock polynomial guide concerns satellite-clock coefficients and their reference epochs. This guide’s b is a receiver term in a velocity observation model. Keep the subject, derivative and units explicit before transferring a clock parameter from one equation to another. The shared word clock does not make the parameters interchangeable.
CHAPTER 6 OF 12 · Test geometry and uncertainty
6. Use geometry to explain a non-unique solution
Suppose both reduced observations have the same coefficient uₓ = 0.6. Each then observes only −0.6v+b. If both residuals are −2.5, v = 5,b = 0.5 fits, but so does v = 0,b = −2.5. Repeating that row cannot separate velocity from clock drift in this simplified model.
The missing information is a contrasting sensitivity, not merely another sample number. A second direction with a different x coefficient can help, provided the common-clock and other model assumptions are appropriate. If the coefficients differ only slightly, the solution may be unique but sensitive to noise.
For two equations y1 = −u1v+b and y2 = −u2v+b, subtraction gives v = (y2−y1)/(u1−u2). A small denominator amplifies a residual difference and its uncertainty. This algebra explains poor geometry without relying on a vague claim that satellites must be spread out.
A real three-dimensional velocity model includes three velocity components and at least one clock-rate parameter in this simplified single-clock form. Its independent rows, weighting and additional needed parameters determine observability. The one-dimensional two-row example is not a universal real-world satellite-count rule.
CHAPTER 7 OF 12 · Test geometry and uncertainty
7. Propagate equal-noise observations through the solution
Return to coefficients ±0.6 and independent reduced-observation errors with equal standard uncertainty σ = 0.2 metres per second. Since v = (y−−y+)/1.2, its variance is (0.2²+0.2²)/1.2² = 0.0555556 square metres per squared second. Its standard uncertainty is approximately 0.235702 metres per second.
The clock-rate estimate b = (y−+y+)/2 has variance (0.2²+0.2²)/4 = 0.02, with standard uncertainty approximately 0.141421 metres per second. The velocity and clock-rate estimates are uncorrelated in this particular symmetric equal-noise construction, because the corresponding covariance contributions cancel.
If each input frequency has standard uncertainty σD at a known fixed wavelength, its converted range-rate standard uncertainty is λσD. For a fictional λ = 0.20 metres and σD = 0.5 hertz, that is 0.10 metres per second before the multi-observation geometry is applied.
These are conditional formal uncertainties. They do not include errors in the satellite model, residual atmospheric rates, timing mismatch or an unsuitable clock account. A small propagated frequency-noise value cannot certify the final velocity independently of the model that gave the observations their meaning.
CHAPTER 8 OF 12 · Test geometry and uncertainty
8. Keep unequal noise and shared errors visible
Now let σ+ = 0.2 and σ− = 0.4 metres per second, still independent. The velocity variance becomes (0.04+0.16)/1.44 = 0.1388889, giving standard uncertainty approximately 0.372678. The clock-rate variance is (0.04+0.16)/4 = 0.05, giving approximately 0.223607.
Their covariance is [Var(y−)−Var(y+)]/(4×0.6) = 0.12/2.4 = 0.05 in the corresponding squared-rate units. The estimates are now correlated despite independent input noise. Equal-noise symmetry was responsible for the earlier zero covariance; it was not a universal property of velocity and clock drift.
If both observations share an additive error a, their difference cancels a in this two-row construction, while their average retains it. That illustrates why the clock-like estimate can carry a common contribution not visible in the velocity difference. A different geometry or additional bias structure can change the cancellation.
Do not interpret that algebra as proof that every common atmospheric or hardware effect is harmless. The error must actually enter the compared observations with the assumed coefficient and timing. Signal-dependent or satellite-dependent contributions generally need their own account. Preserve covariance and definitions rather than inferring independence from separate channel numbers.
CHAPTER 9 OF 12 · Inspect timing and continuity
9. Distinguish an instantaneous rate from a time-window average
A reported frequency estimate may have a defined tracking or averaging interval. Its time label should identify what quantity that interval represents. A finite-window rate is not automatically the instantaneous velocity at the displayed endpoint. The interpretation must follow the actual product, not a guess based on the number of decimal places.
For an original idealised example, suppose a one-dimensional velocity follows v(t) = 2+0.4t metres per second over t = 0 to5 seconds. Its average over that interval is3 metres per second, while its endpoint value is4. Assigning the interval average to the endpoint as an instantaneous rate would introduce a one-metre-per-second mismatch in this model.
With constant acceleration, the interval average equals the instantaneous value at the midpoint, here t = 2.5. That equality is special to the linear velocity history. A more complicated acceleration history can require another relationship between the reported average, its timestamp and the desired instantaneous state.
This example does not claim every GNSS Doppler field is a simple five-second average. It teaches why the measurement definition and effective epoch matter. Aligning velocity with another sensor or a position difference requires compatible time accounts, especially during changing motion.
CHAPTER 10 OF 12 · Inspect timing and continuity
10. Separate carrier continuity from derivative information
A constant carrier ambiguity contributes a constant length offset within a supported arc. Its ideal time derivative is zero, so that constant does not enter a smooth range-rate calculation in the same way it enters an absolute carrier observation. This helps explain why rate information and absolute phase positioning have different ambiguity relationships.
A discontinuity is not a constant over the affected interval. Differencing carrier observations across a slip or gap can create a spurious rate. The existing carrier-continuity guide explains why arcs and independent evidence must be retained. A derivative calculation does not legitimise joining incompatible phase histories.
Tracked Doppler and a finite difference of stored carrier phase are also not identical products by definition. Their noise, filtering, timing and handling of interruptions can differ. Agreement can be a useful check when conventions are compatible; disagreement should be investigated rather than automatically assigned to receiver acceleration.
A changing antenna-orientation term or propagation contribution can also have a derivative. Our toy equations assumed the specified non-velocity contributions were handled or omitted explicitly. For a real solution, identify which derivatives matter at the intended precision and retain their model provenance. The word Doppler does not erase the rest of the observation equation.
CHAPTER 11 OF 12 · Transfer and report
11. Keep short-term velocity separate from long-term station motion
A Doppler-based velocity estimate concerns signal rates over its effective observation interval. A geodetic station velocity inferred from a long coordinate series describes another estimation problem, involving time span, seasonal terms, discontinuities and temporal covariance. The two outputs can use the same unit while answering different physical questions.
The existing station-velocity guide develops those long-term issues. A short-term receiver rate near zero does not establish the absence of slow crustal motion over years. Conversely, a fitted millimetres-per-year trend does not describe instantaneous motion during one receiver update or prove a specific short-lived acceleration.
Choose the target before choosing the comparison. A navigation task may need current receiver velocity, while a deformation study may need a frame-compatible secular rate. Matching units is necessary but insufficient; the time definition, reference frame and uncertainty model must also match.
A useful teaching exercise places a single range-rate estimate beside a multi-year slope and asks what additional information would make them comparable. The answer should discuss scale, averaging and reference rather than merely convert metres per second into metres per year. A unit conversion cannot supply missing physical assumptions.
CHAPTER 12 OF 12 · Transfer and report
12. Report the frequency convention and inferred state separately
A complete result identifies the Doppler definition, wavelength, effective epoch or interval, satellite-state model, estimated velocity components and receiver clock-rate convention. It distinguishes raw frequency, converted range rate, a reduced residual and a fitted receiver velocity. Each stage can be checked without pretending it already equals the next.
For the main fictional example, reduced observations −2.5 and+3.5 metres per second with coefficients ±0.6 imply v = 5 and b = 0.5. Their corresponding residual frequencies at the fictional wavelength are+12.5 and−17.5 hertz under our sign convention. The observations do not individually equal the receiver’s speed.
For real work, use current signal and receiver documentation and validated processing. Independent reference comparisons should be aligned in time and frame and should account for their own uncertainty. A low scatter in repeated estimates does not by itself establish accuracy during dynamics, interference or a model change.
The transferable lesson is to identify what a derivative measures and what common term it shares with other observations. Geometry can separate motion from clock drift when the model provides independent signatures. Without that separation, a frequency shift remains useful evidence—but not an unconditional statement about how fast the receiver moved.
Independent worked case and evidence checks
Use a fresh fictional one-dimensional model with line-of-sight x coefficients ±0.8. Let receiver velocity v = 2 metres per second and clock-rate term b = −0.2. The reduced observations are y+ = −1.6−0.2 = −1.8 and y− = +1.6−0.2 = +1.4. Recover v = [1.4−(−1.8)]/1.6 = 2 and b = (1.4−1.8)/2 = −0.2.
At fictional wavelength λ = 0.25 metres, the residual frequencies are D+ = +7.2 hertz and D− = −5.6. Reverse both conversions. Then change the wavelength to0.20 while keeping the same range-rate observations. The frequencies become+9 and−7 hertz, showing why a frequency value cannot be transferred between signals without its wavelength.
For independent equal observation standard uncertainties σ = 0.16 metres per second, the velocity standard uncertainty is √(2×0.16²)/1.6, approximately0.141421. The clock-rate standard uncertainty is0.16/√2, approximately0.113137. Explain why the geometric denominator affects velocity but not the symmetric average in this construction.
Now change both direction coefficients to+0.8 while retaining one observed combination. The equations no longer provide an independent velocity–clock separation. Find two parameter pairs with the same predicted y = −1.8: v = 2,b = −0.2 and v = 0,b = −1.8. More identical rows can strengthen that combination but cannot select one pair uniquely.
As an independent timing test, consider a reported ideal average over t = 10 to14 seconds with v(t) = 1+0.5(t−10). Its average is2 metres per second and its endpoint is3. The average equals the midpoint value at t = 12 under this linear model. State which event a comparison sensor should represent before treating its difference from the reported value as a measurement error.
Finally add a fictional shared +0.3-metre-per-second error to both original reduced observations. Their difference and inferred v remain unchanged, while the inferred b increases from−0.2 to+0.1. The cancellation holds because the same additive term entered both rows at the same effective event. A mismatched or signal-dependent contribution would require a fresh calculation.
A second independent geometry case removes the symmetric average shortcut. Use reduced equations y1 =−0.2v+b and y2 =−0.8v+b. With fictional v =5 and b =0.5, their values are−0.5 and−3.5. Subtracting gives y1−y2 =0.6v =3, recovering v =5. Substitution then returns b =0.5.
Averaging these two observations gives−2, not the receiver clock-rate term. Their geometric coefficients do not cancel in the average. This exposes why the earlier average worked: it used equal and opposite coefficients, not a universal property of Doppler data. A useful explanation names that symmetry rather than memorising an averaging recipe.
If each of these observations has independent standard uncertainty0.2 metres per second, the velocity standard uncertainty is√(0.04+0.04)/0.6, approximately0.471405 metres per second. It is twice the corresponding0.235702 value in the main ±0.6 example. The same input noise produced different velocity uncertainty because the coefficient contrast was halved.
Now suppose the modelled satellite contribution in the first row alone is too large by0.12 metres per second, so the reduced y1 is too small by0.12. The inferred velocity changes by−0.12/0.6 =−0.2 metres per second in this constructed geometry. It would be wrong to call the original0.12 a receiver-speed error directly. The geometry maps an observation-model discrepancy into the fitted parameter.
For a final sign test, reverse the declared Doppler definition to reference minus received frequency. The reported D values reverse, and the conversion to the same physical range-rate quantity must reverse its sign rule as well. Changing only the frequency signs while retaining the old conversion changes the inferred motion incorrectly. A complete convention transformation should preserve the reconstructed range rates and velocity under the same model.
A final unit check converts the main fictional clock-rate term into a change over a defined interval. If b =0.5 metres per second remains constant for four seconds in the simplified model, its clock-range contribution changes by two metres. Dividing that change by c gives the corresponding clock-offset change. The initial offset remains unknown unless separately supplied; integrating a derivative requires an initial condition.
This is why one should not turn a Doppler-derived drift estimate into an absolute receiver clock offset by multiplying by time since an arbitrary file start. The integration interval, starting offset and any changes in drift must be justified. The same distinction applies to using velocity to reconstruct position: a rate history needs an initial state and a compatible time account before it becomes a trajectory.
Record the starting state separately from the rate observations. A later reader can then change the initial condition without altering the measured rates, and can see which part of a reconstructed trajectory came from observation and which part came from an external reference.
Worked comparison
| Quantity | Meaning in the declared model | Common overstatement |
|---|---|---|
| Doppler frequency D | A signed signal-frequency observable | Receiver speed directly |
| −λD | Converted range-rate-like contribution | A complete velocity vector |
| Reduced residual y | Observation after specified model contributions | Pure receiver motion without a clock term |
| Receiver clock drift b | A common range-rate parameter | A constant clock offset |
| Velocity estimate | A state inferred from geometry and observations | A long-term geodetic station trend |
| Formal covariance | Uncertainty under the stated error model | Independent accuracy certification |
Questions and practical checks
Can a stationary receiver observe Doppler shifts?
Yes. Satellite motion and clock effects contribute. Raw Doppler does not directly identify receiver motion.
Is positive Doppler always positive range rate?
Not under the convention defined here. Our conversion is q = −λD. Actual files must follow their own documented sign definitions.
Why solve receiver clock drift with velocity?
A common frequency-error contribution can enter the same observations. Contrasting geometry helps distinguish it from motion.
Does differentiating remove every carrier-phase problem?
No. A constant ambiguity has zero derivative within a supported arc, but slips, gaps and changing model terms can affect rate information.
Can I compare a window average with an endpoint velocity directly?
Only with an appropriate time and motion account. The worked acceleration examples show why the effective epoch matters.
Continue with the connected subjects
Establish signal and clock definitions
Preserve Doppler-related signal identities, units and epochs in RINEX records
Distinguish clock offsets, derivatives and reference epochs
Check satellite state and derivative evaluation between orbit samples
Separate continuity and time-scale questions
Keep interrupted carrier arcs separate from movement claims
Use the long-term station-velocity guide for secular motion and seasonal uncertainty
Sources and scope
ESA Navipedia: Doppler, relative antenna motion and receiver clock-frequency error
ESA Navipedia: geometry and conditional formal uncertainty
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 31 October 2026.
Related observation models
Use these companion guides to compare a different observation relationship; each keeps its own units, uncertainty assumptions and limits.
Compare orbit-vector projections and orbit–clock covariance
Compare inter-system offsets and identifiable clock combinations