The Gold Standard of Radar | Doppler Velocity Folding and Why a Sign Can Reverse

We model one point target and a fixed radar whose transmitter and receiver occupy the same position. During the observation, radial velocity v is constant, the echo remains associated with that target and the wavelength is λ. We adopt positive v for increasing range.

For clarity, choose a processed phase convention in which increasing range produces a positive phase advance. An actual receiver or weather product can use another sign convention; check its definition rather than carrying ours into its display.

The target’s radial velocity is the rate of change of its range. It is not necessarily its complete speed. A target crossing the line of sight can have substantial sideways speed with a small radial component.

The radar subject root develops that projection. Here we keep the geometry fixed so that a different question becomes visible: can sampling uniquely identify even the radial component?

Contents · Next chapter

If one-way range changes by ΔR, the round-trip path changes by 2ΔR. One wavelength of path change corresponds to one complete phase turn, so the model phase advance is Δφ = 4πΔR/λ.

Between pulses separated by T seconds, ΔR = vT. Therefore Δφ = 4πvT/λ. Dividing by 2π gives the number of phase turns between samples: 2vT/λ.

Use λ = 0.10 metre and T = 0.001 second, corresponding to a pulse repetition frequency of 1,000 per second. At v = 10 metres per second, range increases by 0.010 metre between pulses. The round-trip path changes by 0.020 metre, one fifth of a wavelength.

The corresponding phase advance is 0.2 turn, or 72 degrees. The factor of two belongs to the round-trip geometry. Removing it would change the inferred velocity even before sampling ambiguity is considered.

Contents · Previous chapter · Next chapter

A complex phase sample identifies a position around a circle. After a full extra turn, that position is the same. A phase advance of 252 degrees therefore ends at the same position as an advance of −108 degrees.

This does not mean time ran backwards or the target physically reversed. The two rotations differ by 360 degrees, which sampled endpoint phases do not distinguish.

We represent the measured phase difference in the interval from −180 degrees up to, but excluding, +180 degrees. This half-open convention gives one representative for every circular position. The boundary values −180 and +180 describe the same endpoint.

Because pulse times are uniformly spaced, adding one full phase turn per interval leaves every sample unchanged, not just one pair. More samples can improve an estimate within an alias class while preserving the ambiguity between its members. To distinguish those members, some additional information or changed measurement structure is needed.

Contents · Previous chapter · Next chapter

Keep λ = 0.10 metre and T = 0.001 second. Let the true radial velocity be +35 metres per second. The range increase per pulse interval is 0.035 metre and the round-trip change is 0.070 metre.

That is 0.7 wavelength, giving a phase advance of 252 degrees. Expressed in our chosen interval, it becomes −108 degrees, or −0.3 turn.

Converting that principal phase value directly back to velocity gives v = λ×(−0.3)/(2T) = −15 metres per second. The displayed representative has the opposite sign to the true velocity.

The sample sequence is compatible with both velocities. The calculation has selected a representative; it has not established which physical member generated the observations. That distinction is the central limit of this example.

Contents · Previous chapter · Next chapter

The principal phase interval has width 2π. Since Δφ = 4πvT/λ, the represented velocity interval has width λ/(2T). Its half-width is vN = λ/(4T) = λ×PRF/4, the Nyquist velocity magnitude for this simple uniform-pulse model.

Our example has vN = 25 metres per second. The chosen interval is −25 ≤ v < 25. Velocities differ by the alias spacing 2vN = 50 metres per second.

Thus an apparent −15 is compatible with v = −15 + 50k for any integer k allowed by the physical context. Candidates include −65, −15, +35 and +85 metres per second.

NOAA’s velocity-aliasing glossary identifies folding as ambiguity outside the Nyquist interval. The wavelength–pulse-rate expression above follows directly from our stated phase model. The precise interval endpoint convention here is a mathematical choice for the worked example; operational products must be read using their own definitions.

Contents · Previous chapter · Next chapter

Suppose an estimator reports −15.0 metres per second with a small within-interval uncertainty. That small uncertainty does not eliminate +35.0 as an alias under the model. Both can fit the same uniformly sampled phases exactly.

Longer coherent observation can help separate nearby Doppler frequencies in an appropriate processing model. It does not by itself remove integer-cycle ambiguity when the sampling structure remains unchanged.

Range ambiguity asks which transmitted pulse produced a return. Velocity ambiguity asks which phase progression fits the sampled echo. The two-pulse-interval range article develops the first question. Its timing-candidate intersections do not automatically solve the second.

Similarly, weak echoes beside strong targets concerns interference and distinguishability. A bright, clean, precisely estimated echo can still carry velocity ambiguity. Identify which limitation is being tested before using the broad word “accuracy”.

Contents · Previous chapter · Next chapter

Observe the same constant-velocity target using a second uniform pulse rate of 1,200 per second, with the wavelength unchanged. Now T = 1/1,200 second, vN = 30 metres per second and alias spacing is 60 metres per second.

For true v = +35, the phase advance is 2vT/λ = 7/12 turn, or 210 degrees. Its principal representation is −150 degrees, giving apparent velocity −25 metres per second.

We now have two candidate families: −15 + 50k from the first observation and −25 + 60j from the second. Inside an independently justified search region −100 ≤ v < 100, their lists are short enough to inspect.

Only +35 is shared in this region. This is a conditional reconstruction, based on a common target, constant radial velocity, known wavelength and a justified search bound.

Contents · Previous chapter · Next chapter

The pair does not identify a unique velocity over all possible numbers. Adding 300 metres per second preserves both residues because 300 is a multiple of both 50 and 60. The shared candidates repeat at +35 + 300m for integer m.

For example, +335 produces apparent −15 in the first observation and −25 in the second. Our search bound excludes it; the phase equations alone do not.

Measurement uncertainty also turns exact candidates into intervals. If both apparent velocities have a stated ±0.5-metre-per-second interval under an appropriate error account, compare the shifted bands rather than rounding everything to exact integers. A unique band intersection is not exact truth.

Actual multi-rate processing faces target association, acceleration, phase noise, missing echoes and clutter. The elementary intersection exercise establishes the arithmetic of a simplified case. It does not validate a de-aliasing algorithm for a changing weather field or a surveillance system.

Contents · Previous chapter · Next chapter

At λ = 0.10 metre, raising a uniform PRF from 1,000 to 2,000 per second raises vN from 25 to 50 metres per second. The +35 target now lies inside the chosen interval and need not fold in this model.

But the simple unambiguous range interval c/(2PRF) falls from 150 to 75 kilometres, using c = 300,000,000 metres per second. Increasing one limit has tightened another.

This comparison assumes the ordinary uniform pulsed monostatic model. Real systems can use more elaborate waveforms and processing; these equations should not be presented as a universal description of every radar.

The radar range, angles and Doppler mathematics article supplies the wider calculation context. Our focused contribution is to show the sign reversal, enumerate the velocity aliases and test the scope of a two-rate reconstruction. A useful design comparison needs its range, motion and detection tasks stated together.

Contents · Previous chapter · Next chapter

Use a spreadsheet with pulse index n and phase in turns. For the +35 case at the first rate, enter 0.7n and retain its fractional part. For the −15 case, enter −0.3n and reduce it modulo one. The two columns coincide for every integer n.

Test several synthetic velocities, including zero, a value near each boundary and a value beyond one alias spacing. Keep the true input hidden from the person interpreting the phases, then compare their reported candidates with the known construction.

Record unresolved cases honestly. An output that lists several possible velocities is more faithful to the evidence than a forced principal-interval answer described as certainly physical.

No transmitter is needed for this activity. Actual verification would require appropriate system data and independent references. A successful paper model checks the reasoning and arithmetic, while leaving noise, propagation, calibration and field performance as separate evidence questions.

Contents · Previous chapter · Next chapter

Check one: change the sign

At λ = 0.10 metre and PRF = 1,000 per second, true v = −40 metres per second has principal representative +10. Add the 50-metre-per-second alias spacing once. The apparent reversal is produced by wrapping, not a change in actual motion.

Check two: change the wavelength

At λ = 0.05 metre with the same PRF, vN is 12.5 metres per second and alias spacing is 25. True +35 becomes +10 in −12.5 ≤ v < 12.5. The alias class changed because the wavelength changed.

Check three: examine the endpoint

At v = +25 for the original settings, the phase step is +180 degrees, identical at the sampled endpoints to −180 degrees. Our half-open convention represents it as −25. Do not claim both endpoints are uniquely distinguishable.

Explain each answer using phase turns before checking the arithmetic. Then ask which additional observation could distinguish the candidates, rather than merely returning more decimal places for the same representative.

Contents · Previous chapter · Next chapter

Can a negative velocity still come from a receding target?

Yes, under our stated sign convention, an aliased representative can be negative even when the true radial velocity is positive. Check the product convention and processing before interpreting a real display.

Will averaging more pulses always fix aliasing?

No. The alternative velocities can generate the same ideal uniform sample sequence. Additional constraints or measurement structure are needed to distinguish them.

Does de-aliasing recover complete wind velocity?

Resolving a radial component does not supply the missing transverse components. Geometry and further evidence remain necessary.

Which guide answers the next question?

For foundations, return to The Gold Standard of Radar and radar mathematics. For related ambiguities, follow range candidate intersections and bistatic paths and excess delay. The latter changes transmitter–receiver geometry, so the monostatic Doppler equation here must not be carried over unchanged.

For evidence across observations, use out-of-order observations and track age, weak echoes beside strong targets and sensor fusion engineering. For a different weather-radar quantity, read reflectivity and rainfall estimation.

Continue through The Gold Standard of Education and How X Works to practise distinguishing a measurement from the inference built on it.

NOAA’s velocity-aliasing glossary supports the distinction between physical radial velocity and a folded display value. Chapter 12 of the US Navy Aerographer’s Mate 1 & C training course, hosted by the National Weather Service, explains velocity aliasing and its relationship with range folding. That 1995 text is used for the enduring sampling principle, not as a statement of current operational configurations. All numerical sequences and candidate tables in this article are original simplified examples; no operational performance is claimed.

Contents · Previous chapter · Return to Radar