A dual-frequency ionosphere-free combination removes the first-order ionospheric delay by weighting measurements from the same satellite at two different frequencies. It preserves their common geometric and timing terms, but it also weights their noise and other errors. “Ionosphere-free” therefore does not mean error-free, ambiguity-fixed or immune to loss of tracking. It names a particular measurement combination, not the complete performance of a multi-frequency receiver.
For a GNSS integration engineer comparing observations or positioning modes, the important distinction is between receiving another frequency and deciding how to use its measurements. The combination below explains one widely used approach. It is not a claim that every receiver uses this approach internally.
Why the two frequencies carry different delays
The ionosphere is dispersive: its effect depends on signal frequency. To first order, code pseudorange has a positive ionospheric range delay proportional to 1/f². Carrier phase, expressed in metres under the usual observation model, has the opposite sign. A second frequency provides a differently scaled observation of that propagation effect. ESA-hosted Navipedia derives this frequency dependence.
This is a propagation difference, not simply a choice between two independent noisy copies of the same range. At the same slant electron content, the lower-frequency signal has the larger first-order code delay. That does not make a lower-frequency signal categorically worse: its signal design, tracking conditions and use in the measurement model are separate considerations.
The pseudorange and carrier-phase guide explains the underlying observables. Here the task is narrower: construct a combination that cancels their specified frequency-dependent term without canceling the range we want to measure.
Choose weights that cancel ionosphere, not geometry
Consider two code measurements P₁ and P₂ in metres, for the same satellite and a common observation epoch. To expose the first-order mechanism, write:
P₁ = R + K/f₁² + e₁; P₂ = R + K/f₂² + e₂.
R contains the common geometric range, clock contribution and non-dispersive tropospheric delay. K represents the common slant-ionosphere factor. e₁ and e₂ collect measurement errors not represented by those terms. Frequency-dependent hardware and antenna effects are omitted from this simplified derivation, not assumed absent in an actual receiver.
The conventional code ionosphere-free combination is:
P_IF = aP₁ + bP₂, where a = f₁²/(f₁² − f₂²) and b = −f₂²/(f₁² − f₂²).
Use distinct frequencies and the same frequency units in both coefficients. The two identities explain the result: a + b = 1, so R remains; a/f₁² + b/f₂² = 0, so the first-order ionospheric term cancels. What remains in this model is P_IF = R + ae₁ + be₂.
This is the combination given in Navipedia's dual-frequency ionosphere-free reference. It is a subtraction with one negative coefficient, not an ordinary average. Separate constellations transmitting at the same centre frequency do not supply the distinct-frequency pair required by this formula.
A hypothetical GPS L1/L2 cancellation
Use GPS L1 at 1575.42 MHz and L2 at 1227.60 MHz, as identified in the GPS signal plan. The resulting coefficients are approximately a = 2.545728 and b = −1.545728.
For an explicitly hypothetical, error-free calculation, let R be 20,000,000 m and let the L1 code delay be 5.000 m. Under the same first-order slant-ionosphere model, the L2 delay is 5 × (1575.42/1227.60)², or approximately 8.234722 m.
Thus P₁ is 20,000,005.000 m and P₂ is approximately 20,000,008.234722 m. Applying the unrounded coefficients removes the assumed ionospheric delay and returns R. It does not remove R's clock or tropospheric components. These numbers demonstrate the algebra, not an observed satellite range or a receiver accuracy result.
For the atmospheric term that remains, see how tropospheric and ionospheric delays differ, including slant mapping and why a residual delay can affect the estimated height.
Cancellation can increase the random noise
For the linear combination above, let σ₁ and σ₂ describe the standard deviations of the two code-noise terms. Its noise variance is:
σ_IF² = a²σ₁² + b²σ₂² + 2ab Cov(e₁,e₂).
The covariance has units of square metres. If the two terms are uncorrelated, that last term is zero. This follows the NIST law of propagation of uncertainty for weighted sums; the negative sign of b does not make its squared noise contribution negative.
Continue the hypothetical L1/L2 example, now assuming zero-mean, uncorrelated code noise with σ₁ = σ₂ = 0.30 m. The combined standard deviation is 0.30 × √(a² + b²), approximately 0.893 m. The noise multiplier is about 2.978, not one. No Gaussian assumption is needed to calculate this variance; a probability or percentile claim would need additional distribution assumptions.
There is no contradiction: the calculation removes an assumed systematic ionospheric term while increasing this random component. A noisier combined observable can still be useful when it removes a significant error. Nor does this example prove that dual-frequency positioning is less accurate. A position solution includes multiple satellites, weighting, biases, corrections and possibly carrier observations. Correlated or unequal input errors require their actual covariance model rather than the example's shortcut.
What the combination leaves in the measurement
The complete GNSS observation model contains more than geometric range and first-order ionosphere. Keep the following interpretation separate from the cancellation calculation.
| Measurement feature | Effect of this combination | Consequence when comparing results |
|---|---|---|
| Common geometry, clock contribution and non-dispersive tropospheric delay | The coefficients sum to one, preserving these common terms | An ionosphere-free range is still not pure geometric distance; the position model must handle the remaining terms. |
| Noise and multipath on the selected signals | They enter with the combination's weights rather than generally canceling | Compare the actual signals, error levels and processing stage; the frequency count alone does not predict the residual error. |
| Higher-order ionospheric effects | They do not follow the single first-order 1/f² term eliminated here | Interpret “ionosphere-free” as first-order removal, not a guarantee under every atmospheric condition. |
| Carrier ambiguity and hardware-related terms | A carrier combination retains weighted ambiguity and bias contributions; code-bias treatment depends on the clock and signal convention | Combining two carrier phases does not by itself establish an integer-fixed position. Use the documented measurement and correction model. |
Navipedia's signal-pair and clock-definition analysis explains why instrumental terms must be interpreted with the adopted clock reference. It would be equally misleading to say that all biases survive unchanged or that every hardware delay automatically cancels. Its PPP modelling reference specifically notes the correction implications of using signal types different from the reference combination.
More frequencies provide choices, not one universal output
The pair determines the coefficients. For f₁ greater than f₂, bringing the frequencies closer together reduces the denominator f₁² − f₂² and increases the magnitudes of the weights. Under the same uncorrelated, equal-noise assumption, that increases noise amplification. This algebraic observation is not a recommendation to select a pair solely by separation: available signals can have different noise, multipath and tracking characteristics.
A receiver with three frequencies has additional possible pairs and combinations. Which observations it actually uses, how it weights them and what happens when a required signal is rejected or lost remain implementation questions. A three-frequency label does not identify a single ionosphere-free observable. Losing one component also makes that particular two-frequency combination unavailable at that epoch; fallback behavior is not established by the formula.
Do not confuse this operation with carrier smoothing. The carrier-smoothed pseudorange guide explains why divergence-free smoothing can remove a smoothing-related ionospheric mismatch while retaining code delay. A smoother trace is not proof that the first-order ionosphere was removed, just as an ionosphere-free combination is not proof that the carrier ambiguities were fixed.
Separate delay correction from continued reception
The calculation requires usable measurements. It cannot reconstruct an absent or corrupted signal merely because another frequency is listed in the receiver specification. NOAA's space-weather explanation distinguishes frequency-dependent positioning errors from disturbed reception and tracking. Correcting a modeled delay is not the same task as maintaining lock through rapid signal fluctuations.
Likewise, subtracting an ionospheric term does not establish RF-interference suppression. If a project also needs GNSS protection, the ResiNav catalogue provides the separate published product records. This measurement explanation adds no ResiNav raw-observation, ionospheric-processing, frequency-pair or positioning-performance claim.
When comparing a receiver mode or an evaluation log, identify the actual signal pair, common epoch, observation units, combination coefficients, bias convention and reported processing stage. Compare position results separately from raw or combined ranges. For the equipment-category decision, the multi-constellation versus multi-frequency guide remains the starting point. The additional understanding here is why a second frequency can remove one major propagation term while still leaving a measurement that needs a complete error model.
NEXT STEP
Turn available platform information into an engineering review.
Use the Technology centre to frame the discussion, review relevant application scenarios, then send the available platform and receiver details for confirmation.