The Ionosphere & Multi-Frequency
Turning the largest error into a measurement you can cancel
Why it matters
The ionosphere is the dominant error in single-frequency GNSS and the reason precise positioning needs two frequencies. Understanding its 1/f² dispersion explains the iono-free combination that PPP and precise RTK depend on, the broadcast models single-frequency receivers use, and why space weather degrades positioning. It is the clearest example of exploiting the physics of an error to remove it.
Prerequisites
The line item that was hiding physics
When we drew up the error budget, the ionosphere was the tallest bar — a few metres at the zenith, growing toward the horizon — and we handled it with a dropdown: "model it," or "use the iono-free combination." Later, PPP leaned on that iono-free combination and casually multiplied its measurement noise by about , a number that appeared from nowhere. Both were promissory notes. This lesson pays them off, and the payment turns out to be one of the most elegant ideas in all of GNSS: the largest error is also the one the physics lets you measure and delete.
The idea, before any mathematics
Up high — roughly 60 to 1000 km — sunlight strips electrons off atoms, leaving a shell of charged plasma: the ionosphere. A GNSS signal has to cross it, and a cloud of free electrons slows the signal's code down. Two things control how much. First, how many electrons lie along the particular ray from that satellite to you — the Total Electron Content, or TEC. A satellite low on the horizon is seen through a long, grazing slice of ionosphere and suffers far more delay than one straight overhead; same sky, longer path. Second — and this is the magic — the frequency of the signal. The ionosphere is dispersive: it bends and delays low frequencies more than high ones. The delay falls off as one over the frequency squared.
Sit with what that means. If the delay were the same at every frequency, you would be stuck: you could never separate it from the true range, because it would look exactly like extra distance. But it is not the same at every frequency. Send two signals down at once — GPS does, on L1 and L2 — and they arrive delayed by different, precisely-related amounts. The gap between them is not a nuisance; it is a ruler. Measure that gap and you have measured the ionosphere itself, directly, with no model and no guesswork.
And once you can measure it, you can remove it. There is a particular weighted difference of the two pseudoranges — the iono-free combination — engineered so the delay terms cancel exactly, leaving range, clock, and noise behind. First-order ionosphere, which is better than 99.9% of the whole effect, simply vanishes.
The ionosphere is the biggest error in GNSS, and also the most honest one: it signs its work. Because it touches each frequency by a different, known amount, a receiver listening on two frequencies can read the signature, measure the error, and subtract it — something no amount of listening on one frequency can ever do.
Nothing is free, of course. The same weighted difference that cancels the ionosphere also adds up the noise of the two measurements, inflating it by a fixed geometric factor — about for GPS L1/L2. There is the from PPP, no longer a mystery but a direct consequence of the combination. A single-frequency receiver, unable to form the combination at all, is left to model the delay from coefficients in the navigation message, and a good model removes only about half.
Seeing it
Picture a curve of delay against frequency: a steep slide, high on the left where the frequencies are low, dropping as you move right. Drop three dots on it for L5, L2, and L1. They sit at different heights — that vertical spread between the dots is the measurable signature of the electron content. Turn up the TEC and the whole curve lifts, the dots spreading further apart; drop the satellite toward the horizon and it lifts again. The single-frequency user lives at one dot and must guess the rest; the dual-frequency user reads the spread and erases it.
Try it yourself
The left panel is the dispersion curve; the right compares what each correction strategy leaves behind.
- Slant TEC
- 57 TECU
- L1 delay
- 9.19 m
- L2 delay
- 15.13 m
- Iono-free noise ×
- 2.98
The ionosphere delays the GPS code by 40.3 · TEC / f² metres — metres at the zenith, far more at low elevation, where the ray cuts a longer slant through the plasma. Raise the TEC and the whole curve lifts. The decisive feature is its shape: because the delay goes as 1/f², the three frequencies are hit by different amounts, and that gap is not noise — it is a direct readout of the electron content. A single-frequency receiver can only guess, modelling away perhaps half with broadcast coefficients. A dual-frequency receiver forms the iono-freecombination that cancels the 1/f² term almost perfectly, collapsing the error bar to nearly zero — paying only a fixed noise amplification (about 2.98× for L1/L2, less for the wider L1/L5). That is the trade PPP quietly made, and why every precise receiver tracks more than one frequency.
Raise the vertical TEC and watch the whole dispersion curve climb and the L1 delay with it. Now drop the elevation toward the horizon: the slant path lengthens and the delay balloons — this is why low satellites are the worst offenders. Look at the three bars. No correction carries the full delay. The broadcast model scrapes off whatever fraction you dial in — rarely more than half. The iono-free bar collapses to nearly zero bias, but grows a noise whisker: toggle L1/L2 versus L1/L5 and watch that whisker shrink for the wider-spread pair, because frequencies farther apart make a stiffer, less noise-amplifying combination. That single trade — perfect removal for amplified noise — is the entire reason precise receivers are multi-frequency.
Now the mathematics
The first-order ionospheric group delay of the code, in metres, is
with the slant total electron content (electrons per m²) and the carrier frequency (Hz). The constant comes from the plasma dispersion relation. Slant content relates to the vertical content through an obliquity (mapping) factor that grows toward the horizon; a simple thin-shell form is .
Two frequencies give two equations sharing the same geometry and clock but different ionosphere:
where is everything non-dispersive (range, clock, troposphere). Subtracting them — the geometry-free combination — leaves only the ionosphere, so TEC is directly observable:
Weighting instead to cancel the dispersive term gives the iono-free combination:
free of first-order ionosphere. Its noise, for independent per-measurement , is amplified to
The amplification shrinks as the two frequencies spread apart — one reason the modern L1/L5 pair is attractive.
From the idea to working code
@gnerd/gnss-core implements the dispersion directly, so each claim above is executable:
import {
ionosphericDelay_m,
ionoFreeCombination,
ionoFreeNoiseAmplification,
estimateSlantTec_TECU,
} from "@gnerd/gnss-core";
import { GPS_L1_FREQ_HZ, GPS_L2_FREQ_HZ } from "@gnerd/gnss-core";
const rho0 = 22_000_000 + 1234.5; // range + clock (non-dispersive), metres
const stec = 40; // slant TEC, TECU
const r1 = rho0 + ionosphericDelay_m(stec, GPS_L1_FREQ_HZ);
const r2 = rho0 + ionosphericDelay_m(stec, GPS_L2_FREQ_HZ); // delayed more (lower frequency)
ionoFreeCombination(r1, GPS_L1_FREQ_HZ, r2, GPS_L2_FREQ_HZ); // -> rho0, ionosphere gone
estimateSlantTec_TECU(r1, GPS_L1_FREQ_HZ, r2, GPS_L2_FREQ_HZ); // -> 40, measured back out
ionoFreeNoiseAmplification(GPS_L1_FREQ_HZ, GPS_L2_FREQ_HZ); // -> ~2.98
The measurement model keeps the dispersive and non-dispersive parts explicit, the combinations are small pure functions, and the noise amplification is a derived quantity rather than a magic constant — the same number PPP spent, now accounted for.
Check your understanding
- Why can two frequencies remove the ionosphere when one cannot? (The delay is dispersive — — so two frequencies see different, known-ratio delays whose combination cancels the term; a single frequency cannot separate the delay from true range.)
- You switch a receiver from the L1/L2 iono-free combination to L1/L5 and the position noise drops. Why? (L1 and L5 are farther apart in frequency, so the iono-free combination has a smaller noise amplification factor.)
- A single-frequency receiver applies the broadcast ionosphere model and still has a metre-level error at low elevation. Is it broken? (No — broadcast models remove only about half the delay, and the slant path at low elevation makes the delay large to begin with.)
Where this goes next
Reading the ionosphere instead of fighting it opens a surprising door: the geometry-free combination that measures TEC is the basis of ionospheric mapping — GNSS receivers worldwide now sense space weather in real time. For positioning, the iono-free combination is the backbone of PPP and precise RTK, while single-frequency mass-market receivers increasingly borrow regional correction services (SBAS) to do better than a broadcast model. The one piece we still left as a line item is the troposphere — which, unlike the ionosphere, is not dispersive, so this whole trick fails on it and a completely different strategy is needed. That contrast is the natural next step.
Common misconceptions
- The ionosphere delays all frequencies equally. It is dispersive: the delay scales as 1/f², so lower frequencies are delayed more — which is exactly what makes it measurable with two frequencies.
- A broadcast model removes the ionosphere. A single-frequency model (e.g. Klobuchar) typically removes only about half of it; full first-order removal requires two frequencies.
- The iono-free combination is free. It cancels the ionosphere but amplifies measurement noise by a fixed factor (~2.98 for GPS L1/L2) and removes the ability to read absolute TEC.
Key takeaways
- Ionospheric delay is 40.3·TEC/f² metres - large, and largest at low elevation.
- The delay is dispersive (1/f²), so two frequencies measure it directly from their difference.
- The iono-free combination cancels the first-order delay almost completely.
- Its price is a fixed noise amplification (~2.98 for L1/L2) - the 2.9 PPP used.
- Single-frequency receivers can only model the delay and keep a large residual.