How GPS Actually Knows Where You Are

Your phone is a dumb radio. It doesn’t contain a map of the world, a ruler, or a particularly accurate clock. And yet, right now, it knows where you are to within five meters. How?

The answer has nothing to do with measuring angles, nothing to do with cell towers, and — here’s the part that surprises people — requires correcting for Einstein’s theory of relativity just to work at all. The real mechanism is a chain of physics: radio signals that carry timestamps, timestamps that become distances, distances that become spheres, and four intersecting spheres that pin a point in three-dimensional space while simultaneously correcting for your receiver’s sloppy internal clock.

GPS is an engineering system, but it was built on pure physics. Understanding how it works is understanding how time becomes location.

20,200 km20,200 km20,200 kmORBITAL ALTITUDEGPS satellite distance from Earth

38.5 µs/day38.5 µs/day38.5 µs/dayRELATIVISTIC DRIFThow fast clocks run ahead without correction

11 km/day11 km/day11 km/dayERROR WITHOUT RELATIVITYposition error if uncorrected

4.9 m4.9 m4.9 mCIVILIAN ACCURACYtypical open-sky smartphone GPS

---

The wrong mental model: triangulation

Before teaching the right model, let’s name the wrong one — because almost everyone starts there.

You’ve probably heard that GPS works by “triangulation.” Triangulation is a surveying technique: you know two points and you measure the angles to a third unknown point, then use trigonometry to compute where the third point must be. It’s how ancient mapmakers measured coastlines and how artillery spotters located enemy guns.

GPS does none of this. Satellites don’t aim at you. They don’t measure angles. A GPS satellite doesn’t know you exist. It just broadcasts a signal continuously, like a lighthouse that repeats “I am Satellite 12, and the time right now is 10:32:47.000000000” — a billion times a second.

Your receiver measures how long that signal took to arrive. That time delay becomes a distance — not an angle, never an angle. The correct word for what GPS does is trilateration, which is position-finding by distance, not bearing.

Here’s what that difference feels like in practice. Run this:

runnable · js▸ run

output

Run the code to see its output here.

Both methods can locate a point, but satellites can only give you distances. There is no angle being measured anywhere in the GPS system.

So how does a time measurement become a distance? This is where the physics starts.

---

Distance from time: how a radio signal carries a ruler

Light — and radio waves, which are light at lower frequencies — travels atthe speed of lightDefined since 1983 as exactly 299,792,458 metres per second (c). Source: NIST/BIPM SI definition. exactly 299,792,458 metres per second in a vacuum. That’s about 30 centimetres per nanosecond — a billionth of a second.

If you know precisely when a GPS satellite transmitted a signal and precisely when your receiver received it, the difference is the travel time. Multiply by the speed of light and you have the distance. Simple in principle; brutal in execution.

$$\text{distance} = c \times \Delta t$$

$$c = 299{,}792{,}458 \text{ m/s}$$whyspeed of light in vacuum

$$\Delta t = t_{\text{receive}} - t_{\text{transmit}}$$whytravel time of the signal

$$d = c \times \Delta t$$whygeometric distance to satellite

The constraint on timing accuracy is merciless. GPS aims for meter-level positioning. One meter of position error corresponds to 1/c = about 3.3 nanoseconds of timing error. To hit 4.9 meters accuracy, your timing system must be correct to within about 16 nanoseconds — sixteen billionths of a second.

Your phone’s clock is not nearly that accurate. Neither is the quartz crystal in your laptop. So where does the nanosecond precision come from?

It comes from the satellites.

What each satellite broadcasts

Every GPS satellite carries acesium or rubidium atomic clockGPS Block II/IIA satellites carried both rubidium and cesium clocks; Block IIR carries rubidium-only (PerkinElmer RAFS). Stability is approximately 1 part in 10¹³ or better over 12 hours. Source: GPS.gov, gps.gov/performance-standards-specifications. accurate to one part in 10¹³ — that’s one nanosecond drift over roughly ten thousand seconds. Each satellite continuously broadcasts:

1. The exact time of transmission (from its atomic clock)
2. Its own precise position in orbit (the ephemeris — orbital parameters uploaded from ground control)
3. A health and status message

Your receiver picks up these signals, records the arrival time on its own (much cheaper) clock, and computes a distance. But this distance is not quite right — and the reason it’s wrong is the central engineering challenge of the whole system.

---

The pseudorange: distances that lie

Here is the core problem. Your receiver has a clock, but that clock is not synchronized to GPS time. It might be running fast, slow, or just offset by an unknown amount. Let’s call that offset clock biasthe difference between the receiver's clock reading and true GPS time. It is unknown and must be solved for as a fourth unknown alongside x, y, z. and label it δt.

When you measure a signal arrival time on a biased clock and compute a distance, you get the true geometric range plus an error term proportional to the bias. This impure, bias-infected distance is called apseudorange'Pseudo' because it is not a true range — it is corrupted by the receiver clock bias. Source: Wikipedia, Pseudorange: https://en.wikipedia.org/wiki/Pseudorange:

(1)

$$\rho_i = d_i + c \cdot \delta t$$

$$\rho_i = c \cdot (t_{\text{receive}} - t_{\text{transmit},i})$$whywhat the receiver computes from its biased clock

$$d_i = \sqrt{(x_s^i - x)^2 + (y_s^i - y)^2 + (z_s^i - z)^2}$$whytrue geometric range to satellite i

$$\rho_i = d_i + c \cdot \delta t$$whypseudorange = true range + clock bias term

The satellite’s position (x_s, y_s, z_s) is known — it’s in the broadcast ephemeris. The unknowns are your receiver’s position (x, y, z) plus the clock bias δt. That’s four unknowns.

With one satellite you have one equation and four unknowns — no solution. With two satellites, two equations, still under-determined. With three satellites, three equations — you can almost solve it, but you’re still one short. With four satellites, you finally have a determined system:

(2)

$$\begin{cases} \rho_1 = \sqrt{(x_s^1-x)^2+(y_s^1-y)^2+(z_s^1-z)^2} + c\,\delta t \\ \rho_2 = \sqrt{(x_s^2-x)^2+(y_s^2-y)^2+(z_s^2-z)^2} + c\,\delta t \\ \rho_3 = \sqrt{(x_s^3-x)^2+(y_s^3-y)^2+(z_s^3-z)^2} + c\,\delta t \\ \rho_4 = \sqrt{(x_s^4-x)^2+(y_s^4-y)^2+(z_s^4-z)^2} + c\,\delta t \end{cases}$$

Here’s the geometric intuition. Each satellite defines a sphere of possible receiver locations — you are somewhere on the surface of a sphere centered on that satellite, with radius equal to the pseudorange. If your clock bias were zero, three spheres would intersect at exactly one point on the Earth’s surface (the other mathematical intersection is up in space and can be discarded). But because δt ≠ 0, the three spheres don’t intersect cleanly — they meet in a tiny volume, not a point. The fourth satellite’s sphere collapses that volume to a single point only when the bias is correct. The system finds the bias that makes all four spheres intersect — and that bias value is the receiver’s clock correction.

Go deeper: why the system is nonlinear and how receivers iterate

The pseudorange equations above are nonlinear because the true range terms involve square roots of position differences. To solve them numerically, receivers use the Gauss-Newton method: start with an approximate position estimate (often the last known position, or the center of the Earth on first fix), linearize the equations around that estimate using a first-order Taylor expansion, solve the resulting linear system, update the position estimate, and repeat until convergence. Convergence typically takes 2–5 iterations. When more than four satellites are visible — modern receivers routinely see 8–12 — the system is over-determined and a weighted least-squares solution minimizes the total squared pseudorange residuals, making the fix more robust to individual measurement errors. Source: VectorNav GNSS Error Budget (vectornav.com) and Wikipedia GPS positioning calculation.

Now run the core solve yourself. This shows how the simultaneous solution of four pseudorange equations yields position and clock correction:

runnable · js▸ run

output

Run the code to see its output here.

---

Atomic clocks: what makes time precise enough

The satellite carries an atomic clock. Your receiver carries a quartz oscillator. The satellite’s clock is trusted; the receiver’s clock is solved for. But why are atomic clocks so much better?

A quartz clock keeps time by counting the oscillations of a vibrating crystal — typically at 32,768 Hz. Acesium atomic clockA clock that counts oscillations of the hyperfine transition in cesium-133 atoms, which resonate at exactly 9,192,631,770 Hz — this frequency defines the SI second since 1967. counts oscillations of cesium-133 atoms’ quantum transition at 9,192,631,770 cycles per second. That number doesn’t change — it’s a fundamental constant of nature. Two cesium clocks anywhere in the universe will tick at the same rate (in the same gravitational environment). A quartz crystal, by contrast, drifts with temperature, aging, and manufacturing variance.

GPS Block II/IIA satellites carryrubidium and cesium clocksGPS Block II/IIA satellites carried both rubidium and cesium oscillators; Block IIR carries rubidium-only (PerkinElmer RAFS). Stability is approximately 1 part in 10¹³ or better over 12 hours. Source: GPS.gov performance standards. with fractional stability around 1 part in 10¹³ — meaning they accumulate less than one nanosecond of error per roughly ten thousand seconds.

Why does it matter so much? Becausethe speed of light is 30 cm per nanosecondc = 299,792,458 m/s, so 1 ns × c ≈ 0.3 m. A 30 ns clock error → ~9 m position error. Source: GPS.gov accuracy page, gps.gov/gps-accuracy. at 30 centimetres per nanosecond, a timing error of 30 nanoseconds — typical for a consumer-grade oscillator over an hour — translates to 9 metres of position error. The government commits to broadcasting GPS timing with accuracy≤30 nanoseconds relative to UTCGPS SPS Performance Standard: time transfer accuracy ≤30 ns, 95% of the time, relative to UTC(USNO). Source: gps.gov/performance-standards-specifications. of ≤30 nanoseconds relative to UTC. To achieve meter-level positioning, Physics Today notes that “to avoid navigation errors of more than a meter, an atomic clock must deviate by less than about 4 nanoseconds from perfect synchronization.”¹1Relativity and the Global Positioning System, Physics Today, 2002. The quote continues: “requiring fractional time stability better than a part in 10¹³.” Physics Today

The ground control segment continuously monitors the satellite clocks and uploads corrections to keep them calibrated. But there’s a deeper problem no ground station can fix — one that the physicists had to handle in the design of the clocks themselves, before any satellite launched.

---

The 38 microseconds: how Einstein saves GPS every day

Here is the problem the original GPS engineers faced: when you put a clock in orbit at 20,200 km altitude moving at 3.87 km/s, two effects from Einstein’s theories work on it simultaneously, in opposite directions.

SR

$$\frac{\Delta f_{\text{SR}}}{f_0} = -\frac{v^2}{2c^2} \approx -8.32 \times 10^{-11}$$

$$v = 3{,}870 \text{ m/s (orbital velocity)}$$whyGPS satellite speed (half-sidereal-day orbit at 20,200 km)

$$\frac{v^2}{2c^2} = \frac{(3{,}870)^2}{2 \times (2.998 \times 10^8)^2}$$whyvelocity-squared over 2c²

$$\approx 8.32 \times 10^{-11}$$whyfractional rate slowdown from special relativity

GR

$$\frac{\Delta f_{\text{GR}}}{f_0} = +\frac{\Delta U}{c^2} \approx +5.29 \times 10^{-10}$$

$$\Delta U = GM \left(\frac{1}{R_E} - \frac{1}{r_s}\right)$$whygravitational potential difference between surface and orbit

$$R_E \approx 6{,}378 \text{ km}, \quad r_s \approx 26{,}560 \text{ km}$$whyEarth radius vs. GPS orbital radius

$$\frac{\Delta U}{c^2} \approx 5.29 \times 10^{-10}$$whyfractional rate speedup from general relativity

The two effects combine to produce anet fractional rate differenceNet = GR blueshift (+5.29×10⁻¹⁰) + SR redshift (−8.32×10⁻¹¹) ≈ +4.465×10⁻¹⁰. Source: Physics Today GPS relativity; ESA Navipedia Relativistic Clock Correction. of approximately +4.465 × 10⁻¹⁰ — the gravitational effect wins. Translated into time: a GPS satellite clock, if designed to tick at the same nominal frequency as a ground clock, would run 38.5 microseconds fast per day relative to a clock at Earth’s surface.

38.5 microseconds sounds small. At the speed of light, it is not: c × 38.5 × 10⁻⁶ seconds ≈ ~11 km per day. A GPS receiver that ignored this correction would show your location drifting by ~11 km every day. In two minutes, you’d have a navigational fix good to a couple of miles — worse than a paper map.

Gravitational and motional frequency shifts are so large that, without carefully accounting for numerous relativistic effects, the system would not work.

The fix is elegant. Before launch, the onboard clock frequency is set not to 10.23 MHz (the nominal GPS fundamental frequency) but to10.22999999543 MHzThe factory frequency offset: f₀' = f₀ × (1 − 4.464×10⁻¹⁰) = 10.22999999543 MHz. When in orbit, relativity brings this back up to the nominal 10.23 MHz. Source: ESA Navipedia, Relativistic Clock Correction; Physics Today GPS relativity. — slightly slower by exactly the amount relativity will speed it up. Once in orbit, the clock ticks at exactly the right rate to stay synchronized with Earth-surface time.

There’s also a smaller, periodic correction: GPS satellites travel in slightly elliptical orbits. When the satellite is closer to Earth (moving faster, deeper in the gravity well), both relativistic effects shift; when it’s farther (moving slower, weaker gravity), they shift back. This eccentricity correction, up to about 1 microsecond in amplitude, is calculated in the receiver itself using the formulaΔ_rel = −2(r · v)/c²The periodic relativistic eccentricity correction. Neglecting it causes positioning errors up to 13 metres in range and over 20 metres vertically. Source: ESA Navipedia, Relativistic Clock Correction. where r and v are the satellite’s position and velocity vectors.

Now you can compute exactly what happens without the correction:

runnable · js▸ run

output

Run the code to see its output here.

This is not a small correction fudged in at the margins. It is baked into the hardware — into the physical oscillation rate of every GPS satellite clock — before it ever leaves the ground. General relativity is not optional equipment in the GPS system. It is load-bearing infrastructure.

---

The error budget: what eats your accuracy

Even with perfect atomic clocks and relativistic corrections handled, real-world GPS accuracy is limited by a stack of physical and geometric effects. Here is what that stack looks like:

chart: bar of meters by source — 6 points

Ionospheric delay is the largest single source. GPS signals are radio waves. When they pass through theionosphereThe layer of Earth's upper atmosphere (roughly 60–1000 km altitude) containing free electrons produced by solar radiation. These slow down radio waves by an amount proportional to the total electron content (TEC) along the signal path., free electrons slow the signal slightly. The delay is proportional to the electron density, which varies with the sun, season, and solar weather. For a signal arriving at a low elevation angle (grazing across a longer path through the ionosphere), a single-frequency receiver might see 5–20 metres of delay.ionospheric delay up to 30 metresMid-latitude ionospheric delay can range 8–30 metres depending on time of day, solar activity, and elevation angle. Source: ESA Navipedia Klobuchar model; VectorNav GNSS error budget.

TheKlobuchar ionospheric modelDeveloped in the 1970s for GPS single-frequency users. Eight parameters broadcast by satellites. Reduces ionospheric range error by about 50% RMS. Source: ESA Navipedia Klobuchar Ionospheric Model. (its parameters broadcast in the navigation message) halves this error for single-frequency receivers. Dual-frequency receivers — now common in flagship phones — do better: because the ionosphere delays different frequencies by different amounts (the delay scales as 1/f²), measuring on L1 and L5 lets the receiver compute and subtract the ionospheric delay directly, with no model required. That single technique drops ionospheric error from ~5 m to a few centimetres.

Ephemeris errors arise when the satellite’s reported orbital position differs slightly from where it actually is. GPS ground control uploads updated ephemeris data to satellites every two hours. The residual error is typically a few metres of range.

Multipath is what happens in cities: signals bounce off buildings before reaching your antenna, arriving on a path longer than the direct line of sight. The receiver can’t tell the reflected signal from the direct one and computes an erroneously long pseudorange. This is why GPS accuracy degrades dramatically in urban canyons — not because the signal is blocked (that just causes a loss of the satellite), but because the reflected copies corrupt the measurement.

Tropospheric delay comes from the lowest 12 km of atmosphere — water vapour and air density slow the signal slightly. Unlike the ionosphere, this delay is roughly the same for all frequencies and can be modelled fairly well from surface meteorological data.

Geometry matters: dilution of precision

Here’s a subtlety that surprises people: even with perfect measurements, the geometry of which satellites you can see determines how accurately the system can locate you.

dilution of precision (DOP)A dimensionless multiplier that quantifies how satellite geometry amplifies (or compresses) ranging errors into position errors. Low DOP (near 1) is ideal; DOP > 5 is poor. works like this: if all four visible satellites are clustered in one part of the sky, the spheres they define are nearly parallel — their intersection is poorly conditioned and any small ranging error creates a large position uncertainty. If the satellites are spread across the sky (one near the horizon in each cardinal direction, one overhead), the spheres intersect at acute angles and the position is tightly constrained.

You can feel this directly. In the sandbox below, each satellite reports a noisy distance and the receiver solves for its position over and over; the amber scatter is the cloud of resulting fixes and the ellipse is its 1σ spread. Drag Sky spread down to ~30° so the satellites bunch into a narrow arc and watch the same range noise smear the fix into a large, elongated cloud — that stretch, produced by geometry alone with the noise held constant, is dilution of precision.

GPS-style trilateration: satellites ring a receiver and each reports a noisy distance. Spread the satellites wide and the same noise gives a tight, round position fix (low DOP); bunch them into a narrow arc and the fix smears into a large, elongated cloud (high dilution of precision).

Tunable parameters

• Satellites5 (range 3–8)
• Sky spread (°)220 (range 20–340)
• Range noise σ (m)6 (range 0–20)

Mathematically, DOP is derived from the covariance matrix of the least-squares position solution. The position error in any dimension isσ_position = DOP × σ_rangeThe relationship between DOP and position accuracy. A DOP of 2 with 3 m range errors gives 6 m position error. Source: Error Analysis for the Global Positioning System, Wikipedia; VectorNav.:

$\sigma_{\text{position}} = \text{DOP} \times \sigma_{\text{range}}$

That is a linear amplifier, and you can turn its knobs. The curve below plots position error against range error; the slider sets the geometry factor DOP. At the read’s cited example — DOP = 2 with 3 m range errors — the readout lands on 6 m of position error. Push DOP toward 5 and the same 3 m of ranging noise balloons into 15 m: identical measurements, worse geometry, a worse fix.

σ_pos at σ_range = 3 m6 m

σ_pos at σ_range = 3 m: 6 m.

geometric dilution (DOP)2

A PDOP (3D position DOP) of 2 is good; a DOP of 1.5 is excellent; above 5 is poor. Urban canyons don’t just create multipath — they also hide satellites behind buildings, raising DOP and compounding the geometry penalty on top of the reflections.

Go deeper: HDOP, VDOP, TDOP, and why vertical accuracy is always worse

GPS position errors are not isotropic. Horizontal accuracy (HDOP) is typically about half the vertical accuracy (VDOP). The reason is geometric: satellites appear in the upper hemisphere only (they don’t pass under you), so the vertical sphere intersections are less evenly distributed. A PDOP of 2 might decompose into HDOP ≈ 1 and VDOP ≈ 1.7. Vertical GPS errors are typically 1.5–2× horizontal errors. TDOP captures timing uncertainty in the same framework — when DOP is high, not only position but also the receiver’s clock solution is degraded. This is why GPS time transfer (used in telecommunications, financial systems, and scientific instrumentation) specifies its own error bound separately from position.

---

How the receiver actually solves it: the full loop

Here is the architecture of what the receiver is actually doing — four satellite signals converging on a single geometry solve, with three correction channels feeding the solver before it outputs a position:

Nodes

• Satellite 1 (timestamped signal)
• Satellite 2 (timestamped signal)
• Satellite 3 (timestamped signal)
• Satellite 4 (timestamped signal)
• Receiver (4 pseudoranges ρ₁…ρ₄)
• Relativistic correction
• Ionospheric model (Klobuchar / dual-freq)
• Ephemeris (satellite positions)
• Least-squares solver 4 equations · 4 unknowns (x, y, z, δt)
• Position + corrected clock (lat / lon / alt + GPS time)

Connections

• Satellite 1 (timestamped signal) → Receiver (4 pseudoranges ρ₁…ρ₄) (ρ₁)
• Satellite 2 (timestamped signal) → Receiver (4 pseudoranges ρ₁…ρ₄) (ρ₂)
• Satellite 3 (timestamped signal) → Receiver (4 pseudoranges ρ₁…ρ₄) (ρ₃)
• Satellite 4 (timestamped signal) → Receiver (4 pseudoranges ρ₁…ρ₄) (ρ₄)
• Receiver (4 pseudoranges ρ₁…ρ₄) → Least-squares solver 4 equations · 4 unknowns (x, y, z, δt) (corrected pseudoranges)
• Relativistic correction → Least-squares solver 4 equations · 4 unknowns (x, y, z, δt) (−4.464×10⁻¹⁰ offset)
• Ionospheric model (Klobuchar / dual-freq) → Least-squares solver 4 equations · 4 unknowns (x, y, z, δt) (delay correction)
• Ephemeris (satellite positions) → Least-squares solver 4 equations · 4 unknowns (x, y, z, δt) (satellite positions)
• Least-squares solver 4 equations · 4 unknowns (x, y, z, δt) → Position + corrected clock (lat / lon / alt + GPS time) (solve (x,y,z,δt))

The key insight the diagram makes visible: the four satellite signals are four equations; the three correction inputs (relativistic offset, ionospheric model, ephemeris) reduce systematic errors before the solver ever runs; and the solver outputs both position and clock correction simultaneously from a single algebraic system.

Here is the end-to-end sequence from radio signal to coordinate:

1. 01

   1. Receive + correlate the PRN code

   Each GPS satellite broadcasts a unique pseudo-random noise (PRN) code — a deterministic sequence of bits that looks random. The receiver generates its own copy of this code and slides it in time until it aligns with the incoming signal. The time shift at peak correlation is the signal travel time Δt. This works even when the signal is weaker than the background noise, because correlation integrates over a long code sequence.

2. 02

   2. Extract the navigation message

   Superimposed on the PRN code is a navigation message transmitted at 50 bits per second: the satellite's precise orbital parameters (Keplerian elements + corrections), its clock corrections (a polynomial describing how the atomic clock deviates from GPS time), ionospheric Klobuchar parameters, and a health flag. The receiver uses the orbital parameters to compute exactly where the satellite was when it transmitted.

3. 03

   3. Compute pseudoranges

   For each tracked satellite, the receiver multiplies the travel time Δt by the speed of light: ρ = c × Δt. This is the pseudorange — the true geometric range plus the receiver clock bias term c·δt. Multiple satellites are tracked simultaneously (typically 8–12 in an open-sky environment).

4. 04

   4. Apply corrections

   The receiver applies satellite clock corrections (from the navigation message), the Klobuchar ionospheric model, and the relativistic eccentricity correction Δ_rel = −2(r·v)/c². Each of these adjustments modifies the pseudoranges before position solving begins.

5. 05

   5. Linearize and iterate (Gauss-Newton)

   With corrected pseudoranges in hand, the receiver linearizes the nonlinear equation system around an initial position estimate using a Taylor expansion, solves the resulting 4×4 (or overdetermined) linear system, updates its position estimate, and repeats. Three to five iterations are typically enough. The solved δt is fed back to the receiver clock — the phone now knows GPS time to sub-microsecond accuracy.

6. 06

   6. Output position + uncertainty

   The final output is an ECEF (Earth-Centered Earth-Fixed) coordinate (x, y, z) converted to latitude, longitude, and altitude. The receiver also outputs DOP values and satellite count, giving applications a quality estimate. A DOP of ≤2 with ≥6 satellites is a reliable fix; higher DOP or fewer satellites degrades confidence.

When all of this works — atomic clocks broadcasting timestamps, four equations solved simultaneously, relativistic corrections applied, ionospheric delays modelled or cancelled, DOP geometry acceptable — the government commits to≤2.0 m user range error 95% of the timeGPS SPS Performance Standard (5th ed., April 2020): daily global average URE ≤2.0 m, 95% probability. Actual April 2021 performance: ≤0.643 m. Source: gps.gov/performance-standards-specifications. across all healthy satellites. The measured performance as of 2021 was 0.643 metres RMS — less than the length of your forearm.

---

Check your understanding

1. Q1Why does GPS need four satellites instead of three to compute a 3D position fix?

   • Three satellites only provide 2D positioning; the fourth adds altitude.
   • The receiver's clock is unknown — the fourth satellite provides the extra equation to solve for x, y, z, and clock bias simultaneously.
   • Three satellites don't provide enough signal strength to get an accurate fix.
   • Four satellites are needed to cancel ionospheric delay.

2. Q2GPS satellite clocks are set to tick at 10.22999999543 MHz instead of 10.23 MHz before launch. Why?

   • Manufacturing tolerances make exact 10.23 MHz impossible to achieve.
   • The slower rate compensates for special relativistic frequency shifts during launch.
   • Relativity will speed them up once in orbit — the factory offset ensures they tick at exactly 10.23 MHz from the ground's perspective.
   • It reduces interference between satellites.

3. Q3A GPS receiver in an urban canyon shows lower accuracy than in an open field, even when it can still see four satellites. What are the two main reasons?

   • Signal attenuation and antenna orientation.
   • Multipath reflections corrupt the pseudoranges, and poor satellite geometry (high DOP) amplifies those errors.
   • Ionospheric scintillation is stronger near buildings.
   • The receiver clock drifts faster near concrete.

---

From concept to constellation: the milestones that made it real

Event 1 of 7: 1 Sep 1973, GPS program initiated

011973opening

1 Sep 1973

GPS program initiated

U.S. Department of Defense launches the Defense Navigation Satellite System (DNSS) project, later renamed NAVSTAR GPS, merging competing Navy and Air Force programs.

021978+4 yr

22 Feb 1978

First Block I satellite

Navstar 1 launches from Vandenberg Air Force Base — the first experimental GPS satellite. Ten Block I prototypes would validate the concept by 1985.

031983+6 yr

16 Sep 1983

Reagan opens civilian access

KAL 007 was shot down over Soviet airspace on September 1, 1983. Sixteen days later, on September 16, President Reagan announces GPS will be made available to civilian aviation once operational — to prevent future navigation tragedies.

041993+10 yr

8 Dec 1993

Initial operational capability

24-satellite constellation in orbit. GPS declared initially operational for civilian use.

051995+1 yr

27 Apr 1995

Full operational capability

Air Force Space Command formally declares Full Operational Capability on April 27, 1995 — all 24 satellites healthy, GPS works everywhere on Earth, continuously.

062000+5 yr

2 May 2000

Selective Availability disabled

President Clinton orders Selective Availability — deliberate civilian GPS degradation to ~100 m — switched off permanently. Civilian accuracy improves roughly tenfold to ~5 m overnight.

072018+19 yr

23 Dec 2018

First GPS III satellite

GPS III SV01 launches — carrying a more powerful L1C signal, triple the accuracy of earlier blocks, and a search-and-rescue payload.

One detail worth noting: the 2000 Selective Availability switch-off shows how much of GPS’s practical usefulness was policy, not physics. The atomic clocks and relativistic corrections were always there. The artificial degradation was a deliberate wobble injected into the satellite clocks to blur civilian coordinates. When that was removed, nothing changed in the physics — the same constellation, the same signals, the same Einstein corrections — just more honesty about what the system could actually do.

---

The mental model to take away

GPS is a time-distribution system. Satellites broadcast clocks; you receive timestamps; you turn time differences into distances; distances become spheres; overlapping spheres pin a location in space.

Three deep features make it work:

1. Four equations, not three — the fourth satellite doesn’t just add position information. It closes the algebraic system that simultaneously locates you and calibrates your clock. The “free atomic clock in every phone” is a consequence of geometry, not extra hardware.

2. Atomic clocks, not because of accuracy alone — you could achieve the same effect with a perfectly synchronized cheap clock. Atomic clocks matter because they stay synchronized across the hours between ground control uploads, without drifting enough to corrupt the pseudoranges. The stability of cesium is the bedrock on which the whole timing chain stands.

3. Relativity is not a correction, it is infrastructure — before the first Block II satellite launched, physicists calculated the relativistic correction and baked it into the hardware as a physical frequency offset. GPS is the only technology in mass daily use where general relativity is not a theoretical nicety but a first-order engineering requirement. Ignore it and you lose ~11 km of accuracy per day.

The next time your phone drops you on a map to within the width of a doorway, what happened is: the laws of physics — special and general relativity, the constancy of the speed of light, the quantum mechanics of cesium — conspired to make the laws of geometry solvable. You just happened to be sitting on the intersection point of four spheres.