GPS Multipath Error

Of all the contributors to the GPS error budget (covered in /learn/gps-accuracy-explained), multipath is the one most variable from place to place and the hardest to remove. It's also the dominant reason smartphone GPS in cities is much less accurate than the ~5 m open-sky figure suggests. This article unpacks the geometry, the magnitudes, and the mitigation strategies.

The geometric mechanism

A GPS satellite broadcasts a radio signal at ~1.5 GHz. The signal takes some path from the satellite to the receiver. In an open environment, the path is a straight line: direct line-of-sight.

In an environment with reflective surfaces — buildings, glass, water, metal roofs, vehicles — the signal also reaches the receiver via reflected paths. The reflected paths are longer than the direct path; the reflected signals arrive at the receiver a few nanoseconds to a few microseconds later.

The receiver, attempting to compute a pseudorange from the signal's arrival time, has two situations:

• Direct + reflected reaching receiver simultaneously: the receiver's code-tracking algorithm sees a distorted correlation peak. The peak's apparent timing is intermediate between the direct and reflected, biasing the pseudorange longer.
• Direct blocked, reflected only: the receiver tracks the reflected signal as if it were the direct one. The pseudorange is too long by the difference in path lengths — often tens of metres.

In both cases, the result is an inflated pseudorange: the satellite appears further away than it really is. The position fix shifts in the direction of the dominant reflectors. In a city with buildings on one side and open sky on the other, the position bias is consistently toward the buildings — sometimes producing systematic errors of 50+ metres.

Typical magnitudes by environment

| Environment | Typical multipath error | | ------------------------ | -------------------------- | | Open sky, no reflectors | < 0.5 m | | Suburban, some reflectors | 1–3 m | | Urban, mid-rise (5 storeys) | 3–10 m | | Urban dense (skyscrapers) | 10–50 m | | Indoor or basement | Variable; often non-functional | | Maritime open ocean | 1–3 m (water-surface reflection) | | Vehicle interior | 5–20 m (windshield + body) |

Multipath is not constant — it's time-varying. As the receiver moves and the satellites move across the sky, the geometry of reflectors changes and the multipath pattern changes too. A smartphone reading in the same downtown location can jump by 50 m within seconds as the multipath geometry shifts.

The frequency dependence

Multipath behaviour varies by GPS frequency:

• L1 (1575.42 MHz): most affected because the wavelength (~19 cm) is small enough that many surfaces produce measurable reflections.
• L5 (1176.45 MHz): longer wavelength (~25 cm), slightly less affected. Also has a wider bandwidth, which lets the receiver use narrow-correlator code-tracking that's more multipath-resistant.
• L2 (1227.60 MHz): similar to L1 in multipath sensitivity.

Dual-frequency GPS (L1 + L5, standard on premium smartphones since ~2018) gives some multipath improvement because the receiver can compare measurements across frequencies and identify which is more affected by multipath.

Receiver-side mitigations

Modern GPS receivers use several techniques to reduce multipath:

Narrow correlator code tracking. Standard GPS receivers correlate the received signal against an internal reference code; the correlation peak gives the timing measurement. A narrow correlator (sampling the peak at smaller offsets) is less affected by multipath-induced peak distortion. Modern receivers use narrow correlators by default.

Carrier-phase smoothing. The receiver uses the carrier-phase measurement (more precise than code-phase) to smooth the code-based pseudorange over time. Multipath errors in carrier-phase are smaller in magnitude (~cm vs ~m) and more frequently change sign, so averaging tends to reduce them.

Multi-frequency observation. With L1 + L5 dual-frequency receivers, the receiver compares pseudoranges across frequencies. Multipath affects each frequency differently; the receiver weights the less-multipath-affected frequency more in the position fix.

Multipath estimator algorithms. Some high-end receivers attempt to explicitly model and remove the multipath component using techniques like the Multipath Estimating Delay Lock Loop (MEDLL). Effective but computationally expensive; reserved for survey-grade and military hardware.

Antenna design. The antenna's gain pattern can suppress signals from below the horizon (the typical multipath direction). Choke-ring antennas (used in geodetic CORS stations) have a series of concentric rings that block ground-reflected signals while preserving the sky-coming direct signals.

Controlled Reception Pattern Antennas (CRPAs). Multi- element antenna arrays with adaptive beam-forming. Originally military for anti-jam; also reduce multipath because reflections usually come from non-up directions. Standard CRPAs are large and expensive ($10K+); smaller commercial versions exist for high-end surveying and autonomous-vehicle applications.

Site-selection rules

For permanent GNSS installations (CORS stations, RTK base stations, geodetic monuments), the receiver location is chosen to minimise multipath:

• Antenna elevated: at least 2 m above ground; preferably 10+ m for survey-grade installations.
• Open sky view: no obstructions within 15° elevation of the antenna.
• No nearby reflective surfaces: no large windows, no water bodies, no metal roofs within ~10 m.
• Stable mounting: the antenna doesn't move (movement introduces additional multipath variability).
• Ground plane: a metal disc beneath the antenna rejects ground-reflected signals.

For temporary survey rovers, the operator typically chooses to work in less-reflective environments when possible, or waits for favourable satellite geometry (avoiding low-elevation satellites that are more affected by multipath).

For smartphone users, the practical mitigation is simple: step outside, away from buildings, with the phone's antenna pointing upward (not lying flat with the screen down).

A worked example

Consider a smartphone GPS in an urban canyon, with a 30-storey building 50 m to the west. A GPS satellite at low elevation broadcasts its signal eastward; the direct path reaches the phone, but a strong reflection off the building's glass facade also arrives at the phone, delayed by:

Direct path:    typical
Reflected path: 2 × 50 m extra = 100 m extra
Delay:          100 m / (3 × 10⁸ m/s) ≈ 333 ns

The receiver sees the satellite signal arriving 333 ns later than the actual direct arrival. The computed pseudorange is inflated by 100 m. The position fix shifts ~30 m east (away from the building) compared to the true position.

In a real environment, multiple reflectors and multiple satellites combine; the position bias is the vector sum of all the contributions. Cities with tall buildings on multiple sides produce position errors that vary unpredictably as the satellite geometry changes.

Multipath in autonomous vehicles

Autonomous vehicles operate in environments multipath-rich by construction (cities, with buildings, other vehicles, road signs). The standard solution: don't rely solely on GPS. Most autonomous vehicles use sensor fusion combining:

• GPS (multi-GNSS, with multipath mitigation)
• Inertial measurement unit (IMU) — gyroscopes and accelerometers for short-term dead reckoning
• Visual odometry — cameras tracking visual features in the environment
• LiDAR — laser scanning for environmental mapping
• HD maps — pre-built models of the road network

The fused position is much more robust against multipath than GPS alone. When GPS shows a multipath-affected jump, the other sensors maintain the smooth trajectory, and the algorithm weights GPS less in those moments. For lane-level positioning (which AV systems require), this multi-sensor architecture is essential.

Multipath in CORS networks

National geodetic networks (NOAA NGS CORS, EUREF EPN, GeoNet NZ, etc.) operate hundreds of permanently-installed GNSS receivers used for surveying, science, and as RTK base stations. Multipath is a major concern because:

• Permanent installations have to stay in one place for decades; site selection matters dramatically.
• Multipath patterns repeat with satellite orbital periods (~12 hours for GPS), allowing characterisation and potentially modelling.
• Survey-grade applications using these stations need every millimetre of accuracy.

CORS station siting follows strict standards: choke-ring antennas, elevated mounting, open sky view, no reflective surfaces within specified distances. Even with all of this, residual multipath remains the largest contributor to CORS positioning noise at the millimetre level.

Common misconceptions

“Multipath only affects code-based positioning.” Code-based multipath is much larger (~10 m typical) than carrier-phase multipath (~cm typical), but both exist. RTK work that depends on millimetre carrier-phase precision is still affected by carrier-phase multipath, just at a smaller scale.

“Modern receivers eliminate multipath.” Modern receivers reduce multipath substantially compared to 1990s hardware, but no civilian receiver eliminates it. Multipath is a property of the environment, not the receiver; the receiver can mitigate but not remove it.

“Multi-GNSS solves multipath.” It helps by providing more satellites and better geometric weighting, but doesn't fundamentally remove multipath. A city with multipath affects all GNSS frequencies, not just GPS L1.

“Multipath only matters in cities.” Maritime multipath (off the water surface) is real and can produce 1–3 m errors on ships. Aviation multipath (off the airframe and nearby aircraft on the ramp) affects ground operations. Vehicle-interior multipath (off the windshield) is in every moving car. Multipath is everywhere; cities are just the worst case.

“You can predict multipath from a static environment.” Multipath is time-varying as the satellites move across the sky. Even in a permanently fixed location with permanent reflectors, the multipath pattern changes hour by hour. You can characterise it statistically but not predict it precisely.

“Multipath only inflates pseudoranges.” Multipath can also produce pseudoranges that are shorter than the true direct distance, in certain reflection geometries (though rarer than inflated cases). The full range of multipath effects is highly geometry-dependent; the “multipath always inflates pseudorange” rule of thumb is approximately but not strictly true.

“Carrier-phase multipath averages out to zero over time.” Code multipath has a strong random component that averages out partially; carrier-phase multipath is periodic with the satellite repeat cycle (about 12 hours for GPS) and doesn't average out over short intervals. Survey-grade RTK work that requires sub-2-cm accuracy has to either characterise the site multipath or wait through a full repeat cycle for accurate position determination.

“Putting the phone in your pocket fixes multipath.” Worse — pocket multipath adds your body as a reflective surface and attenuator. Best smartphone GPS reception is with the phone held away from your body, antenna upward, open sky.