GNSS code multipath · Rust core · Python API

Measure multipath. Map where it comes from.

geoveil-mp reads RINEX 2, 3 and 4 and estimates code multipath for every pseudorange signal on GPS, GLONASS, Galileo, BeiDou, QZSS and NavIC, with cycle slips detected per carrier. Version 1.0 adds SNR wavelet analysis and first Fresnel zone mapping, so the numbers point back to the ground that caused them.

$pip install geoveil-mp

Read the API Open the live platform

14–28×faster SNR wavelets

1.8e-15max difference vs numpy

10new Python functions

First Fresnel zone explorer

eq. 6 · Hunegnaw & Teferle, Sensors 2022

Antenna height 2.0 m

Satellite elevation 15°

Carrier

L1 L2 L5

Semi-major a: 4.896 m
Semi-minor b: 1.267 m
Specular distance d: 7.464 m

<15° 15–30° 30–60° >60°

gm.fresnel_zone(15.0, 2.0) # → (4.896, 1.267, 7.464)

b = √(λh / sin θ + (λ / 2 sin θ)²), a = b / sin θ, d = h / tan θ. Satellite tracks are illustrative; the geometry is the library's.

Release1.0.1 · stable

What 1.0 adds

The wavelet and Fresnel analysis that the GeoVeil batch platform ran in numpy now lives in the library, in Rust, with the Python API you already use. Everything from 0.2 works as before.

SNR residuals

Wavelet spectra per satellite

SNR is converted from dB-Hz to linear amplitude, detrended arc by arc, and passed through a Morlet transform. You get band power in periods matched to your sampling interval and a 95 % red-noise significance level.

analyze_snr_waveletsscalogramband_power

Reflector geometry

First Fresnel zones

For every 10° × 10° cell a satellite crosses, the library returns the zone's semi-axes and its distance from the antenna, for your antenna height and carrier. Large multipath now has an address on the ground.

fresnel_zonefresnel_map

Parity

Same numbers, 20× faster

Each function was checked against the numpy reference on synthetic series and on a real station hour. Satellites run in parallel and the GIL is released, so thread pools keep working.

morlet_cwtdetrend_arcsplit_arcs

Linux container, release build, same RINEX 3.04 file for both, checked against the numpy implementation the GeoVeil platform used before 1.0.
Measured on	numpy reference	geoveil-mp 1.0
SNR wavelets, 51 satellites, 3,600 epochs at 1 Hz	7.05 s	0.35 s
Largest relative difference in wavelet power	reference	1.8 × 10⁻¹⁵
Largest relative difference in detrended residuals	reference	4.3 × 10⁻¹²

SpeedIntel Xeon E5-2620 v3 @ 2.40GHz · 12 threads

A station day in 2.0 seconds

Wall-clock time for the full pipeline on public IGS files from 31 December 2025, measured with the geoveil-mp 1.0.1 wheel from PyPI on a 2014 server CPU. Parsing and multipath analysis run on one core. SNR wavelets use every core.

Seconds per file with all 12 threads, median of three runs. Hatanaka files are decompressed beforehand; the table lists that step.

Stress test

24 hours at 1 Hz: a 948 MB file with 86,400 epochs and 127 satellites takes 66 s and peaks at 10.7 GB of memory. Split high-rate days into hours to keep memory near 562 MB.

Hatanaka: decompression to plain RINEX with the hatanaka Python package, not part of geoveil-mp. vs numpy: SNR wavelet stage against the numpy implementation the GeoVeil platform used before 1.0. Scripts and raw results are in the benchmarks folder.
File	Epochs × sats	Hatanaka	Parse	Multipath	SNR wavelets	Total	1 thread	Peak memory	vs numpy
RINEX 2.11 · 24 h · 30 s ZIMM · Hatanaka + gzip · 5.4 MB	2,880 × 32	0.22 s	0.15 s	0.42 s	0.07 s	0.64 s	0.90 s	129 MB	18×
RINEX 3.02 · 24 h · 30 s BOR1 · Hatanaka + gzip · 19.4 MB	2,880 × 106	0.69 s	0.34 s	1.47 s	0.20 s	2.01 s	3.43 s	398 MB	22×
RINEX 3.02 · 24 h · 30 s BOR1 · gzip, read directly · 5.8 MB	2,880 × 106	–	0.45 s	1.75 s	0.21 s	2.42 s	3.59 s	401 MB	23×
RINEX 3.05 · 15 min · 1 Hz BUCU · Hatanaka · 9.6 MB	900 × 44	0.30 s	0.14 s	0.44 s	0.08 s	0.66 s	0.98 s	135 MB	28×
RINEX 3.05 · 1 h · 1 Hz BUCU · Hatanaka · 39.9 MB	3,600 × 52	1.18 s	0.61 s	1.94 s	0.35 s	2.90 s	4.57 s	562 MB	19×
RINEX 3.05 · 24 h · 1 Hz BUCU · Hatanaka · 948.1 MB	86,400 × 127	28.4 s	14.6 s	46.4 s	5.13 s	66.1 s	96.3 s	10.7 GB	14×

In productionHAR1 · 2025-12-31 · 30 s

Running on batch.geoveil-rinex.eu

The GeoVeil batch platform runs this analysis on every station day it receives. These screenshots show one day of the HAR1 station, a daily 30 s RINEX file. Select a chart to see it full size.

GeoVeil platform, multipath overview tab for HAR1 on 31 December 2025, with stat cards and a bar chart of multipath RMS per signal code. — **Multipath overview.** 11 signals, 199,548 multipath estimates and 2,938 cycle slips. The worst signal, GLONASS M3Q, sits at 0.395 m RMS.

Polar map of first Fresnel zone footprints around the antenna, coloured by elevation band. — **Fresnel footprints.** 3,029 zones for a 1 m antenna. The empty wedge to the north matches the gap in satellite coverage.

Morlet wavelet scalogram of detrended SNR for satellite R02, period against time, with significance marks and cone of influence. — **Morlet scalogram, GLONASS R02.** White marks pass the 95 % red-noise test. The dashed curve is the cone of influence, below which edge effects dominate.

Skyplot of Galileo E1 multipath magnitude along satellite tracks, most points green, larger values at low elevation. — **Multipath skyplot, Galileo E1.** Low passes in the north-west and east carry most of the large values.

Line chart of multipath RMS against elevation in 2 degree bins for GPS, GLONASS and Galileo L1 codes. — **RMS against elevation.** 2° bins, one L1 code per constellation.

Scale-averaged wavelet power over time for four period bands from 60 seconds to 30 minutes. — **Band power, R02.** Four period bands from 60 s to 30 min.

Bar chart of cycle slips counted per carrier-phase signal. — **Cycle slips** counted per carrier-phase signal.

A year of daily multipath RMS for HAR1 with 30-day rolling medians per constellation, all below a dashed 0.50 metre line. — **A year of HAR1.** Daily RMS per constellation with 30-day rolling medians, all well under the 0.50 m alert line.

Explore the platformNo sign-in needed to browse.

MethodAnubis / TEQC style

From observables to reflectors

Two processing lanes share one parsed file. The code lane measures how much multipath each signal carries. The SNR lane finds its periods and the Fresnel step places it around the antenna.

MP_k = P_k − (1 + 2/(α−1)) Φ_i + (2/(α−1)) Φ_j, α = (f_i / f_j)² Φi is the phase on the code's own band, Φj a partner band from a fixed per-system priority list. GLONASS uses each satellite's channel number from the header.

Code multipath

Parse RINEX 2, 3 or 4 into code, phase and SNR per satellite and epoch.
Form MP_k for every pseudorange code with a deterministic phase pairing.
Detect cycle slips with geometry-free, code-phase and loss-of-lock tests whose thresholds scale with the interval.
Remove the ambiguity bias per arc, resetting at gaps, slips and jumps.
Report per signal: RMS, elevation-weighted RMS and slip counts, with elevations from SP3 or broadcast orbits.

SNR and reflectors new

Pick the band-1 S observable for each satellite.
Convert dB-Hz to linear amplitude and split into arcs at gaps.
Detrend each arc with the polynomial of order 2 to 9 that leaves the smallest residual.
Run the Morlet transform and average power into four period bands, with red-noise significance.
Place first Fresnel zones along each track for your antenna height and carrier.

24 h of BOR1 (EPN, Poland), 30 s multi-GNSS RINEX 3, default settings. The ranking reproduces published station-monitoring results.
Constellation	Best signal	RMS	Worst signal	RMS
Galileo	E5 AltBOC (M8X)	0.165 m	E5b (M7X)	0.395 m
GPS	L2C (M2X)	0.312 m	L1 C/A (M1C)	0.342 m
BeiDou	B2I (M7I)	0.295 m	B3I (M6I)	0.354 m
GLONASS	L2 P (M2P)	0.318 m	L1 C/A (M1C)	0.566 m

APIPython 3.9+ · Rust 1.75+

Five lines to a station report

Install from PyPI, point it at a RINEX file, and read plain Python lists and dicts back. The Rust crate exposes the same functions.

import geoveil_mp as gm

obs = gm.read_rinex_obs("HAR100ROU_R_20253650000_01D_30S_MO.rnx")

# Code multipath and cycle slips per signal
res = gm.MultipathAnalyzer(obs, 10.0, ["G", "R", "E"]).analyze()
for s in res.statistics:
    print(s.signal, round(s.rms, 3), s.cycle_slips)   # GLOM3Q 0.395 ...

# New in 1.0: SNR wavelets for every satellite, in parallel
w = gm.analyze_snr_wavelets(obs)
w["bands"]                       # [[60, 120], [120, 300], [300, 900], [900, 1800]]
w["band_power"]["R02"]["power"]  # binned power, one series per band

r = w["residuals"]["R02"]
scal = gm.scalogram(r["t"], r["v"], w["interval"])   # power, significance, coi

# First Fresnel zone for a 2 m antenna at 15° elevation, GPS L1
a, b, dist = gm.fresnel_zone(15.0, 2.0)                    # (4.896, 1.267, 7.464) m

use geoveil_mp::analysis::advanced::{fresnel_zone, morlet_cwt, snr_wavelets, SnrWaveletConfig, DJ, DEFAULT_WAVELENGTH};
use geoveil_mp::RinexObsReader;

let obs = RinexObsReader::new().read("HAR100ROU_R_20253650000_01D_30S_MO.rnx")?;

// SNR wavelets for all satellites (rayon-parallel)
let sats = snr_wavelets(&obs, &[], None, &SnrWaveletConfig::default());

// Band power of one residual series
let cwt = morlet_cwt(&sats[0].residual, 30.0, DJ, None);
let power = cwt.band_power((120.0, 300.0));

// First Fresnel zone: a, b and distance in metres
let z = fresnel_zone(15.0, 2.0, DEFAULT_WAVELENGTH);

New in 1.0	Returns
analyze_snr_wavelets(obs)	Residuals and binned band power per satellite, in the layout the platform stores
scalogram(t, values, interval)	Power matrix, periods, 95 % significance and cone of influence for the longest arc
morlet_cwt(y, dt)	Wavelet power, scales, Fourier periods and cone of influence
band_power(y, dt, bands)	Scale-averaged power for each (low, high) period band
red_noise_significance(y, dt, periods)	Lag-1 red-noise significance level per period
period_bands(dt)	The four default bands for a sampling interval
detrend_arc(t, values)	Residual and the chosen polynomial order
split_arcs(times, dt)	(start, stop) index pairs of continuous arcs
fresnel_zone(elevation, height)	Semi-major axis, semi-minor axis and specular distance in metres
fresnel_map(tracks, height)	Zone footprints along satellite tracks, one per 10° × 10° cell

LicensePolyForm Noncommercial 1.0.0

Free for science. Licensed for business.

From 1.0, geoveil-mp is source-available under PolyForm Noncommercial 1.0.0 with an attribution term.

Non-commercial use

Free, with credit

Research, teaching and personal projects.
Universities, public research bodies, charities and government agencies.
Credit "geoveil-mp by Miluta Dulea-Flueras" with a link to the repository, and cite it in publications.

Commercial use

Ask for a license

Products, paid services and work done for a company need a separate commercial license. Write to:

miluta.flueras@cartografie.ro

Versions 0.2.1 and earlier remain available under the MIT license. The full terms are in the LICENSE file.

CitationBibTeX

geoveil_mp.bib

@software{geoveil_mp,
  author  = {Dulea-Flueras, Miluta},
  title   = {geoveil-mp: GNSS Code Multipath Analysis Library},
  year    = {2026},
  version = {1.0.1},
  url     = {https://github.com/miluta7/geoveil-mp},
  license = {PolyForm-Noncommercial-1.0.0}
}

Method: Hunegnaw, A.; Teferle, F.N. Evaluation of the Multipath Environment Using Electromagnetic-Absorbing Materials at Continuous GNSS Stations. Sensors 2022, 22, 3384. Wavelets: Torrence, C.; Compo, G.P. A Practical Guide to Wavelet Analysis. Bull. Amer. Meteor. Soc. 1998, 79, 61–78.