GPS: Global Positioning and Signal Processing

• GPS is a U.S. space-based radionavigation system that uses satellite signals for accurate three-dimensional positioning, velocity, and timing.
• The system relies on pseudorange trilateration and advanced filtering methods like Kalman filters to mitigate errors from atmospheric and urban environments.
• GPS integration with inertial sensors and multimodal data pipelines enhances performance in urban canyons, spoofing detection, and time transfer applications.

The Global Positioning System (GPS) is a U.S. space-based radionavigation system that provides reliable positioning, navigation, and timing services to civilian users on a continuous worldwide basis, freely available to all (Ahamed, 2010). It provides specially coded satellite signals that can be processed in a GPS receiver, enabling the receiver to compute position, velocity and time, and it basically works by using four GPS satellite signals to compute positions in three dimensions together with the time offset in the receiver clock (Ahamed, 2010). GPS is now both a core infrastructure for navigation and a timing reference for banking, mobile phone operations, and power grid control, while contemporary research treats it as a probabilistic sensing system whose performance depends on geometry, propagation, signal integrity, fusion with other sensors, and the structure of downstream data pipelines (Ahamed, 2010, Ranacher et al., 2015, Abrar et al., 2024).

1. System architecture and measurement principle

GPS is commonly described through three segments: the space segment, the Operational Control Segment, and user equipment (Ahamed, 2010). The space segment comprises a constellation of 24 NAVSTAR satellites, plus spares, in six orbital planes with nearly circular orbits at an altitude of about 20,200 km and inclination $55^\circ$ to the equator, providing global coverage (Ahamed, 2010). GPS satellites broadcast radio signals at L1 and L2 frequencies together with pseudo-random code, ephemeris data, and almanac data, and GPS receivers ranging from wristwatch size to shipboard units process those signals to compute position, velocity, and time (Ahamed, 2010).

The fundamental positioning model is pseudorange-based trilateration. If the receiver measures signal travel time from satellite $i$, the corresponding distance estimate is expressed as

$$\text{Distance}_i = c \cdot (t_{\mathrm{Rx}} - t_{\mathrm{Tx}_i} + \delta t),$$

where $c$ is the speed of light and $\delta t$ is receiver clock bias (Ahamed, 2010). With four satellites, the receiver solves for the unknown three-dimensional coordinates and the receiver clock offset. In the usual formulation,

$$(x-x_i)^2 + (y-y_i)^2 + (z-z_i)^2 = [c \cdot (t_{\text{Rx}}-t_{\text{Tx}_i}+\delta t)]^2,$$

for $i=1,\dots,4$ (Ahamed, 2010). Velocity can be derived from change in position or Doppler shift, and time is obtained by correcting the receiver clock against satellite atomic clocks (Ahamed, 2010).

The satellite platform itself is an active control problem rather than a passive signal source. GPS satellites operate in medium Earth orbit and must maintain precise Earth-pointing attitudes to transmit signals effectively (Samir, 3 Aug 2025). Recent review work describes three-axis stabilization, reaction wheels for fine torque control, magnetorquers for momentum dumping and coarse or backup control, hydrazine thrusters for momentum management and station-keeping, and estimation architectures based on star trackers, Sun and Earth sensors, gyroscopes, magnetometers, and Extended Kalman Filter sensor fusion (Samir, 3 Aug 2025).

2. Accuracy limits, error sources, and statistical structure

Although GPS provides accurate location and time information in all weather, day and night, anywhere in the world, practical performance is limited by satellite clock or ephemeris errors, ionospheric and tropospheric delay, multipath, receiver noise, and satellite geometry (Ahamed, 2010). Geometric Dilution of Precision (GDOP) summarizes the effect of satellite configuration on solution accuracy, and later timing-focused work similarly emphasizes Time Dilution of Precision (TDOP) and the number of visible satellites as major quality factors (Ahamed, 2010, Park et al., 2024).

Urban and partially shadowed environments are a recurring failure mode. GPS location imprecision in dense areas of cities is attributed to proximity to walls and buildings, while indoor spaces and urban canyons degrade signal reliability through obstruction, multi-path errors, and non-line-of-sight conditions (Salarian et al., 2015, Alaba, 2024, Charan et al., 2024). In such settings, research increasingly treats GPS error as site-specific rather than homogeneous. A recent vehicle-to-infrastructure study reports that GPS error variance is grid- and site-specific and not uniform across a scene, motivating location-dependent denoising strategies (Charan et al., 2024).

A particularly important statistical result is that GPS trajectory measurements do not merely fluctuate randomly around truth; they can be systematically biased. For movement sampled at high frequencies, measurement error dominates interpolation error, and the distance between two points recorded with a GPS is, on average, bigger than the true distance between these points (Ranacher et al., 2015). If the two true positions are separated by $d_0$, then

$$\mathbb{E}[(d^m)^2] = d_0^2 + 2\sigma_X^2 + 2\sigma_Y^2 - 2\mathrm{Cov}(X_1,X_2) - 2\mathrm{Cov}(Y_1,Y_2),$$

which implies $\mathbb{E}[d^m] > d_0$ under realistic assumptions (Ranacher et al., 2015). The overestimation can be written as

$i$0

with $i$1 and $i$2 (Ranacher et al., 2015). In that framework, $i$3 is interpreted as a quality measure for movement data recorded with a GPS: strong spatio-temporal autocorrelation means that consecutive errors are similar and partly cancel in derived quantities such as distance, speed, and direction (Ranacher et al., 2015).

Classical mitigation strategies remain important. Differential GPS uses a reference station at a known position to compute corrections, and carrier phase GPS can achieve centimeter- or even millimeter-level accuracy in survey-grade equipment (Ahamed, 2010). This suggests that GPS accuracy is not a single scalar property of the system but an interaction among signal environment, sampling regime, measurement model, and correction architecture.

3. Filtering, fusion, and estimation architectures

A large fraction of recent GPS research concerns estimation rather than raw measurement. Even simple filters can materially reduce apparent noise. In a real-time tracking tool built from a SIM908 shield and Arduino card, positional error from GPS satellites was processed with Kalman and Average filters; in clear-weather experiments over 30 sequential readings, the best raw error margin was 9.39 meters, the Kalman filter reduced the minimum error to 3.47 meters with an improvement rate of 63.04%, and the Average filter reduced it to 4.18 meters with an improvement rate of 55.48% (Güzel et al., 2018). The simplified Kalman update used there was

$i$4

with $i$5 (Güzel et al., 2018).

For autonomous vehicles, fusion with inertial sensing has become standard because GPS provides absolute positioning with global coverage, whereas IMUs provide relative motion information without reliance on external signals but suffer from drift (Alaba, 2024). A GPS-IMU fusion architecture based on the Unscented Kalman Filter reported that, on the KITTI GNSS and IMU datasets, RMSE decreased from 13.214, 13.284, and 13.363 to 4.271, 5.275, and 0.224 for the $i$6-, $i$7-, and $i$8-axes, respectively (Alaba, 2024). The state evolution was written as

$i$9

with GNSS measurements modeled as $$\text{Distance}_i = c \cdot (t_{\mathrm{Rx}} - t_{\mathrm{Tx}_i} + \delta t),$$0 (Alaba, 2024).

A more tightly coupled ground-robot formulation is GPS-aided visual-wheel odometry, which fuses camera, wheel encoder, and GPS measurements through a Multi-State Constraint Kalman Filter (Song et al., 2023). That work estimates the extrinsic rotation parameter between the GPS global coordinate frame and the VWO reference frame online, gives an observability analysis showing that the yaw-only parameterization is observable for ground vehicles, and reports better accuracy than GPS through fusion in large-scale urban driving scenarios (Song et al., 2023). It also presents the theoretical finding that the variance of an unobservable state could converge to zero for a specific Kalman filter system (Song et al., 2023). A plausible implication is that covariance behavior alone can be misleading as a proxy for identifiability in tightly coupled GPS fusion systems.

4. Operational systems, timing, and GPS data engineering

GPS is also an operational software and embedded-systems domain. A prototype integrating GPS, GSM, and cellular phone communication uses a client-server model in which a GPS receiver, GSM modem, microcontroller, and C/C++ software acquire location and speed information and return it through SMS commands such as “SPEED” or “LOC” (Sahoo et al., 2013). A related real-time geolocation tool built from SIM908 and Arduino transmits data via GSM/GPRS to a web server and visualizes current location on a web page with Google Maps integration (Güzel et al., 2018).

In measurement management, one software implementation for integrated GPS/Loran receivers is organized into recording, classification, and conversion modules, each running automatically without user intervention (Kim et al., 2020). The recording module captures raw streaming text losslessly, the classification module separates NMEA 0183 and proprietary messages by heading, and the conversion module parses, timestamp-sorts, and outputs analysis-ready structures, verified on GPS and Loran measurements collected over 24 hours (Kim et al., 2020). GPS measurements are explicitly used there as ground truth data for performance analysis of navigation methods (Kim et al., 2020).

At larger scales, GPS data engineering becomes a performance problem. For GPS trajectory rasterization, spatial join workflows based on QGIS and PostGIS+QGIS were compared with custom Python implementations, including a parallelized variant (Gengec et al., 2024). QGIS and PostGIS+QGIS showed relatively low performance with respect to the custom method using total processing time, the PostGIS+QGIS method achieved the best results for spatial join alone, while the Python methods scaled directly with GPS point count rather than test-area size and remained feasible for very large grids where the GIS-based methods crashed (Gengec et al., 2024). Public-transit processing raises analogous issues. The gps2gtfs Python package converts raw GPS trajectory data of public transit vehicles into GTFS format through preprocessing, trip extraction, stop matching, feature extraction, and export, relying on geo-buffer mapping, parallel processing, and filtering to address high volume, discontinuities, and localization errors (Ratneswaran et al., 2024).

GPS is equally central to time transfer. One low-power receiver architecture introduces an Instant-On GPS receiver system implemented on an ARM based FPGA board and operated under repeated sleep mode (Yoon et al., 2015). By storing navigation state and using a high resolution RTC to estimate frame sync timing without waiting for the current frame sync preamble, the receiver reports a first navigation solution within about 1 second after wake-up, whereas Hot Start typically required 1.2–6 seconds; for a 15-minute sleep and 2–3 second active cycle, estimated power consumption was 1/300th of a continuously running GPS receiver (Yoon et al., 2015).

5. Security, integrity, and resilience to outages

GPS is vulnerable to both unintentional interference and deliberate attack. Work on autonomous vehicles emphasizes that spoofing and jamming pose risks such as navigation errors and system failures (Abrar et al., 2024). In software-defined receiver research, SPREE extends a conventional receiver with auxiliary peak tracking and navigation message inspection in order to detect spoofing attacks, including seamless takeover attacks (Ranganathan et al., 2016). The key empirical bound is that auxiliary peaks separated by more than 500 ns can be detected, which constrains even a strong attacker from spoofing the receiver to a location not more than 1 km away from its true location; the average maximum location offset reported was 455 m (Ranganathan et al., 2016). This directly contradicts the common assumption that a receiver either fully trusts GPS or must abandon it entirely under spoofing pressure.

Vehicle-level anomaly detection takes a different route. GPS-IDS integrates a GPS navigation model into the dynamic bicycle model, extracts temporal features from GPS and vehicle behavior, and applies machine learning to classify normal and abnormal navigation behaviors (Abrar et al., 2024). On the AV-GPS-Dataset, MLP, RF, and XGB achieved F1 scores over 90%, MLP reached an average F1 score of 94.4%, and attacks were detected about 10–16 seconds after onset, compared with about 23 seconds for EKF; false positive and false negative rates were reported below 0.2% at optimal settings (Abrar et al., 2024).

Resilience also includes graceful degradation when signals disappear. For satellite vehicles, especially in Low Earth Orbit, CSAC drift modeling considering GPS signal quality uses the number of visible satellites and TDOP as weights in a regression model so that a stand-alone Chip-Scale Atomic Clock can maintain its error less than 4 microseconds even when no GPS signals are received for 12 hours (Park et al., 2024). The weighted TDOP model achieved RMSE values of 3.548, 0.775, and 1.930 microseconds across the three reported scenarios, improving on unweighted linear models (Park et al., 2024). This suggests that GPS integrity is inseparable from timekeeping architecture: satellite geometry and visibility affect not only position error but also the quality of holdover models during denial.

6. Augmentation in shadowed environments and nontraditional uses

A major research direction is the extension of GPS into environments where conventional reception is unreliable. In urban visual refinement, a retrieval-based method uses camera images, Google Street View, and sensor parameters to confine search to areas of 0.01 sq. Km with experimental accuracy less than 10 meters, compared with search areas of 1–2 sq. Km and typical accuracy of 25–100 meters for more regular approaches (Salarian et al., 2015). The method leverages approximate query position together with GPS SOS and camera orientation, and it also reports that retrieval-based techniques are less accurate in sparse open areas compared with purely GPS measurement (Salarian et al., 2015).

A hardware augmentation route is GPSMirror, described as the first GPS-strengthening system that works for unmodified smartphones with the assistance of newly-designed GPS backscatter tags (Dong et al., 2023). The tags operate at microwatt-level power consumption, provide coverage up to 27.7 m, and, in reported experiments, GPSMirror achieved median positioning accuracy of 3.7 m indoors and 4.6 m in urban canyon environments (Dong et al., 2023). The associated positioning algorithm introduces virtual satellites through the extra propagation path created by the tag and solves the resulting equations using weighted least squares (Dong et al., 2023).

Infrastructure-side multimodal denoising pushes further. A vision-aided framework combining RGB cameras with mmWave and sub-THz basestation signals performs site-specific GPS characterization and GPS position de-noising, and in a realistic V2I scenario it reports consistent reduction of mean localization error to sub-meter levels across all site regions in both vehicle directions (Charan et al., 2024). That work uses grid-based characterization, YOLOv3-based object detection, beam-aided transmitter identification, and either a lookup table or a feed-forward neural network to map visual-spatial cues to denoised geographic coordinates (Charan et al., 2024).

GPS has also appeared in non-navigation fundamental physics. In one proposal, a GPS photon traversing Earth’s magnetic field may be transformed into an axion, with the relevant small-mixing conversion probability scaling as $$\text{Distance}_i = c \cdot (t_{\mathrm{Rx}} - t_{\mathrm{Tx}_i} + \delta t),$$1 (Nicolaidis, 2017). The proposal exploits the fact that Earth’s geomagnetic field is weak but the magnetized length $$\text{Distance}_i = c \cdot (t_{\mathrm{Rx}} - t_{\mathrm{Tx}_i} + \delta t),$$2 between satellites is very large, yielding a $$\text{Distance}_i = c \cdot (t_{\mathrm{Rx}} - t_{\mathrm{Tx}_i} + \delta t),$$3 scale larger than existing or proposed terrestrial reaches, and it considers both signal dimming and a “light shining through the Earth” configuration (Nicolaidis, 2017). This suggests that GPS is not merely a navigation utility but also a distributed experimental platform whose precisely timed, globally propagated signals can support wider scientific inquiry.