PaperlayerView exact revision on arXiv

Source-linked AI summary

GNSS Anomaly Detection in Urban Canyons: A Comprehensive Survey of Multimodal Visibility Inference

Abdelgabar Ahmed, Tarig Ballal, Nour Kouzayha, Tareq Y. Al-Naffouri

arXiv:2610.34468v1eess.SP

TL;DR

Urban buildings and infrastructure make multipath and NLOS reception major challenges for reliable GNSS positioning. This survey synthesizes receiver-side, map-based, and auxiliary-sensor evidence for reception-state inference and visibility determination. It concludes that no method family is universally dominant and identifies calibrated measurement reliability as an open challenge.

  • Problem

    Urban multipath and NLOS effects are localized and tied to immediate physical surroundings, complicating reliable positioning in urban canyons.

    Source: §I · p. 1, §II · p. 6

  • Method

    The survey organizes reception-state inference into signal-domain methods, 3D map-aided visibility determination, and auxiliary-sensor spatial screening.

    Source: §VII · p. 26

  • Results

    No method family is universally dominant; deployment trade-offs depend on platform constraints, receiver access, map availability, and integrity requirements.

    Source: §VI · p. 25, §VII · p. 26

  • Takeaways & Limitations

    Higher-integrity applications can use multimodal fusion for complementary or redundant evidence, while accounting for its calibration, synchronization, uncertainty, failure-detection, and resource requirements.

    Source: §VI · p. 25

    Hard LOS/NLOS or multipath decisions provide limited information when propagation states overlap or observations are ambiguous, motivating calibrated measurement-level reliability.

    Source: §VII · p. 26

Abstract

from arXiv · show

Reliable GNSS positioning in dense urban environments depends on assessing satellite visibility and identifying measurements degraded by non-line-of-sight (NLOS) reception and multipath. This survey organizes reception-state inference into three complementary evidence families: receiver-side signals, 3D maps and environmental models, and auxiliary sensors. Receiver-side approaches span antenna/electromagnetic, tracking/baseband, and observable-domain evidence, interpreted through rule-based, statistical, machine-learning, and deep-learning methods. Map-based approaches are examined by reviewing environmental representations and methods such as geometric screening, shadow matching, 3D ray tracing, and neural implicit visibility prediction. IMU, camera, and LiDAR measurements provide complementary kinematic, visual, and geometric evidence for online screening. Across these families, we compare accuracy, interpretability, hardware and calibration requirements, map dependence, computational cost, and suitability for real-time deployment. We identify open challenges in dataset quality and labeling, domain generalization, map freshness, and resource-efficient implementation. The resulting synthesis connects signal processing, environmental modeling, and sensor fusion to guide the design of robust urban GNSS measurement-screening systems.

I. INTRODUCTION

Urban buildings and infrastructure make multipath and NLOS reception major threats to GNSS positioning, with effects that depend on the local propagation environment. The survey treats reception state as a latent state inferred from receiver signals, maps, and auxiliary sensors, distinguishing tracked anomalous signals from complete signal unavailability.

A. Impact of Multipath and NLOS on GNSS Accuracy

  • Multipath combines direct and reflected signals, while NLOS tracks an indirect component after the direct path is blocked; without a trackable indirect component, the satellite is unavailable.These distinctions define the tracked reception states used in the survey.Source: A. Impact of Multipath and NLOS on… · pp. 2–3
  • Multipath distorts the code correlation response and shifts the DLL tracking point, producing a pseudorange bias that can vary in either direction as geometry changes.The shift depends on reflected-signal delay, amplitude, and carrier phase, and can change as the receiver or reflectors move.Source: A. Impact of Multipath and NLOS on… · p. 2
  • Multipath also perturbs carrier phase through direct–reflected signal superposition, and evolving biases can degrade carrier-phase continuity, ambiguity resolution, and high-precision positioning.Both code and carrier errors depend on changing multipath geometry.Source: A. Impact of Multipath and NLOS on… · p. 3
  • Because urban multipath and NLOS depend on the immediate environment and lack a generally applicable deterministic correction, contaminated observations must be assessed before use in navigation.The survey frames reception state as a latent state inferred from receiver measurements, map-based visibility, and complementary spatial or kinematic sensor evidence.Source: A. Impact of Multipath and NLOS on… · p. 3

B. Related Works and Identified Gaps

Existing reviews cover GNSS receiver processing, machine learning, environmental modeling, and sensor integration unevenly, often without a shared reception-state framework. This survey connects those domains and treats datasets and labeling as a distinct methodological concern.

B. Related Works and Identified Gaps

  • Prior reviews span receiver and signal processing, machine learning, 3D mapping, and multisensor integration, but their methodological coverage is uneven.Receiver-focused reviews discuss signal indicators and mitigation; learning-focused reviews cover data-driven inference, while mapping and sensor reviews emphasize their respective domains.Source: B. Related Works and Identified Gaps · pp. 3–4
  • Existing reviews generally do not organize these domains under a shared framework, making cross-domain comparison difficult.The survey considers both prior-review coverage and the historical development of reception-state and visibility methods.Source: B. Related Works and Identified Gaps · pp. 3, 5
  • No unified benchmark compares signal-, map-, camera-, and LiDAR-based methods using common labels, receiver configurations, and urban scenarios.The survey therefore treats datasets, labeling strategies, and evaluation protocols as a distinct methodological component.Source: B. Related Works and Identified Gaps · p. 5
  • The survey unifies receiver evidence, learning-based inference, 3D visibility, and camera/LiDAR/IMU verification under a common per-satellite reception-state framework.It also traces environmental representations from building models and LiDAR or photogrammetry to crowdsourced signal representations and learned implicit models.Source: B. Related Works and Identified Gaps · pp. 4–5

C. Contributions and Survey Structure

The survey organizes GNSS reception-state inference around receiver signals, 3D environmental maps, and auxiliary sensing, then compares their practical trade-offs and identifies open challenges.

C. Contributions and Survey Structure

  • The survey organizes reception-state classification around receiver-side signals, 3D map-based environmental information, and real-time auxiliary sensing.It situates these information sources within the historical evolution of the methods.Source: C. Contributions and Survey Structure · p. 5
  • It reviews signal features separately from the rules, classical machine learning, deep learning, sequence models, and multimodal strategies used to classify them.Source: C. Contributions and Survey Structure · p. 5
  • It surveys map representations from static city models and LiDAR or vision-derived maps to neural implicit and crowdsourced signal-based maps.The review considers detail, freshness, memory, updateability, and real-time query cost.Source: C. Contributions and Survey Structure · p. 5
  • It synthesizes map-based methods including geometric LOS screening, ray tracing, shadow matching, 3DMA ranging, and neural implicit visibility inference.Source: C. Contributions and Survey Structure · pp. 5–6
  • It examines IMU, camera, and LiDAR for real-time reception-state verification before measurement integration.Source: C. Contributions and Survey Structure · p. 6
  • It compares practical trade-offs across method families and identifies challenges in datasets, domain generalization, edge deployment, calibration, and trustworthy learning-based detection.Source: C. Contributions and Survey Structure · p. 6

II. GNSS SIGNAL FEATURES AND OBSERVABLES

Receiver-side GNSS evidence reveals degradation through physical wavefield, correlation-response, and observable-level signatures. The section reviews these feature domains and their detection mechanisms, hardware requirements, and practical constraints.

II. GNSS SIGNAL FEATURES AND OBSERVABLES

  • Multipath and NLOS are localized, environment-dependent effects whose structural reflections imprint distinctive physical and statistical signatures on GNSS waveforms.The section therefore focuses on direct signal-based detection rather than treating degradation as generic white noise.Source: §II · p. 6

A. Antenna / EM Level

1) Direction of Arrival (DoA):

  • DoA methods compare incident-wave direction with the expected LOS direction derived from satellite geometry, requiring array-frame attitude knowledge and accurate calibration.Azimuth can be particularly informative in street canyons dominated by vertical facades, while elevation complements other reflection geometries.Source: 1) Direction of Arrival (DoA) · p. 7

2) Polarization Diversity:

  • Dual-polarized measurements use RHCP–LHCP power imbalance or C/N0 differences to identify multipath and NLOS reception.Reflections and diffraction can alter the transmitted RHCP signal’s polarization, often introducing stronger LHCP or elliptical components.Source: 2) Polarization Diversity · p. 7

3) Spatial Correlation:

  • Spatial-correlation methods detect multipath through covariance departures and eigenvalue dispersion, without necessarily estimating arrival direction.Array processing can also resolve and suppress reflected components, but requires calibration, synchronization, stable geometry, and often more power and computation than mass-market devices allow.Source: 3) Spatial Correlation · p. 8

B. Tracking Level

Tracking-level methods detect propagation distortion from correlator behavior, while observable-level methods use accessible measurements such as C/N0 to assess signal quality. Each provides useful evidence but has receiver-specific constraints and can be ambiguous in urban conditions.

B. Tracking Level

  • In multipath, distorted correlation peaks can unbalance early and late correlator outputs; the normalized early–late slope is expected to be zero for ideal LOS and departs from zero under distortion.Dynamic confidence thresholds can account for changing C/N0-dependent variance; a ±3-standard-deviation envelope corresponds to a nominal 0.0027 per-trial false-alarm probability under a Gaussian assumption.Source: B. Tracking Level · pp. 8–9
  • Multi-correlator, area-based, and multi-frequency metrics capture ACF distortion, while vector tracking, parametric path estimation, and learned correlation representations offer complementary detection strategies.MEDLL estimates path delays and powers; correlation-domain methods require tracking outputs and may be computationally costly, while short-delay multipath can evade single-frequency metrics.Source: B. Tracking Level · p. 9

C. Observables Level

  • Observable-domain features use widely accessible measurements, and C/N0 and its windowed variability are common indicators of signal quality and fading.Urban LOS and NLOS C/N0 distributions may overlap, and absolute statistics vary with receiver hardware and environment, so C/N0 is commonly combined with other observables.Source: C. Observables Level · p. 9, 1) Carrier-to-Noise Ratio · p. 10

2) Pseudorange Residuals and Double-Differencing:

Pseudorange residuals and double differences measure range inconsistency, while Code-Minus-Carrier divergence provides an indicator of code multipath. Their usefulness depends on measurement conditions, especially in urban environments.

2) Pseudorange Residuals and Double-Differencing:

  • Pseudorange residuals capture mismatch between measured and predicted range and can flag NLOS or strongly multipath-contaminated measurements, –.Double differences suppress common-mode clock and atmospheric errors over short baselines, making rover-specific multipath and NLOS effects more visible,.Source: 2) Pseudorange Residuals and… · p. 10
  • Residual-based screening weakens when many satellites are affected, because NLOS biases can be absorbed into position and clock estimates; it also depends on sufficient LOS redundancy and suitable reference-satellite selection.In deep urban canyons, signal strength and elevation alone may not reliably identify a suitable reference satellite.Source: 2) Pseudorange Residuals and… · p. 10

3) Code-Minus-Carrier Divergence:

  • Code-Minus-Carrier divergence compares code pseudorange with carrier-phase range and indicates code distortion, while time-differenced CMC cancels constant carrier-phase ambiguity.Absolute CMC depends on continuous phase tracking; cycle slips and urban loss-of-lock events fragment the time series and can mask multipath features,,.Source: 3) Code-Minus-Carrier Divergence · pp. 10–11

4) Pseudorange–Doppler Consistency:

Observable-domain screening draws on Doppler consistency, satellite geometry, temporal and multivariate features, and cross-antenna spatial constraints. These signals add complementary evidence, with reliability and implementation trade-offs that vary by method.

4) Pseudorange–Doppler Consistency:

  • Pseudorange rate should align with Doppler-derived range rate under LOS, while multipath or NLOS can perturb the code- and frequency-tracking estimates differently,.The metric depends on time alignment, short differencing intervals, and stable tracking; receiver dynamics, noise, and loss of lock can also raise inconsistency,.Source: 4) Pseudorange–Doppler Consistency · p. 11

5) Geometric and Satellite Ephemeris Indicators:

  • Elevation and azimuth provide spatial context, but elevation alone cannot determine reception state in urban canyons, where high satellites may be blocked and low satellites may remain LOS, –,.Evaluated urban scenarios found C/N0-based weighting outperformed purely elevation-based weighting,.Source: 5) Geometric and Satellite Ephemeris… · p. 11

6) Temporal and Multivariate Sequence Features:

  • Temporal windows reveal persistent attenuation and fluctuations that single-epoch measurements may miss, while multivariate sequence and graph models capture temporal and cross-satellite relationships –,, –.Longer windows add historical context but increase latency and computational demands; shorter windows respond faster but are more vulnerable to tracking jitter, and complex models may be opaque –.Source: 6) Temporal and Multivariate Sequence… · pp. 11–12

7) Multi-Antenna and Spatial Observables:

  • Multi-antenna methods use carrier-phase consistency with array-baseline geometry to identify contaminated observations; some jointly estimate calibration and multipath, while others exclude geometric violations –.These methods require multi-antenna hardware and greater system complexity, but provide robustness when many urban observations are corrupted,,.Source: 7) Multi-Antenna and Spatial Observables · p. 12

A. Algorithmic Approaches

The surveyed approaches range from fixed statistical rules to flexible learned classifiers, with generalization and ground-truth quality remaining central concerns. The section also connects dataset labeling resources with the online use of 3D environmental models.

A. Algorithmic Approaches

  • Thresholds are transparent but depend on valid assumed statistics, while learned classifiers capture nonlinear relationships yet may lose performance across receivers, trajectories, or environments,,,,.Sequence models, graph methods, domain adaptation, and branched or ensemble architectures address temporal, satellite-relational, transfer, and missing-measurement challenges,,,,,.Source: A. Algorithmic Approaches · pp. 12, 14

B. Labeling Strategies and Multimodal Datasets

  • Navigation-centric datasets use perception sensors and reference poses for independent visibility labels, but label accuracy depends on pose synchronization, calibration, model completeness, and dynamic or boundary occlusions, –.GNSS-centric collections support receiver diversity, while prepared benchmarks and supplied indicators improve reproducibility; label fidelity and independence still matter, –.Source: B. Labeling Strategies and Multimodal… · p. 14
  • Raw-IF datasets expose tracking-level signal features, but generally lack native per-satellite reception-state truth and require geometry-, imagery-, or path-based annotation –.Controlled synthetic datasets provide known propagation states for ablations, but their simulated channels cannot reproduce the full diversity of real urban channels.Source: B. Labeling Strategies and Multimodal… · pp. 14–15
  • A reproducible labeling pipeline can derive geometric LOS/NLOS labels, cross-check them with independent imagery and corrected residuals, and exclude ambiguous epochs,.Static signal thresholds such as 35 dB-Hz may misclassify strong reflections or mildly attenuated LOS signals.Source: B. Labeling Strategies and Multimodal… · p. 15

IV. 3D MAP-BASED VISIBILITY DETERMINATION

  • 3D environmental models support deterministic visibility prediction and evaluation of reflected paths, enabling exclusion, down-weighting, or geometric correction of degraded measurements,,,.Map-based methods include geometric screening, shadow matching, ray tracing, and neural occlusion models, with computational trade-offs and potential fusion with signal-based methods.Source: §IV · p. 15

A. 3D Mapping Modalities for GNSS Visibility Analysis

The survey reviews environmental representations for predicting GNSS visibility, from explicit geometry and image-derived sky masks to learned continuous scene fields. Their fidelity, computational cost, and information content vary with representation and intended propagation model.

1) Explicit Geometric Maps:

  • Explicit building models support interpretable visibility queries, with LoD1 geometry suited to scalable LOS/NLOS screening and richer surfaces needed for explicit multipath simulation.LiDAR point clouds and photogrammetric meshes add local detail but increase acquisition, storage, preprocessing, and maintenance requirements, creating a scalability–fidelity trade-off.Source: 1) Explicit Geometric Maps · pp. 15, 17

2) Image-Derived Sky Representations:

  • Georeferenced panoramas can be segmented into hemispherical sky masks and queried with satellite azimuth and elevation to label candidate LOS or direct-path blockage.These labels are geometric predictions, not RF ground truth; sky-visible signals may still contain multipath, and blocked directions need not have a tracked reflection.Source: 2) Image-Derived Sky Representations · p. 17

3) Neural Scene Representations:

  • Neural scene representations encode urban structure as continuous learned fields, enabling GNSS visibility queries without requiring an explicit building mesh.Prediction reliability depends on training-image coverage and quality; learned fields add training cost and reduce geometric transparency.Source: 3) Neural Scene Representations · p. 17

4) Crowdsourced Signal-Based Mapping:

The section connects environmental mapping with reception-state inference, describing both GNSS-derived map construction and map-based visibility or propagation prediction. It contrasts efficient geometric screening with higher-fidelity ray tracing and notes their respective dependencies and limitations.

4) Crowdsourced Signal-Based Mapping:

  • Repeated GNSS reception-state observations can reveal building boundaries or sky-obstruction profiles and support map construction, height estimation, or monitoring of model changes.Crowdsourced methods depend on measurement density and diversity, accurate inferred reception states, and sufficient satellite-view geometry; they complement rather than guarantee replacement of surveyed maps.Source: 4) Crowdsourced Signal-Based Mapping · pp. 17–18

B. Map-Based Visibility Prediction and Detection

  • Map-derived predictors estimate expected visibility or propagation from receiver position, satellite geometry, and an environmental model; verifiers compare those priors with GNSS observations.Disagreement may reflect anomalous reception, map incompleteness, dynamic occlusion, or pose uncertainty.Source: B. Map-Based Visibility Prediction and… · p. 18

1) Geometric LOS Screening:

  • A geometric LOS test predicts direct-path visibility when the satellite–receiver ray clears building geometry and blockage when it intersects the model.Precomputed skylines and texture-based ray marching can accelerate repeated queries, but predictions remain sensitive to receiver position and map completeness.Source: 1) Geometric LOS Screening · p. 19

2) Explicit 3D Ray-Tracing:

  • Ray tracing models direct and indirect paths to predict LOS, NLOS, or multipath and can provide excess delay, Doppler shift, phase rotation, and attenuation priors.GPU-based pixel-shader ray tracing can reduce computational growth for multiple reflections, while prediction fidelity remains dependent on map geometry, surface properties, and modeled propagation.Source: 2) Explicit 3D Ray-Tracing · pp. 19–20

3) Shadow Matching:

Shadow matching and 3DMA ranging use consistency between map-derived expectations and GNSS evidence, while neural fields provide continuous learned visibility priors. The methods accommodate different levels of propagation detail and uncertainty.

3) Shadow Matching:

  • Shadow matching scores candidate positions by comparing map-predicted satellite visibility with receiver-derived evidence, with disagreement indicating inconsistency but not uniquely NLOS reception.Likelihood, Bayesian, and set-valued approaches address visibility uncertainty and boundary sensitivity.Source: 3) Shadow Matching · p. 20

4) 3DMA Ranging:

  • 3DMA ranging compares map-conditioned propagation expectations with pseudoranges, ranging from explicit reflected-path delay prediction to statistical LOS/NLOS likelihoods.The resulting range-consistency evidence is distinct from downstream measurement correction, weighting, or exclusion.Source: 4) 3DMA Ranging · p. 20

5) Neural-Field Visibility Prediction:

  • Neural-field visibility prediction integrates learned density along a satellite-directed ray to produce a soft visibility score rather than a hard geometric LOS/NLOS decision.Continuous querying and richer scene representation come with dependence on training coverage, offline learning, and reduced geometric transparency.Source: 5) Neural-Field Visibility Prediction · p. 21

V. AUXILIARY SENSORS FOR URBAN NLOS AND MULTIPATH DETECTION

The survey frames auxiliary sensors as independent evidence for checking GNSS measurements against receiver motion or the observed environment. It reviews inertial kinematic checks and camera-based direct-path screening, each with a distinct role.

V. AUXILIARY SENSORS FOR URBAN NLOS AND MULTIPATH DETECTION

  • Auxiliary sensors provide time-varying kinematic or geometric evidence, structurally decoupled from the GNSS RF chain, for reception-state verification before measurements enter navigation.The survey focuses on their verification role rather than their conventional use in coordinate estimation.Source: §V · pp. 21–22

A. IMU-Aided Kinematic Consistency Checking

  • IMUs act as short-term kinematic consistency checkers, supplying motion and attitude references against which GNSS measurements can be tested.Reviewed examples use inertial aiding for multiple-fault exclusion, dynamic multipath estimation, and array-phase consistency checks.Source: A. IMU-Aided Kinematic Consistency… · p. 22

B. Camera-Aided NLOS Detection

  • Cameras project satellite directions onto segmented sky views to identify direct-path blockage, while current-epoch onboard imagery can capture transient occluders missed by prior maps.Screening requires calibration, alignment, synchronization, and robust segmentation; visible sky does not rule out multipath.Source: B. Camera-Aided NLOS Detection · pp. 22–23

1) Sky-Mask Construction:

Camera and LiDAR methods turn local scene information into geometric evidence for GNSS visibility and reception-state screening. Cameras offer angular sky/occlusion cues, while LiDAR adds metric depth and surface geometry for blockage and reflected-path inference.

1) Sky-Mask Construction:

  • Camera sky-mask construction separates open sky from surrounding occluders using interpretable, efficient rules, but performance can depend on illumination, appearance, vegetation, and boundary detail.These sensitivities motivated learning-based segmentation.Source: 1) Sky-Mask Construction · p. 23

2) Learning-Based Semantic Sky Segmentation:

  • Learning-based segmentation replaces hand-crafted rules with pixel-wise sky/occluder masks, while lightweight classical processing remains an option when computational efficiency is prioritized.Reviewed approaches range from CNN, fully convolutional, and Swin Transformer models to Otsu thresholding; the resulting mask supports visibility screening regardless of architecture.Source: 2) Learning-Based Semantic Sky… · p. 23

3) Satellite Projection and NLOS Detection:

  • Satellite projection combines ephemerides, approximate receiver state, calibrated camera geometry, and attitude to test whether a satellite maps to sky or an occluding region.Onboard images can capture transient occluders, but a visually unobstructed direction may still contain multipath; calibration, attitude, synchronization, and segmentation errors can affect classification.Source: 3) Satellite Projection and NLOS… · p. 23

C. LiDAR-Based GNSS Anomaly Detection

  • For reception-state verification, onboard LiDAR supplies viewpoint-specific 3D geometry for checking satellite blockage and supporting hypotheses about reflected NLOS or multipath paths.This differs from using LiDAR primarily for localization or SLAM; inferred reflected paths rely on environmental geometry and, where available, GNSS measurements.Source: C. LiDAR-Based GNSS Anomaly Detection · pp. 23–24

1) Point-Cloud Scene Construction:

  • LiDAR point-cloud scene construction represents nearby structures, but viewpoint limits and sparse or noisy returns can leave urban geometry incomplete.Methods filter, segment, and organize points into surfaces, facades, or local maps; coarse building heights, sliding-window maps, and reconstructed facades address coverage limits.Source: 1) Point-Cloud Scene Construction · p. 24
  • Onboard LiDAR captures current and transient obstacles that static city models may miss, bridging static-map visibility prediction and signal-domain classification.Vehicles and temporary roadside structures can block line of sight or contribute reflected paths even where a pre-built map predicts open sky.Source: 1) Point-Cloud Scene Construction · p. 24

2) Visibility and Reflected-Path Inference:

  • LiDAR visibility testing classifies an obstructed satellite direction as candidate NLOS when a signal is nevertheless tracked, with onboard sensing reflecting current scene conditions.Later methods combine LiDAR visibility with GNSS measurement consistency, including pseudorange innovations.Source: 2) Visibility and Reflected-Path… · p. 24
  • LiDAR surface geometry supports inference of feasible reflectors and reflected paths, while pairing it with a sky-pointing camera combines angular visibility with metric range and depth.Reflected-path inference can also incorporate Doppler-derived direction information; camera and LiDAR limitations differ, including visual-condition sensitivity versus LiDAR range, reflectivity, sparsity, and field-of-view constraints.Source: 2) Visibility and Reflected-Path… · p. 24

VI. COMPARATIVE INSIGHTS AND DESIGN TRADE-OFFS

The survey compares receiver-side, map-based, and auxiliary-sensor methods by their evidence sources and deployment trade-offs, rather than identifying one universally dominant family. It concludes that robust reception-state assessment faces open challenges in transfer, uncertainty, benchmarking, and practical deployment.

VI. COMPARATIVE INSIGHTS AND DESIGN TRADE-OFFS

  • Fig. 11 presents relative trade-offs in implementation burden and dependence on labeled data or environmental priors, not quantitative performance rankings.Reported performance should be interpreted in light of validation setting, receiver hardware, and study assumptions.Source: Figure 11 · p. 25, §VI · p. 25
  • Signal-observable rules and classical learning on engineered features use standard GNSS measurements without exteroceptive infrastructure and suit low-latency screening, but their indicators can overlap across reception states.Their decisions are consequently sensitive to receiver, environment, and operating scenario.Source: §VI · p. 25
  • Correlator-domain learning, antenna arrays, and polarization-aware methods provide evidence more directly tied to propagation mechanisms, but often require internal data or calibrated RF hardware.Some also require software-defined radio architectures or multi-antenna installations.Source: §VI · p. 25
  • Maps and auxiliary sensors add environmental awareness: maps predict visibility, cameras provide real-time occlusion cues, and LiDAR supplies 3D geometry for blockage and reflected-path inference.These sources constrain propagation hypotheses geometrically rather than directly measuring the RF path; combining them with GNSS observables can help distinguish reflected NLOS or multipath.Source: §VI · p. 25
  • No method family is universally dominant; deployment choices depend on platform constraints, receiver access, map availability, and integrity requirements.The survey describes layered architectures that combine evidence streams before unreliable observations enter the navigation solution.Source: §VI · p. 25, §VII · p. 26

VII. CONCLUSION AND OPEN DIRECTIONS

  • Reliable transfer of data-driven detectors across receivers, environments, and operating conditions with little or no relabeling remains an open challenge.The survey also identifies calibrated measurement-level reliability for ambiguous or overlapping states as an unresolved need.Source: §VII · p. 26
  • Jointly modeling satellite and urban-occluder distributions is proposed as a route to predicting LOS availability and positioning localizability without a deterministic 3D map for every environment.This direction combines stochastic satellite-distribution and urban-geometry models.Source: §VII · p. 26
  • Reception-state labels from maps, cameras, LiDAR, ray tracing, or GNSS measurements can disagree, motivating standardized state definitions, labeling uncertainty, and cross-domain evaluation.Richer methods also bring latency and integration constraints, while map-based approaches need ways to detect and update stale environmental information.Source: §VII · p. 26

Related papers

Loading 2610.34468v1…