Source-linked AI summary
Position Location for Futuristic Cellular Communications -- 5G and Beyond
Ojas Kanhere, Theodore S. Rappaport
TL;DR
5G and beyond require precise positioning that remains difficult for GPS and conventional methods in obstructed environments. The article surveys mmWave localization and improvements from NLoS mitigation, machine learning, tracking, data fusion, cooperative localization, and map-based techniques. It reports 23 cm median accuracy after suppressing NLoS multipath and outlines future map-enabled positioning using real-time multipath.
Problem
Existing GPS and localization approaches lose accuracy in obstructed environments, while future applications require precise positioning for devices and objects.
Method
The article surveys mmWave localization and combines geometric, machine-learning, tracking, data-fusion, cooperative, and map-based approaches.
Results
23 cm median localization accuracy was achieved with six 2.4 GHz Wi‑Fi access points by suppressing NLoS multipath and using the LoS path.
Takeaways & Limitations
Future robust positioning is expected to combine machine learning, multiple sensor measurements, cooperative localization, and seamless transitions across wireless technologies.
Abstract
from arXiv · showhide
With vast mmWave spectrum and narrow beam antenna technology, precise position location is now possible in 5G and future mobile communication systems. In this article, we describe how centimeterlevel localization accuracy can be achieved, particularly through the use of map-based techniques. We show how data fusion of parallel information streams, machine learning, and cooperative localization techniques further improve positioning accuracy.
I. INTRODUCTION
Precise positioning is a key 5G and beyond application, but existing GPS, Wi‑Fi, and other approaches face accuracy or environmental limitations. MmWave communication combines wide bandwidth, directional antennas, and shared communication infrastructure to support centimeter-level localization.
- Limitations of Existing Systems: GPS is accurate to about 5 m but becomes unreliable indoors, underground, and in urban canyons because signals are attenuated and reflected.Wireless-assisted GPS and Wi‑Fi hotspot databases improve indoor positioning, but Wi‑Fi estimates typically achieve only tens of meters.
- Limitations of Existing Systems: Future positioning applications require errors below 3 m, tighter than the FCC’s less-than-50-m requirement for 80 percent of E911 callers.
- Alternative Technologies: Vision, ultrasound, UWB, RFID, visible light, and Bluetooth offer alternatives, while cellular-frequency systems remain useful when visibility is hampered.UWB ranging can achieve accuracy on the order of centimeters, whereas camera-based localization can be affected by low visibility.
- Article Focus: MmWave systems use the same infrastructure for communication and positioning, with wide bandwidth, steerable directional MIMO antennas, cooperative localization, machine learning, tracking, and multipath.
II. FUNDAMENTAL LOCALIZATION TECHNIQUES
Fundamental localization estimates a UE position from geometric constraints involving known base stations. AoA uses signal direction, while ToA and TDoA use distance or distance differences derived from propagation time.
- Geometric Localization: Geometry-based localization determines UE position from known BS locations, BS–UE distances, and physical angular orientations.
- Fundamental Techniques: AoA estimates the angle of the strongest received signal, whereas ToA and TDoA estimate distance or distance differences from reference-signal travel times.
- Fundamental Techniques: ToA and TDoA localize the UE where distance circles or distance-difference hyperbolas corresponding to BS measurements intersect.
A. Accurate Localization in 5G Networks with Directional Antenna Arrays and Wide Bandwidths
5G mmWave improves localization through wide bandwidths and narrow, high-gain antenna beams. Electrically large arrays enable precise angle estimates, while fine multipath time resolution supports accurate ranging, subject to propagation and beam-management constraints.
- 5G mmWave Positioning: MmWave 5G NR spans 24.25–52.6 GHz, while IEEE 802.11ad supports 57–71 GHz indoors, enabling highly directional arrays with narrow beamwidths.
- Fundamental Measurements: Localization can use ToA, TDoA, or AoA measurements, represented respectively by intersecting circles, hyperbolas, or angular lines.
- Directional Antenna Arrays: For URAs with half-wavelength spacing, HPBW is approximately (102/N)°, so larger planar arrays produce narrower beams; 8×8 through 64×64 arrays have HPBWs from 12.76° to 1.55°.
- Directional Antenna Arrays: Narrower antenna HPBWs enable more precise AoA estimation, with further signal processing improving angular accuracy.
- Propagation Conditions: Directional antenna gain and smaller 100–200 m cells can compensate for mmWave’s higher path and blockage losses, supporting outdoor localization.
- Wide Bandwidths: Wide mmWave bandwidths provide about 1 ns multipath time resolution, corresponding to 30 cm spatial resolution before additional processing.
B. Performance of Fundamental Localization Techniques in Dense Multipath Environments
Traditional ToA, TDoA, and AoA methods are designed for line-of-sight propagation and can incur large errors in dense multipath and non-line-of-sight environments.
- NLoS Multipath: NLoS multipath arrives from different angles with larger delays, producing positioning errors; traditional indoor methods have reported mean errors of 8–10 m.
C. NLoS Mitigation for Accurate Positioning
NLoS mitigation identifies and excludes unreliable propagation paths using motion-sensitive AoA, distance variance, channel features, and SVM classification. These methods improve localization reliability, with SVM-based classification reducing NLoS identification error to 5%.
- Traditional ToA, TDoA, and AoA methods require NLoS mitigation because reflected paths create larger delays and angular errors.The described mitigation strategy retains LoS signals while discarding signals classified as NLoS.
- A 5° AoA change over a 5 cm UE movement flags an NLoS path for exclusion, enabling 23 cm median accuracy with six WiFi access points.The criterion exploits stable AoA under small LoS movements and variable AoA under NLoS conditions.
- Running variance of BS-UE distance estimates identifies NLoS links when σ2 exceeds a calibrated threshold γ.Distance-estimate variance is higher in NLoS than LoS conditions.
- NLoS links can also be recognized from lower received power, larger RMS delay spread, Rician-K factor, and angular-spread characteristics.Using multiple channel characteristics together improves NLoS classification compared with relying on individual features.
- NLoS identification error fell from 10 percent to 5 percent when an SVM combined multiple channel characteristics.The SVM separates LoS and NLoS links using an optimized hyperplane over channel features.
III. SUB-METER PRECISE POSITION LOCATION
Discarding NLoS signals preserves classical positioning assumptions but wastes multipath energy and requires dense base-station deployment with multiple LoS links.
- Using only LoS signals wastes multipath energy and requires the UE to see two or more BSs, making dense deployment potentially cost-prohibitive.
A. Cooperative Localization
Cooperative localization uses D2D measurements between neighboring UEs and can estimate positions centrally or distribute iterative refinement across the network. Centralized and distributed approaches achieved 2.5 m and 3 m RMS error, respectively, in an indoor deployment.
- 5G D2D protocols let UEs communicate directly and measure range and angular relationships for cooperative localization.UEs are typically closer to one another, increasing the likelihood of LoS links and higher SNR.
- Centralized cooperative localization sends relative UE measurements to a central unit and jointly estimates all UE positions by nonlinear least squares.Optimization methods such as Levenberg-Marquardt can solve the resulting constrained estimation problem.
- Distributed algorithms avoid routing all localization messages to a central server by iteratively refining neighboring UEs’ estimates until agreement.This reduces the centralization bottleneck but is described as less accurate than infrared methods.
- 2.5 m and 3 m RMS error were achieved with centralized and distributed cooperative localization in an indoor 40 m × 20 m environment.The experiment used four BSs with known locations and 13 UEs with unknown locations.
B. Machine Learning for Localization
Machine-learning localization maps measured channel fingerprints to positions using reference-point databases and learned nonlinear functions. Ray tracing, transfer learning, and environment maps are proposed to reduce the burden of collecting and updating dense fingerprints.
- Machine-learning localization matches online RSS, CSI, and AoA measurements against a fingerprinting database collected at known reference points.The database stores each channel-parameter vector with the corresponding reference-point coordinates.
- k-NN estimates position as the weighted average of the k reference points whose channel measurements are most similar to the online measurements.Similarity can be measured with Euclidean or Manhattan distance.
- Neural networks learn a nonlinear transformation from measured channel parameters to user coordinates during offline training.Training adjusts network weights so outputs approximate reference-point positions.
- Map-based localization can exploit real-time multipath, with future UEs generating or loading environmental maps and augmenting vision in low-visibility settings.The article illustrates on-the-fly map generation and seeing through walls using narrow beams and multipath.
- Dense fingerprint databases are time-intensive to create, and localization accuracy typically follows the spacing between reference points.Environmental changes such as new furniture can require database recreation.
- Transfer learning and calibrated ray tracing can reduce data collection and accelerate fingerprint-database updates after minor environmental changes.A neural network can first train on ray-traced synthetic data before refinement with additional measurements.
C. User Tracking and Data Fusion
User tracking smooths position estimates for mobile targets, while KF/EKF fusion combines sensor and channel measurements to correct inertial drift.
- C. User Tracking and Data Fusion: Continuous user tracking smooths sudden apparent position changes caused by positioning errors in mobile UEs.Stationary-target accuracy can instead improve by averaging position estimates to reduce variance.
- C. User Tracking and Data Fusion: IMU-based position estimates drift because a constant acceleration offset produces a quadratic position error over time.
- C. User Tracking and Data Fusion: KF/EKF recursively fuses sensor and channel measurements to update the position estimate and correct inertial drift.The KF models linear processes, while the EKF locally linearizes nonlinear relations using a Taylor expansion.
D. Localization Algorithms Exploiting Multipath
Map-based multipath localization turns reflected and scattered signals into location information by associating them with virtual anchors and enforcing feasible movement.
- D. Localization Algorithms Exploiting Multipath: Map-based multipath processing detects and discards forbidden UE transitions, such as movement through walls or between floors in consecutive time steps.
- D. Localization Algorithms Exploiting Multipath: Virtual anchors model successive BS reflections as line-of-sight anchors, enabling AoA, ToA, or TDoA localization after multipath association.An EKF can use the previously estimated UE location to associate received multipath components with virtual anchors.
- D. Localization Algorithms Exploiting Multipath: Ray tracing can exploit non-line-of-sight multipath for single-shot location estimation using BS AoA, source-ray ToA, and an environmental map.Large mmWave bandwidths resolve more multipath components but make their association with virtual anchors more difficult.
- D. Localization Algorithms Exploiting Multipath: Without a map, UE position, orientation, and single-bounce scatterer locations can be estimated jointly through nonlinear least squares.Particle swarm optimization and the Levenberg-Marquardt algorithm may solve this nonlinear least-squares problem.
IV. CONCLUSION AND FUTURE RESEARCH
The article projects that mmWave systems, real-time maps, and multipath-aware processing can support unprecedented localization accuracy, alongside future fusion, cooperation, and privacy safeguards.
- IV. CONCLUSION AND FUTURE RESEARCH: Wide mmWave bandwidths can support unprecedented localization accuracy, but narrow antenna beams require smart beam management and multipath exploration.The article identifies joint communication and localization as requiring further study.
- IV. CONCLUSION AND FUTURE RESEARCH: Future robust positioning is predicted to combine machine learning, multi-sensor data fusion, and cooperative localization.
- IV. CONCLUSION AND FUTURE RESEARCH: Centimeter-level localization will raise privacy concerns, requiring tracking opt-out mechanisms and protection of stored user-location data.
- IV. CONCLUSION AND FUTURE RESEARCH: Localization systems must resist malicious interference, including replicated cellular reference signals used to obtain unauthorized location information.
- IV. CONCLUSION AND FUTURE RESEARCH: Real-time mapping and ray tracing could let future phones exploit multipath to measure environmental features and view around or behind obstructions.Phones may generate or download maps and act like radars measuring walls, doors, and other prominent features.