PaperlayerView exact revision on arXiv

Source-linked AI summary

Joint Device Positioning and Clock Synchronization in 5G Ultra-Dense Networks

Mike Koivisto, Mário Costa, Janis Werner, Kari Heiska, Jukka Talvitie, Kari Leppänen, Visa Koivunen, Mikko Valkama

arXiv:1604.03322v3cs.IT

TL;DR

Accurate, low-power device positioning and tracking in 5G networks requires network-side processing that handles both propagation measurements and clock offsets. The paper proposes cascaded EKFs using uplink DoA/ToA estimates, evaluates them in realistic ray-traced scenarios, and finds sub-meter positioning feasible even with time-varying offsets and potentially unsynchronized network elements.

  • Problem

    The paper addresses accurate device positioning and tracking in 5G radio access networks while limiting user-device computing and energy demands and accounting for relative clock offsets.

  • Method

    Cascaded EKFs estimate and track DoA and ToA at access nodes, fuse measurements into user-node position estimates, and jointly estimate clock offsets on the network side.

  • Results

    Sub-meter positioning and tracking of moving devices is technically feasible at sub-6GHz frequencies despite realistic time-varying clock offsets and potentially unsynchronized network elements.

  • Takeaways & Limitations

    The proposed scheme can provide accurate positioning together with user and network clock-offset estimates while keeping user-device processing and power requirements low.

Abstract

from arXiv · show

In this article, we address the prospects and key enabling technologies for highly efficient and accurate device positioning and tracking in 5G radio access networks. Building on the premises of ultra-dense networks as well as on the adoption of multicarrier waveforms and antenna arrays in the access nodes (ANs), we first formulate extended Kalman filter (EKF)-based solutions for computationally efficient joint estimation and tracking of the time of arrival (ToA) and direction of arrival (DoA) of the user nodes (UNs) using uplink reference signals. Then, a second EKF stage is proposed in order to fuse the individual DoA/ToA estimates from one or several ANs into a UN position estimate. Since all the processing takes place at the network side, the computing complexity and energy consumption at the UN side are kept to a minimum. The cascaded EKFs proposed in this article also take into account the unavoidable relative clock offsets between UNs and ANs, such that reliable clock synchronization of the access-link is obtained as a valuable by-product. The proposed cascaded EKF scheme is then revised and extended to more general and challenging scenarios where not only the UNs have clock offsets against the network time, but also the ANs themselves are not mutually synchronized in time. Finally, comprehensive performance evaluations of the proposed solutions on a realistic 5G network setup, building on the METIS project based outdoor Madrid map model together with complete ray tracing based propagation modeling, are provided. The obtained results clearly demonstrate that by using the developed methods, sub-meter scale positioning and tracking accuracy of moving devices is indeed technically feasible in future 5G radio access networks operating at sub-6GHz frequencies, despite the realistic assumptions related to clock offsets and potentially even under unsynchronized network elements.

I. INTRODUCTION

The paper develops network-centric cascaded EKFs for accurate device positioning, tracking, and clock synchronization in 5G ultra-dense networks. It combines uplink reference signals, multiantenna access nodes, and multicarrier waveforms, with evaluations indicating feasibility under realistic clock-offset conditions.

  • Motivation: 5G networks are expected to support highly accurate device positioning beyond existing approaches that typically provide positioning accuracy of a few tens of meters.The motivation includes enhanced observed time difference, uplink-time difference of arrival, observed time difference of arrival, GPS, and WiFi fingerprinting approaches.
  • System benefits: All essential processing is performed on the network side, keeping computing requirements and power consumption at user devices low.The framework targets connected-vehicle scenarios and supports applications including location-aware communications, IoT positioning, proactive radio resource management, and mobility management.
  • System setting: 5G ultra-dense networks support network-based positioning because dense access nodes increase the likelihood that user nodes have line-of-sight links to multiple access nodes.The considered network has access nodes deployed below rooftops with a maximum inter-site distance of around 50 m.
  • Proposed approach: The proposed cascaded EKFs first estimate and track direction of arrival and time of arrival at individual access nodes, then fuse these estimates into a user-node position.The first stage operates at individual access nodes, while the second stage combines measurements from multiple access nodes.
  • Synchronization: The EKF framework accounts for relative clock offsets between user nodes and receiving access nodes, producing clock-offset estimates as a by-product.The approach is designed for sequential estimation and tracking in mobile scenarios rather than batch localization alone.
  • Synchronization: The solution is extended to scenarios where access nodes are mutually unsynchronized in addition to user nodes having offsets from network time.The paper evaluates these solutions in a realistic 5G setup based on the METIS Madrid map model and ray-tracing propagation modeling.

B. Channel Model for DoA/ToA Estimation and Tracking

The channel model supports EKF-based DoA and ToA estimation from multicarrier, multiantenna uplink signals. The paper uses a computationally efficient single-path EKF model while evaluating performance with detailed ray-tracing-based multipath channels.

  • Channel modeling: The numerical evaluations use detailed ray-tracing channel modeling, while the proposed EKF fits a single-path model to the estimated multipath channel.The single-path approximation is motivated by computational efficiency and the dominant-path characteristics of the propagation environment.
  • Channel response: The EKFs exploit uplink SIMO multicarrier-multiantenna channel response estimates obtained from uplink reference signals.The channel response includes the multicarrier and multiantenna dimensions used for DoA and ToA estimation.
  • Channel model: The channel response separates polarimetric array response, path weights, and additive complex-circular white-Gaussian noise.The polarimetric response is denoted B(ϑ, ϕ, τ), path weights by γ, and noise variance by σ^2.
  • Channel response: The multichannel vector dimension is M = M_fM_AN, where M_f is the number of subcarriers and M_AN is the number of antenna elements.This dimension links the frequency and spatial observations used by the channel model.
  • Array model: The array model supports planar or conformal arrays, including nonuniform element placement, through an effective aperture distribution function representation.The representation uses horizontal and vertical array responses and spatial harmonics.
  • Assumptions: The model assumes identical RF chains and frequency-flat angular response for clarity, with extensions to nonidentical chains and frequency-dependent responses described as computationally more demanding.Timing and frequency synchronization needed to avoid inter-carrier and inter-symbol interference is assumed, while more rigorous ICI treatment is left for future work.
  • Tracking state: The tracking state includes azimuth and co-elevation DoAs, ToA, and their rates of change, although only azimuth DoAs are fused for 2D positioning.In OFDM, the modeled τ represents the ToA difference relative to the FFT-window start before that offset is added back.

C. Clock Models

The paper models time-varying clock offsets recursively through clock skew and uses first-order autoregressive clock-skew models within EKF-based tracking. The proposed EKF framework supports DoA/ToA tracking while allowing extensions to higher-order clock models for low-grade oscillators.

  • Clock-offset evolution: Clock offset ρ is modeled as time-varying because of clock-oscillator imperfections.The recursive model updates the offset using the clock skew over each measurement interval.
  • Clock-skew evolution: The clock skew α[n] may be treated as constant on average, although measurements also support time-dependent skew.The paper adopts a first-order autoregressive model for clock skew based on reported performance gains over constant-skew modeling.
  • Model extension: Higher-order autoregressive clock models can be incorporated through state augmentation, particularly for low-grade clock oscillators.The paper notes that poorer frequency stability motivates considering clock-skew models beyond first order.
  • Information-form EKF: The information-form EKF derives its observed Fisher information matrix and score function from a concentrated likelihood after eliminating linear path-weight parameters.This reduces processing to the DoA and ToA parameters rather than explicitly tracking path weights.
  • DoA/ToA tracking: The DoA/ToA tracking EKF uses a continuous white noise acceleration model for arrival angles and time of arrival.Its six-dimensional state includes DoA and ToA values together with their rates of change, propagated using discretized transition and covariance matrices.

B. Positioning and Synchronization EKF at Central Processing Unit

The central-processing-unit EKF fuses DoA and ToA estimates from line-of-sight access nodes to jointly track user position, velocity, clock offset, and clock skew. A two-phase initialization improves coarse position estimates before full DoA/ToA positioning and synchronization, while the formulation assumes nearly constant velocity with small random perturbations.

  • Cascaded EKF: The cascaded EKF tracks DoAs and ToAs at individual access nodes, then jointly estimates user position and clock offset at the network side.For 2D positioning, the second stage fuses azimuth DoAs and ToAs from line-of-sight access nodes.
  • State and measurement model: The joint Pos&Clock EKF estimates user position, velocity, clock offset, and clock skew from DoA and ToA measurements.The measurement model is nonlinear, while the state transition is linear under the assumed motion and clock models.
  • State evolution: The state evolves under a continuous white noise acceleration model, with separate process-noise parameters for velocity and clock-skew dynamics.The state-transition matrix combines constant-movement dynamics with the clock-evolution model.
  • Outputs: At each time step, the EKF returns a two-dimensional user-position estimate and a clock-offset estimate as a by-product.The position covariance is taken from the upper-left 2×2 state-covariance submatrix.
  • EKF initialization: The proposed two-phase initialization uses only normal UN–AN communication, first obtaining coarse position and velocity estimates before refining the state with a DoA-only EKF.The second phase avoids immediately relying on potentially unreliable ToA updates and improves both position and velocity initialization.
  • Coarse positioning: The first initialization phase can use centroid localization from known line-of-sight access-node positions, with weighted centroid localization offered as an improvement.The unweighted estimate is the mean of the line-of-sight access-node positions and may be poor depending on the user’s relative location.

A. Positioning and Network Synchronization EKF at Central Unit

The proposed DoA/ToA Pos&Sync EKF extends joint positioning and clock tracking to mutually unsynchronized access nodes by augmenting the state with access-node clock offsets. Its measurement and transition models incorporate these offsets while retaining the user-node motion and clock states.

  • The DoA/ToA Pos&Sync EKF tracks mutual clock offsets of line-of-sight access nodes using available ToA measurements.
  • The augmented state contains user position, velocity, and clock parameters together with line-of-sight access-node offsets relative to a chosen reference access-node clock.
  • The state transition model combines the existing user-node transition with an identity block representing the assumed access-node clock-offset evolution.
  • The ToA measurement equations are revised by adding the clock offset of the considered line-of-sight access node.
  • The EKF uses a Kalman-gain formulation, modified Jacobians, and initialization procedures to obtain user position and clock-offset estimates with corresponding covariance elements.
  • A reference access node must be selected because synchronization in the unsynchronized network is defined relative to that node.

B. Propagation of Universal Network Time

The paper considers how estimated relative clock offsets can support propagation of a common network time across multiple tracked user nodes. It also identifies storage and memory demands at the central node as a practical consideration.

  • With multiple user nodes, clock-offset estimates may have different time references, requiring treatment of their relative timing information.
  • Storing clock-offset information increases computational load and memory use in the network’s central node.
  • If relative offset information is available when tracking a new user node begins, it would most probably speed EKF convergence and improve the user-node clock-offset estimate.

V. NUMERICAL EVALUATIONS AND ANALYSIS

The numerical evaluations quantify device-positioning performance in an urban outdoor METIS Madrid environment using a 3.5 GHz ultra-dense network and a connected-car scenario.

  • The evaluations use the METIS Madrid grid model in an urban outdoor environment with a 3.5 GHz ultra-dense network.
  • The connected-car scenario models vehicles driving through the city at velocities in the order of 50 km h^-1.

A. Simulation and Evaluation Environment

The evaluation environment combines the METIS Madrid urban map, realistic vehicle motion, ray-traced propagation, cylindrical antenna arrays, interference modeling, and specified 5G radio parameters. Positioning updates are evaluated under multiplexed uplink reference signals and varying access-node conditions.

  • Madrid map: The Madrid map models dense urban buildings, roads, and sidewalks using a two-dimensional layout derived from the METIS environmental model.
  • Channel and antenna models: Ray tracing models reflected and diffracted uplink-reference-signal paths through the three-dimensional Madrid environment.
  • Channel and antenna models: Uncoordinated interference is modeled with randomly placed interferers using a geometry-based stochastic channel model and spatially correlated receive covariance.
  • Channel and antenna models: The access nodes use cylindrical arrays with 10 dual-polarized patch elements, arranged as two circles of five elements.
  • UN motion model: Vehicles accelerate toward 50 km h^-1 on straight segments and turn at a constant 20 km h^-1.
  • 5G radio interface: The radio interface uses OFDMA with 75 kHz subcarrier spacing, 100 MHz bandwidth, and 1280 active subcarriers.
  • Evaluation procedure: Orthogonal uplink reference signals are assumed within each coordination area, while co-channel interference produces receiver SINR values from 5 dB to 40 dB.
  • Evaluation procedure: The EKFs update once per 100 ms, initially fusing the two closest line-of-sight access nodes to study pilot allocation and access-node spacing effects.

B. DoA and ToA Estimation

The proposed DoA/ToA EKF tracks azimuth direction and arrival time across multiple LoS-AN configurations, with ToA benefiting from wider bandwidth and shorter inter-site distance while DoA accuracy remains consistently high.

  • Evaluation setup: DoA and ToA RMSEs are averaged over 15 random routes, comparing the closest and second-closest LoS-ANs.Colored bars represent closest LoS-ANs, while gray bars represent second-closest LoS-ANs.
  • ToA estimation: 96 MHz bandwidth provides more accurate ToA estimation than 19.2 MHz because it improves time-domain resolution.The comparison uses uplink beacons transmitted with different bandwidths and subcarrier configurations.
  • ToA estimation: Reducing inter-site distance improves ToA estimates, especially with the narrower bandwidth, through higher average SINRs at the ANs.The improvement is less pronounced with 96 MHz bandwidth.
  • DoA estimation: Azimuth DoA estimation is generally highly accurate and varies little across inter-site distances or between the closest and second-closest ANs.More distant UNs can provide more favorable geometry for azimuth estimation.

C. Positioning, Clock and Network Synchronization

The cascaded EKFs are evaluated on moving UNs across the Madrid map under different numerologies, synchronization assumptions, LoS-AN counts, and detection quality. They achieve sub-meter positioning and accurate clock-offset tracking, although unsynchronized ANs and rapid handovers reduce synchronization performance.

  • Evaluation setup: The evaluation tracks UNs along 15 random Madrid-map routes using positioning and synchronization EKF updates at the central unit.The positioning and synchronization EKFs update every 500th radio sub-frame, while individual AN measurements are communicated every 100 ms.
  • Positioning: The proposed Pos&Clock and Pos&Sync EKFs significantly outperform the DoA-only EKF across all considered evaluation scenarios.The proposed methods use DoA and ToA information, whereas the comparison method is DoA-only.
  • Positioning: Sub-meter positioning is achieved in every test scenario, with RMSE below 0.5 m using 96 MHz bandwidth and approximately 25 m ISD.The proposed methods also avoid the unfavorable near-collinear geometry that degrades the DoA-only EKF.
  • Clock synchronization: With a synchronized network, the Pos&Clock EKF estimates UN clock offsets with RMSE below 2 ns in every test scenario.An initial clock-offset standard deviation of 100 µs is generally reduced by 5 orders of magnitude.
  • Clock synchronization: With unsynchronized ANs, clock-offset estimation is worse than in the synchronous case, and reducing ISD from 50 m to 25 m degrades UN clock-offset accuracy somewhat.At 50 km/h and 25 m ISD, each LoS-AN remains visible for only 1.8 s, limiting the available measurements and EKF iterations.
  • LoS-AN availability and detection: Fusing measurements from three closest LoS-ANs improves positioning and synchronization over using two, achieving below-30-cm positioning accuracy even with unsynchronized network elements.The proposed methods retain sub-meter positioning accuracy despite imperfect LoS detection.
  • LoS-AN availability and detection: Rapid handovers, especially when only one LoS-AN is used, degrade UN and AN clock-offset estimation in the Pos&Sync EKF.With K[n] = 1, Pos&Clock outperforms Pos&Sync because early UN clock-offset errors accumulate in the unsynchronized network.

VI. CONCLUSION

The paper presents a network-centric cascaded EKF framework for joint device positioning and clock synchronization in realistic 5G ultra-dense-network scenarios. Evaluations on the Madrid map show sub-meter positioning and nanosecond-level synchronization while keeping user-device processing requirements low.

  • Contribution: All essential processing is performed on the network side to minimize power consumption and computing requirements at user devices.The cascaded structure estimates DoA and ToA, then fuses them for positioning and clock-offset estimation.
  • Contribution: The cascaded EKFs jointly estimate DoA, ToA, device position, device clock offsets, and mutual clock offsets between network elements.The evaluations use realistic movement models on the Madrid grid with full ray-tracing propagation modeling.
  • Conclusion: Sub-meter positioning and nanosecond-level clock-offset estimation are technically feasible in realistic 5G radio access networks.The conclusion identifies these results under realistic clock-offset assumptions and potentially unsynchronized network elements.
  • Future work: Future work extends the solutions to 3D positioning and applies accurate positioning information to mobility management and location-based beamforming.These directions are stated as planned extensions of the proposed solutions.
Loading 1604.03322v3…