Abstract
Location-aware networks will enable new services and applications in fields such as autonomous driving, smart cities, and the Internet-of-Things. One promising solution for ubiquitous localization is network localization and navigation (NLN), where devices form a network that cooperatively localizes itself, reducing the infrastructure needed for accurate localization. This paper introduces a real-time NLN system named Peregrine, which combines distributed NLN algorithms with commercially available ultra-wideband (UWB) sensing and communication technology. The Peregrine software application, for the first time, integrates three NLN algorithms to jointly perform the tasks of localization and network operation in a technology agnostic manner, leveraging both spatial and temporal cooperation. Peregrine hardware is composed of low-cost, compact devices that comprise a microprocessor and a commercial UWB radio. This paper presents the design of the Peregrine system and characterizes the performance impact of each algorithmic component. Indoor experiments validate that our approach to realizing NLN is both reliable and scalable, and maintains sub-meter-level accuracy even in challenging indoor scenarios.
Introduction
While Global Navigation Satellite Systems (GNSS) provide reliable outdoor positioning, they are ineffective in indoor environments or "urban canyons" where signals are obstructed. Traditional indoor positioning systems (IPS) typically rely on trilateration, requiring a mobile agent to maintain line-of-sight (LOS) with at least three or four fixed anchors. This dependency necessitates a dense and often prohibitively expensive infrastructure.
Network Localization and Navigation (NLN) addresses these limitations by enabling agents to communicate not only with fixed anchors but also with one another. This cooperative approach allows the network to localize itself even when individual agents lack sufficient anchor visibility. However, implementing NLN at scale introduces significant challenges: the computational complexity of 3D tracking (the "curse of dimensionality") and network congestion caused by excessive inter-node communication.
The Peregrine System
The Peregrine system overcomes these hurdles through a combination of custom hardware—utilizing Raspberry Pi Zero microprocessors and Ultra-Wideband (UWB) radios—and a suite of three specialized algorithms designed for scalable, real-time operation.
Scalable Bayesian Inference
To manage the complexity of 3D localization, Peregrine employs Sigma Point Belief Propagation (SPBP). Unlike traditional Particle Filters, which are computationally intensive, SPBP provides a robust framework for non-linear, non-Gaussian tracking that is efficient enough to run on low-power microprocessors. This approach leverages both temporal cooperation (tracking an agent's movement over time) and spatial cooperation (sharing information between agents).
Efficient Network Operation
To mitigate network congestion and interference, Peregrine introduces two "smart" management layers:
- Holistic Threshold-based Node Activation (HTNA): Agents evaluate the potential information gain of a new measurement against the "cost" of channel congestion. Communication only occurs if the expected improvement in location certainty exceeds a specific threshold.
- Conic Programming-based Node Prioritization (CPNP): When an agent decides to communicate, it selects the most informative neighbors based on geometric configuration and signal quality, effectively filtering out noisy non-line-of-sight (NLOS) paths.
Experimental Evaluation and Results
The system was validated in challenging real-world environments at MIT, including a metal-walled room and a complex multi-floor lounge. Ground truth was established using high-precision landmarks to calculate Root-Mean-Square Error (RMSE).
Key findings include:
- High Precision: Peregrine consistently achieved 3D positioning accuracy within 30–40 cm.
- Cooperation Benefits: Enabling spatial cooperation between agents improved accuracy by 53% compared to non-cooperative baselines.
- Operational Efficiency: The HTNA algorithm reduced the measurement rate by 46% without sacrificing accuracy, significantly lowering power consumption.
- Robustness: In multi-floor scenarios, the prioritization algorithm (CPNP) reduced localization outages by over 60% by successfully identifying and ignoring obstructed signals.
Conclusion
Peregrine demonstrates that high-accuracy, real-time 3D indoor localization is achievable using low-cost, off-the-shelf hardware. By integrating scalable inference with intelligent network management, the system provides a viable path for large-scale cooperative localization. Future work will focus on extending these methods to high-speed dynamic environments, such as autonomous drone swarms, and further refining hardware-level UWB interactions.
Summary updated