Probabilistic snapshot GNSS for low-cost wildlife tracking

Thesis

Snapshot GNSS is more energy-efficient than conventional localisation methods based on global navigation satellite systems (GNSS), like the GPS. This is beneficial for long deployments on battery such as in wildlife tracking. However, only a few snapshot GNSS systems that could be used for wildlife tracking have been presented and all have disadvantages. Most significantly, they are closed-source and either not available or expensive. A reason is that they typically require GNSS signals to be captured with good resolution, which demands complex receiver hardware capable of capturing multi-bit data at sampling rates of 16 MHz and more. By contrast, this thesis presents fast algorithms that reliably estimate locations from twelve-millisecond signals that are sampled at just 4 MHz and quantised with only a single bit. This allows to build a snapshot receiver at an unmatched low cost of less than $30 and with particularly low power consumption, outperforming existing systems and enabling low-budget and long-term field work. The system can acquire two positions per hour for a year on a tiny 40 mAh battery. On a challenging public dataset with thousands of snapshots from real-world scenarios, median accuracy is 11 m, comparable to more complex and expensive solutions with higher energy consumption. Additionally, the system has been deployed for several wildlife tracking studies, including on sea turtles, where brief signal acquisition times are crucial to obtain location fixes during surfacing events lasting only 1–2 s. For the first time, (i) snapshot GNSS receiver hardware and (ii) an accompanying cloud-based processing platform are open-source. This allowed several third parties to independently replicate the system. In total, several hundred receivers have been built and millions of locations estimated for those. As three additional contributions, this thesis presents (i) the first evaluation of snapshot GNSS for wildlife tracking across a variety of species and habitats, (ii) the first snapshot GNSS system with cloud-offloading via a low-power narrow-band cellular connection, and (iii) a demonstration of the potential of smoothing for snapshot GNSS. A final contribution are factor graph optimisation algorithms to (i) smooth snapshot GNSS data and (ii) tightly fuse raw GNSS data with inertial measurements and, optionally, lidar observations for precise and smooth localisation. In several environments with little sky visibility, such as a forest, the accuracy of the fused location estimates in the global Earth frame is still 1–2 m, while the estimated trajectories are discontinuity-free and smooth. This requires a professional-grade (non-snapshot) GNSS receiver, but, unlike traditional differential GNSS, no connection to a base station.

Authors: Beuchert, J

• DOI: 10.5287/ora-dae7ap229
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English