Time synchronisation for millisecond-precision on bio-loggers

Abstract

Time-synchronised data streams from bio-loggers are becoming increasingly important for analysing and interpreting intricate animal behaviour including split-second decision making, group dynamics, and collective responses to environmental conditions. With the increased use of AI-based approaches for behaviour classification, time synchronisation between recording systems is becoming an essential challenge. Current solutions in bio-logging rely on manually removing time errors during post processing, which is complex and typically does not achieve sub-second timing accuracies.We first introduce an error model to quantify time errors, then optimise three wireless methods for automated onboard time (re)synchronisation on bio-loggers (GPS, WiFi, proximity messages). The methods can be combined as required and, when coupled with a state-of-the-art real time clock, facilitate accurate time annotations for all types of bio-logging data without need for post processing. We analyse time accuracy of our optimised methods in stationary tests and in a case study on 99 Egyptian fruit bats (Rousettus aegyptiacus). Based on the results, we offer recommendations for projects that require high time synchrony.During stationary tests, our low power synchronisation methods achieved median time accuracies of 2.72 / 0.43 ms (GPS / WiFi), compared to UTC time, and relative median time accuracies of 5 ms between tags (wireless proximity messages). In our case study with bats, we achieved a median relative time accuracy of 40 ms between tags throughout the entire 10-day duration of tag deployment. Using only one automated resynchronisation per day, permanent UTC time accuracies of ≤ 185 ms can be guaranteed in 95% of cases over a wide temperature range between 0 and 50 °C. Accurate timekeeping required a minimal battery capacity, operating in the nano- to microwatt range.Time measurements on bio-loggers, similar to other forms of sensor-derived data, are prone to errors and so far received little scientific attention. Our combinable methods offer a means to quantify time errors and autonomously correct them at the source (i.e., on bio-loggers). This approach facilitates sub-second comparisons of simultaneously recorded time series data across multiple individuals and off-animal devices such as cameras or weather stations. Through automated resynchronisations on bio-loggers, long-term sub-second accurate timestamps become feasible, even for life-time studies on animals. We contend that our methods have potential to greatly enhance the quality of ecological data, thereby improving scientific conclusions.

Keywords: Animal tracking; Embedded systems; GPS; Internet of animals; IoT; Movement ecology; Proximity; Real time; Telemetry; WiFi; Wireless sensors.

Fig. 1

Error model for measuring time on bio-loggers. Drifts (T_drift) and time offset errors (T_off) of clocks integrated into bio-loggers lead to deviations from the reference time t. Different clocks have different errors, resulting in relative time differences (T_{error; relative}) between devices

Fig. 2

Overview of the three evaluated methods (GPS [A], WiFi and NTP servers [B], wireless proximity messages [C]) to automatically (re)synchronise time onboard bio-loggers

Fig. 3

Achieved time accuracies with optimised methods for automatic time synchronisation on bio-loggers in stationary experiments (GPS [A], WiFi and NTP [B], wireless proximity messages [C-D]). The achieved accuracies with WiFi (B) are separated by the internet speed (broadband, mobile LTE, mobile 2G). During the 16-day-long relative time synchronisation experiment between two tags (one positioned outside, the other inside), an automatic resynchronisation was triggered because the tags measured a relative time difference of more than 50 ms onboard (D). The tag temperatures were measured with an onboard temperature sensor (Bosch Sensortec BME680)

Fig. 4

Time synchronisation accuracies achieved on 99 bio-loggers attached to Egyptian fruit bats. Time differences were calculated using the time of arrival of proximity messages on data gateways placed inside the bat cave (B). Precise time synchronisation permitted the alignment of 800-ms-long 50 Hz acceleration bursts across multiple tags (exemplified in A; bursts are drawn with real time offsets). This capability facilitates the analysis of collective bat behaviour at an exceptionally proximate temporal resolution