Understanding the behavior of marine mammals is quite limited by the observation technology used. Surface tracking using either geolocation or Argos satellite tags have shown that these mammals range much farther than previously thought. Relatively simple time/depth recorders (TDR's) have shown dives to depths of over 1000 meters for durations of over one hour. To further the understanding of these aquatic creatures, a smaller and more capable tag is being developed that can be deployed for longer durations and with more sensing capabilities. This tag utilizes a sensor suite consisting of temperature, depth, speed, salinity, three axes of magnetic field, three axes of acceleration, and GPS. The three-axis magnetometers and three-axis accelerometers are used to reconstruct the full attitude quaternion of the creature. Fusing this attitude measurement with water speed, and both initial and final position estimates from GPS, a full three dimensional underwater trajectory can be reconstructed (distributing the error from the return surface as an estimate of the ocean currents). This paper looks at three types of dead reckoning filters used to process this data via simulation: (1) pure open loop integration, (2) "scaled" integration that feeds back measured depth, and (3) a Kalman filter used to estimate the position of the creature, with a correction term from the measured depth (pressure). Comparison of these three navigation filters on simulated data for a 20 minute dive shows that the Kalman filter works best, in terms of position drift, though at the cost of high frequency noise in the trajectory. At the end of a 2 Km path, it shows a total offset error of approximately two meters (or 0.1% drift). The scaled filter is the worst, demonstrating instability at certain points, and the open loop integration yields a position estimate that is off approximately 5 meters at the end of the dive.
