Description
(4) Discussion and Conclusion
An ongoing analysis of our results revealed that we have significantly layering effects, meaning that we see significant temperature differences along the vertical axis (about 1°C). We believe this to be due to heating from the DAQs on the outside of the bowl holding the water. We are currently investigating these two phenomena further.
(2) Material and Methods
In a 1st step temporal oversampling reduces the ADC noise and achieve sub-bit “super resolution” by a low-order polynomial fit which also models temperature drifts. In the 2nd step, locality & symmetricity for the sensors surrounding the HQ sensors is utilized. A global compensation is created by utilizing many measurements. A final step integrates all data in a mode assuming a low rate of change over space and time, by that removing individual deviations.
For added robustness, we expand our method to include procedure variations and unknown influences now: air temp., device heat up, water temp. Therefore we installed additional sensors and created a setup which applied fast temperature changes to gain data about the temperature inertia.
(3) Results
The accuracy and precision of the temp. data has been improved by more than an order of magnitude. This is well-beyond the original hardware-limited 1-bit resolution. We achieved a sub-bit accuracy of 0.005°C for a confidence interval of 95% as opposed to the original $\pm$1°C raw accuracy. Our method automatically minimizes and substitutes outliers and faulty data using a robust model, reduces the sizeable hardware-related variation of about 0.6°C among individual sensors & runs without additional time-intensive delays on the USCT measurement process.
However, that we do on occasion see significant deviation ($\pm$1°C) from expected results using our method. For this reason, we are currently working to incorporate more data.
