Improved Temperature Compensation of Atomic Clocks and INS Instruments using Multivariate Model-based Design Optimized for Real-world Operating Conditions.

Atomic clocks, gyroscopes, magnetometers, and related INS precision equipment are often performance-limited by temperature stability. Because of the cumulative effect of systematic errors, squeezing even small reductions in temperature sensitivity (tempco) can produce significant improvements in holdover performance. One common method for reducing tempco is by directly applying a correction based on predicted output errors. The predicted output error is computed from internal temperature sensors in the instrument. The correction uses thermal chamber experiments to directly measure temperature sensitivity. This paper will look at existing methods and propose an improve prediction algorithm based on Deep Neural Networks.