Dynamic clock calibration via temperature measurement

We study a clock calibration problem for an ultra-low power timer on a sensor node platform. When the sensor is put into sleep mode, this timer is the only thing left running, so power consumption is extremely low. However, this power savings comes at the expense of loss in timing accuracy, which can result in unnecessary energy consumption from two unsynchronized devices trying to communicate. The speed of the timer is dependent on the ambient temperature. Because this dependence can be measured off-line, one way to improve timing accuracy is to periodically wake the processor up to take temperature measurements, and use these measurements in combination with the local clock ticks to obtain a more accurate estimate of the elapsed real time. The goal of this work is to dynamically schedule a limited number of temperature measurements in a manner most useful to improving the accuracy of the timer. We present a stochastic control formulation to solve this problem.