Like members of an orchestra require a conductor to keep time to make sure their notes correlate with each other, the nodes of wireless sensor networks require time synchronization to make sure their data correlates with one another.
Farzad Asgarian, a University of Michigan PhD student, researches how to make sure wireless sensors stay in sync, and has developed a way to reduce synchronization error while still reducing the power required. This can open up Internet of Things applications from body sensors for athletes to health and usage monitoring systems, or HUMS, for large vehicles such as helicopters and trains.
“With increased performance in correlating data across sensors, which precise time synchronization allows, sensor networks could be more effectively applied to an athlete, such as a rower, to help evaluate and improve their movement,” Asgarian said.
“And, for machinery like a train where cost of repairs is high and operation is critical, improved time synchronization in these sensor networks is necessary for analyzing the condition of parts in order to ensure they undergo maintenance when necessary.”
Sensor networks could be more effectively applied to an athlete, such as a rower, to help evaluate and improve their movement. Farzad Asgarian
Time synchronization in wireless sensor networks has typically relied on high-resolution timestamps, requiring a higher frequency clock, for example at 16 megahertz. In addition, previous methods of time synchronization have required a base node to synch with other nodes every few seconds to help keep timing errors low.
However, both of these synchronization methods, the higher frequency clock and need to constantly transmit resynchronizing messages, require extra energy.
To save on power, a lower frequency clock can be used, such as one megahertz as in Asgarian’s research. However, this lower frequency clock causes the synchronization error between nodes to deviate wildly over time, with more than five times the error than with the higher frequency 16 megahertz clock.
To address this, Asgarian, who is advised by Khalil Najafi, Schlumberger Professor of Engineering and Arthur F. Thurnau Professor of Electrical Engineering and Computer Science, introduced frequency scaling to the problem.
“Frequency scaling means that the processor dynamically adjusts its clock frequency based on the work load. For example, when the processor is only interfacing a low-speed peripheral, it runs at a lower clock speed, and when it has to complete a lot of tasks, it runs at a higher clock speed,” Asgarian explained. “Ultimately, this saves energy.”
Asgarian broke up the task of time synchronization into two phases. The first phase synchronizes the sensor nodes in a few seconds, and runs at the higher frequency clock of 16 megahertz. The second phase is the measuring task, in which over a few minutes the sensor reports its data at time intervals calculated based on the first phase parameters. This second phase runs at one megahertz, and consumes less power.
This approach is implemented alongside BlueSync, a synchronization protocol that was introduced by Asgarian for Bluetooth Low Energy (BLE).
By utilizing the accuracy of 16 megahertz to sync the sensors, and then the lower power requirement of one megahertz for the running task, the synchronization error remains low while using seven times less power for the timer and two times less power for the oscillator. This means sensors can report measurements using dramatically less power, while still reporting data at a quality found in higher-powered systems.
For this work, outlined in the paper, “Frequency Scaling in Time Synchronization for Wireless Sensor Networks,” Asgarian received a Best Poster Award at the 2018 IEEE International Conference on Distributed Computing in Sensor Systems (DCOSS).
Posted September 10, 2018