Offset Tracking of sensor clock using Kalman filter for wireless network synchronization

International audienceWireless Sensors Networks (WSN) are more and more used in structural health monitoringapplications since they represent a less expensive and non-invasive way to monitorinfrastructures. Most of these applications work by merging or comparing data fromseveral sensors located across the structure. These data often comprise measurementsof physicals phenomenons evolving with time, such as acceleration and temperature.To merge or compare time-dependent data from different sensors they need to be synchronizedso all the samples are time-stamped with the same time reference. An initialsynchronization of the sensors is needed because sensors are independent and thereforecan not be all started at the same time. Subsequent re-synchronizations are also neededsince the sensors keep track of time using their imperfect local clock. A quartz clock willdrift in time due to the sensitivity of the quartz oscillator to its environmental conditions ;thus, synchronization accuracy depends on the quality of the oscillator, environmentalconditions, re-synchronization frequency and time reference quality. The required accuracyof the synchronization depends on applications, for instance, one needs millisecondaccuracy to analyze vibration data, microsecond for acoustic data and nanosecond totime-stamp electromagnetic propagation. Sensors can be synchronized by exchangingtiming information as in the RBS [1], TPSN [2] and FTSP [3] protocols or through anexternal timing reference such as the PPS signal transmitted by the GPS as in [4] and [5].While the first option might be less power hungry, our work in this paper focus on thesecond option as it allows for a better synchronization accuracy. This paper presents asmart-sensor able of time-stamping samples as well as measuring its clock offset andfrequency from a noisy PPS signal. We implemented it on an FPGA to get high speedcounters without software overhead.We then used a Kalman filter to track the offset withmore accuracy and to adjust the sample time-stamps of the sensors. This work is aimingat having the highest time accuracy as possible while minimizing its consequence onsensor power-consumption