Distributed Wireless MEMS Networks

Today many position aware devices are demanded. Micro-Electro-Mechanical Systems (MEMS) can detect their own acceleration. Out of these values the movement, and thus the new position, can be calculated. The problem of these systems is, they are very error-prone. In order to minimize the fault this paper introduces a new way of combining different MEMS measurements to achieve a more accurate relative positioning in regard of the object. This is done by combining measurements of different sensors. All data communication will be wireless over the IEEE 802.15.4 standard.

We started from the idea that macroscopical properties could help filtering the data from a MEMS accelerometer, without the use of other hardware like gyroscopes or GPS. The problem with the macroscopical information is that it is presented as positions and the incoming messages are accelerations. On this accelerations there is a lot of drift, some drift is even bigger then the real accelerations, resulting in large errors after converting the accelerations and its drift into new positions.

Converting the macroscopical properties into accelerations is also not an option, because the integrations in the conversion functions produce faults and then the integrations need to be used twice. First to calculate the accelerations of the macroscopical properties and second to get the new position.

A solution to minimize the drift with only MEMS devices was needed. Therefor we started combining data of different sensors. the outcome is a filter that uses a weighted mean of the different sensors and adepts the weight in time. This mean is not yet perfect but the drift is already decreased with 50 to 80 percent. Adapting the weight to more factors could decrease the drift even more but at a certain point the complexity of the weight can start to have its downsides. A Kalman filter can offer a solution to keep the complexity of the weight low and still decrease the drift majorly. If a third or fourth order Kalman filter is implemented after the adaptable weighted mean filter drift could even diminish more.

Since the original idea still hasn’t been researched, research to using the macroscopical properties to filter the converted data can still be done. To turn the disadvantage of the constant 1G force on the z-axis in an advantage, one can stack the sensors on each other, each single stack representing a single sensor module. Placing these modules on strategically chosen parts of an object could result in even measuring rotations of the object, while this normally can not be done with a single MEMS accelerometer.

All the research can be found in the final paper.

The poster: