Fixed point performance of interpolation/extrapolation algorithms for resource constrained wireless sensors

Date collection in wireless networks is performed by sampling the phenomenon to be measured at recurring intervals in time. In situations where the sensors must operate over a very long time without the possibility of battery replacement, the sleep cycle for these sensors must be as long as possible (and the longest the design allows). Longer sleep cycles, while extending battery life may be too long for the sampling rate required to collect data - i.e. more frequent wake ups would be needed from the sensor. On the other hand, the sensors can collect samples at a lower than required rate, but interpolate data-points in between the actual measured sampled data. Interpolation implemented in floating point can be demanding in terms of energy cost to the sensors. Therefore as it is generally the case, fixed point implementations have to be introduced to reduce the energy demands due to excessive computations/instruction count due to floating point arithmetic. In this paper we present a fixed point implementation of the Neville's interpolation scheme, and present performance results as compared to the floating point (ideal) results.