Data Compression for Activity Detection with Wireless Sensor Nodes (Masterarbeit)
Wireless sensor networks (WSN) are gaining popularity in part due to their high relia- bility, dynamic reconfigurability, low cost and ability to measure and analyze distributed sources. Examples include environmental monitoring, sound (object) localization or ac- tivity detection in form of body area networks (BAN). For the latter application large quantities of acceleration data are processed and transmitted to collection devices which can result in substantial power expenditures from wireless communication alone. Data compression on the sensing nodes could be used to reduce the amount of transmitted data as long as activity detection is not compromised. The activity analysis algorithm used with an existing multi-sensor mote was revised and optimized for low-complexity hard- ware without compromising its sensitivity. Then selected data compression methods were evaluated against this algorithm. Slepian-Wolf Coding and run-length encoding were not effective in compressing acceleration data while differential encoding, a dictionary method and foremost wavelet based compression were highly effective. A multistage compression algorithm (wavelet filtering, differential encoding, exponential Golomb code) was derived from these analyses. Finally a new BAN mote with typical limited processing power and storage capability was designed and the components of the proposed compression scheme implemented to demonstrate feasibility and measure the current consumption. Differential encoding with a variable length code resulted in a ca. 45% increase in power consumption as the increase in computation (largely due to inefficiently implemented bit manipula- tions) exceeded the decrease in communication. Wavelet filtering to 25% of the original data rate resulted in a ca. 17% decrease in power consumption exclusively due to the reduction in communication while the expense of computation was nearly unaltered.