Long-term Audio Observation by Wireless Sensor Networks with Filtering Strategies

Due to the wireless transmission capability and ease of replaceability of wireless sensor networks (WSN) nodes, they can be deployed in harsh environments for monitoring environmental changes. However, a sensor network faces the challenge of limited residual energy, particularly in audio sensing where large amounts of data are communicated and drainage of sensor energy is significant. In order to conserve energy, it is thus desirable for sensor nodes to intelligently analyze and filter sensed data to be delivered to the server. Nevertheless, data processing on the computationally limited sensor node needs to be lightweight. In other words, the data processing technique needs to operate efficiently and accurately under memory and processing constraints. In this paper, two approaches are proposed to filter redundant sensed audio data in order to prolong WSN lifetime. The first approach analyzes the fundamental frequencies in the distribution of audio data to prune unobserved audio, while the second approach compares the amplitudes of multiple audio signals to filter out duplicated data. Our experimental results prove that through applying the proposed approaches for wildlife detection, the sensor network lifetime can be significantly prolonged.