Large Distributed Sensor Networks and Bandwidth Constraints for Time Critical Applications
One of the major issues of paramount importance that has precluded the widespread use of the distributed wireless sensors in many applications that involve real-time sensing and processing streaming of data is the lack of a simple sensor-level detection, feature extraction and data compression for low bit rate transmission of essential signal attributes to the base station. This is exacerbated by several practical limitations including, cost, size, power consumption, and onboard DSP and communication system.
In moderately large (with more than 100 nodes) distributed sensor networks with sensor nodes that use communication protocols such as zigbee-based communication protocols, data rates of less than 2 kbps per sensor node are needed in order to meet the bandwidth limitations, while guaranteeing the usefulness of the data for accurately locating moving sources. Thus, a system for sensor-level detection, feature extraction and data compression in large distributed sensor networks can significantly enhance the capabilities of the existing sensor nodes and make their widespread use in many applications a practical reality.
Our Innovation
ISTI's innovative solution involve,
- Designing and building a multi-channel sensor board using a field programmable gate array (FPGA) processor, memories, onboard radio, and several additional auxiliary channels (see Products). The sensor board can be interfaced with zigbee-based sensor nodes to augment local processing and data collection capabilities.
- Developing and implementing dedicated yet simple signal processing algorithms.
- A novel collaborative subband selection process is also developed to allow generating consistent (among multitudes of distant sensor nodes) data compression.
The proprietary algorithms developed by ISTI allow sensor-level frequency subband feature extraction, data compression and encoding for streaming data e.g., for ground and airborne vehicle tracking and personnel tracking applications. The system offers excellent bit rates of less than 2 kilo bit per second (kbps) even for encoding acoustic signatures of several moving sources without sacrificing any noticeable accuracy in the localization and tracking at the base station.
Figure below shows the estimated locations of a moving truck and its true path (based upon hand-held GPS) as well as the locations of four clusters of distributed sensor nodes with a total of twenty one nodes. This result was obtained by using the reconstructed versions of the compressed signals (bit rate of 1.5 kilo bit per second ) at the base station. The results clearly indicate that even at this low compression rate the estimated truck locations are accurate most of the time. Experiments revealed that accuracies of the estimated source locations using the compressed and uncompressed data are very close. This implies that using the compressed data achieves comparable performance to that of the uncompressed data while allowing for deployment of much larger number of sensor nodes to cover a much larger surveillance area. This, in turn enables one to detect and locate sources at larger distances. Additionaly, further studies indicated that the direction of arrival (DOA) estimation accuracy of the proposed method when implemented on real compressed data is very close to those of model generated data for the same operating and environmental conditions.
